rust/compiler/rustc_errors/src/lib.rs

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

2051 lines
79 KiB
Rust
Raw Normal View History

//! Diagnostics creation and emission for `rustc`.
//!
//! This module contains the code for creating and emitting diagnostics.
// tidy-alphabetical-start
#![allow(incomplete_features)]
#![allow(internal_features)]
#![allow(rustc::diagnostic_outside_of_impl)]
#![allow(rustc::untranslatable_diagnostic)]
2020-09-23 21:51:56 +02:00
#![doc(html_root_url = "https://doc.rust-lang.org/nightly/nightly-rustc/")]
2023-11-13 07:39:17 -05:00
#![doc(rust_logo)]
#![feature(array_windows)]
2024-08-11 12:10:36 -04:00
#![feature(assert_matches)]
#![feature(associated_type_defaults)]
#![feature(box_into_inner)]
2024-01-14 20:08:47 +00:00
#![feature(box_patterns)]
#![feature(default_field_values)]
#![feature(error_reporter)]
#![feature(if_let_guard)]
#![feature(let_chains)]
#![feature(negative_impls)]
#![feature(never_type)]
#![feature(rustc_attrs)]
#![feature(rustdoc_internals)]
#![feature(trait_alias)]
2023-01-05 03:39:07 +01:00
#![feature(try_blocks)]
#![feature(yeet_expr)]
// tidy-alphabetical-end
extern crate self as rustc_errors;
2024-08-11 12:10:36 -04:00
use std::assert_matches::assert_matches;
use std::backtrace::{Backtrace, BacktraceStatus};
use std::borrow::Cow;
use std::cell::Cell;
use std::error::Report;
use std::ffi::OsStr;
use std::hash::Hash;
use std::io::Write;
2024-01-29 23:59:09 +01:00
use std::num::NonZero;
use std::ops::DerefMut;
use std::path::{Path, PathBuf};
use std::{fmt, panic};
use Level::*;
pub use codes::*;
pub use diagnostic::{
BugAbort, Diag, DiagArg, DiagArgMap, DiagArgName, DiagArgValue, DiagInner, DiagStyledString,
Diagnostic, EmissionGuarantee, FatalAbort, IntoDiagArg, LintDiagnostic, StringPart, Subdiag,
SubdiagMessageOp, Subdiagnostic,
};
pub use diagnostic_impls::{
DiagArgFromDisplay, DiagSymbolList, ElidedLifetimeInPathSubdiag, ExpectedLifetimeParameter,
IndicateAnonymousLifetime, SingleLabelManySpans,
};
pub use emitter::ColorConfig;
use emitter::{DynEmitter, Emitter, is_case_difference, is_different};
use rustc_data_structures::AtomicRef;
use rustc_data_structures::fx::{FxHashSet, FxIndexMap, FxIndexSet};
use rustc_data_structures::stable_hasher::StableHasher;
use rustc_data_structures::sync::{DynSend, Lock};
pub use rustc_error_messages::{
DiagMessage, FluentBundle, LanguageIdentifier, LazyFallbackBundle, MultiSpan, SpanLabel,
SubdiagMessage, fallback_fluent_bundle, fluent_bundle,
};
use rustc_hashes::Hash128;
use rustc_lint_defs::LintExpectationId;
pub use rustc_lint_defs::{Applicability, listify, pluralize};
use rustc_macros::{Decodable, Encodable};
pub use rustc_span::ErrorGuaranteed;
pub use rustc_span::fatal_error::{FatalError, FatalErrorMarker};
use rustc_span::source_map::SourceMap;
2025-02-20 23:05:38 +00:00
use rustc_span::{BytePos, DUMMY_SP, Loc, Span};
pub use snippet::Style;
// Used by external projects such as `rust-gpu`.
// See https://github.com/rust-lang/rust/pull/115393.
pub use termcolor::{Color, ColorSpec, WriteColor};
use tracing::debug;
use crate::registry::Registry;
pub mod annotate_snippet_emitter_writer;
pub mod codes;
2017-08-19 03:09:55 +03:00
mod diagnostic;
mod diagnostic_impls;
pub mod emitter;
2023-01-05 03:39:07 +01:00
pub mod error;
pub mod json;
mod lock;
pub mod markdown;
pub mod registry;
2019-12-22 17:42:04 -05:00
mod snippet;
mod styled_buffer;
#[cfg(test)]
mod tests;
pub mod translation;
pub type PResult<'a, T> = Result<T, Diag<'a>>;
2019-10-11 13:06:36 +02:00
rustc_fluent_macro::fluent_messages! { "../messages.ftl" }
2019-10-11 13:06:36 +02:00
// `PResult` is used a lot. Make sure it doesn't unintentionally get bigger.
#[cfg(target_pointer_width = "64")]
rustc_data_structures::static_assert_size!(PResult<'_, ()>, 24);
#[cfg(target_pointer_width = "64")]
rustc_data_structures::static_assert_size!(PResult<'_, bool>, 24);
2019-10-11 13:06:36 +02:00
#[derive(Debug, PartialEq, Eq, Clone, Copy, Hash, Encodable, Decodable)]
pub enum SuggestionStyle {
/// Hide the suggested code when displaying this suggestion inline.
HideCodeInline,
/// Always hide the suggested code but display the message.
HideCodeAlways,
/// Do not display this suggestion in the cli output, it is only meant for tools.
CompletelyHidden,
/// Always show the suggested code.
/// This will *not* show the code if the suggestion is inline *and* the suggested code is
/// empty.
ShowCode,
2019-10-03 13:22:18 -07:00
/// Always show the suggested code independently.
ShowAlways,
}
impl SuggestionStyle {
fn hide_inline(&self) -> bool {
2020-10-26 21:02:48 -04:00
!matches!(*self, SuggestionStyle::ShowCode)
}
}
/// Represents the help messages seen on a diagnostic.
#[derive(Clone, Debug, PartialEq, Hash, Encodable, Decodable)]
pub enum Suggestions {
/// Indicates that new suggestions can be added or removed from this diagnostic.
///
/// `DiagInner`'s new_* methods initialize the `suggestions` field with
/// this variant. Also, this is the default variant for `Suggestions`.
Enabled(Vec<CodeSuggestion>),
/// Indicates that suggestions cannot be added or removed from this diagnostic.
///
/// Gets toggled when `.seal_suggestions()` is called on the `DiagInner`.
Sealed(Box<[CodeSuggestion]>),
/// Indicates that no suggestion is available for this diagnostic.
///
/// Gets toggled when `.disable_suggestions()` is called on the `DiagInner`.
Disabled,
}
impl Suggestions {
/// Returns the underlying list of suggestions.
pub fn unwrap_tag(self) -> Vec<CodeSuggestion> {
match self {
Suggestions::Enabled(suggestions) => suggestions,
Suggestions::Sealed(suggestions) => suggestions.into_vec(),
Suggestions::Disabled => Vec::new(),
}
}
}
impl Default for Suggestions {
fn default() -> Self {
Self::Enabled(vec![])
}
}
#[derive(Clone, Debug, PartialEq, Hash, Encodable, Decodable)]
pub struct CodeSuggestion {
/// Each substitute can have multiple variants due to multiple
/// applicable suggestions
///
/// `foo.bar` might be replaced with `a.b` or `x.y` by replacing
/// `foo` and `bar` on their own:
///
2022-04-15 15:04:34 -07:00
/// ```ignore (illustrative)
/// vec![
2017-11-03 16:17:33 +01:00
/// Substitution { parts: vec![(0..3, "a"), (4..7, "b")] },
/// Substitution { parts: vec![(0..3, "x"), (4..7, "y")] },
/// ]
/// ```
///
/// or by replacing the entire span:
///
2022-04-15 15:04:34 -07:00
/// ```ignore (illustrative)
2017-11-03 16:17:33 +01:00
/// vec![
/// Substitution { parts: vec![(0..7, "a.b")] },
/// Substitution { parts: vec![(0..7, "x.y")] },
/// ]
/// ```
2017-11-03 16:17:33 +01:00
pub substitutions: Vec<Substitution>,
pub msg: DiagMessage,
/// Visual representation of this suggestion.
pub style: SuggestionStyle,
/// Whether or not the suggestion is approximate
///
/// Sometimes we may show suggestions with placeholders,
/// which are useful for users but not useful for
/// tools like rustfix
2018-04-25 14:51:06 -07:00
pub applicability: Applicability,
}
#[derive(Clone, Debug, PartialEq, Hash, Encodable, Decodable)]
2017-05-11 15:26:22 +02:00
/// See the docs on `CodeSuggestion::substitutions`
pub struct Substitution {
2017-11-03 16:17:33 +01:00
pub parts: Vec<SubstitutionPart>,
}
#[derive(Clone, Debug, PartialEq, Hash, Encodable, Decodable)]
2017-11-03 16:17:33 +01:00
pub struct SubstitutionPart {
2017-05-11 15:26:22 +02:00
pub span: Span,
2017-11-03 16:17:33 +01:00
pub snippet: String,
2017-05-11 15:26:22 +02:00
}
/// Used to translate between `Span`s and byte positions within a single output line in highlighted
/// code of structured suggestions.
#[derive(Debug, Clone, Copy)]
pub(crate) struct SubstitutionHighlight {
start: usize,
end: usize,
}
impl SubstitutionPart {
pub fn is_addition(&self, sm: &SourceMap) -> bool {
!self.snippet.is_empty() && !self.replaces_meaningful_content(sm)
}
pub fn is_deletion(&self, sm: &SourceMap) -> bool {
self.snippet.trim().is_empty() && self.replaces_meaningful_content(sm)
}
pub fn is_replacement(&self, sm: &SourceMap) -> bool {
!self.snippet.is_empty() && self.replaces_meaningful_content(sm)
}
/// Whether this is a replacement that overwrites source with a snippet
/// in a way that isn't a superset of the original string. For example,
/// replacing "abc" with "abcde" is not destructive, but replacing it
/// it with "abx" is, since the "c" character is lost.
pub fn is_destructive_replacement(&self, sm: &SourceMap) -> bool {
self.is_replacement(sm)
2025-02-20 23:05:38 +00:00
&& !sm
.span_to_snippet(self.span)
.is_ok_and(|snippet| as_substr(snippet.trim(), self.snippet.trim()).is_some())
}
fn replaces_meaningful_content(&self, sm: &SourceMap) -> bool {
sm.span_to_snippet(self.span)
.map_or(!self.span.is_empty(), |snippet| !snippet.trim().is_empty())
}
/// Try to turn a replacement into an addition when the span that is being
/// overwritten matches either the prefix or suffix of the replacement.
fn trim_trivial_replacements(&mut self, sm: &SourceMap) {
if self.snippet.is_empty() {
return;
}
let Ok(snippet) = sm.span_to_snippet(self.span) else {
return;
};
2025-02-20 23:05:38 +00:00
if let Some((prefix, substr, suffix)) = as_substr(&snippet, &self.snippet) {
self.span = Span::new(
self.span.lo() + BytePos(prefix as u32),
self.span.hi() - BytePos(suffix as u32),
self.span.ctxt(),
self.span.parent(),
);
self.snippet = substr.to_string();
}
}
}
2025-02-20 23:05:38 +00:00
/// Given an original string like `AACC`, and a suggestion like `AABBCC`, try to detect
/// the case where a substring of the suggestion is "sandwiched" in the original, like
/// `BB` is. Return the length of the prefix, the "trimmed" suggestion, and the length
/// of the suffix.
fn as_substr<'a>(original: &'a str, suggestion: &'a str) -> Option<(usize, &'a str, usize)> {
let common_prefix = original
.chars()
.zip(suggestion.chars())
.take_while(|(c1, c2)| c1 == c2)
.map(|(c, _)| c.len_utf8())
.sum();
let original = &original[common_prefix..];
let suggestion = &suggestion[common_prefix..];
if suggestion.ends_with(original) {
let common_suffix = original.len();
Some((common_prefix, &suggestion[..suggestion.len() - original.len()], common_suffix))
} else {
None
}
}
impl CodeSuggestion {
/// Returns the assembled code suggestions, whether they should be shown with an underline
/// and whether the substitution only differs in capitalization.
pub(crate) fn splice_lines(
&self,
sm: &SourceMap,
) -> Vec<(String, Vec<SubstitutionPart>, Vec<Vec<SubstitutionHighlight>>, bool)> {
// For the `Vec<Vec<SubstitutionHighlight>>` value, the first level of the vector
// corresponds to the output snippet's lines, while the second level corresponds to the
// substrings within that line that should be highlighted.
use rustc_span::{CharPos, Pos};
/// Extracts a substring from the provided `line_opt` based on the specified low and high
/// indices, appends it to the given buffer `buf`, and returns the count of newline
/// characters in the substring for accurate highlighting. If `line_opt` is `None`, a
/// newline character is appended to the buffer, and 0 is returned.
///
/// ## Returns
///
/// The count of newline characters in the extracted substring.
2016-10-18 23:13:02 +05:30
fn push_trailing(
buf: &mut String,
2019-02-07 03:53:01 +09:00
line_opt: Option<&Cow<'_, str>>,
2016-10-18 23:13:02 +05:30
lo: &Loc,
hi_opt: Option<&Loc>,
) -> usize {
let mut line_count = 0;
// Convert CharPos to Usize, as CharPose is character offset
// Extract low index and high index
2016-10-18 23:13:02 +05:30
let (lo, hi_opt) = (lo.col.to_usize(), hi_opt.map(|hi| hi.col.to_usize()));
if let Some(line) = line_opt {
if let Some(lo) = line.char_indices().map(|(i, _)| i).nth(lo) {
// Get high index while account for rare unicode and emoji with char_indices
let hi_opt = hi_opt.and_then(|hi| line.char_indices().map(|(i, _)| i).nth(hi));
match hi_opt {
// If high index exist, take string from low to high index
Some(hi) if hi > lo => {
// count how many '\n' exist
line_count = line[lo..hi].matches('\n').count();
buf.push_str(&line[lo..hi])
}
Some(_) => (),
// If high index absence, take string from low index till end string.len
None => {
// count how many '\n' exist
line_count = line[lo..].matches('\n').count();
buf.push_str(&line[lo..])
}
}
}
// If high index is None
if hi_opt.is_none() {
buf.push('\n');
}
}
line_count
}
2017-11-03 16:17:33 +01:00
assert!(!self.substitutions.is_empty());
self.substitutions
.iter()
2020-01-16 18:27:18 -08:00
.filter(|subst| {
// Suggestions coming from macros can have malformed spans. This is a heavy
// handed approach to avoid ICEs by ignoring the suggestion outright.
2020-02-22 16:07:05 +02:00
let invalid = subst.parts.iter().any(|item| sm.is_valid_span(item.span).is_err());
2020-01-16 18:27:18 -08:00
if invalid {
debug!("splice_lines: suggestion contains an invalid span: {:?}", subst);
}
!invalid
})
2017-11-03 16:17:33 +01:00
.cloned()
.filter_map(|mut substitution| {
2017-11-03 16:17:33 +01:00
// Assumption: all spans are in the same file, and all spans
// are disjoint. Sort in ascending order.
substitution.parts.sort_by_key(|part| part.span.lo());
2019-12-22 17:42:04 -05:00
2017-11-03 16:17:33 +01:00
// Find the bounding span.
let lo = substitution.parts.iter().map(|part| part.span.lo()).min()?;
let hi = substitution.parts.iter().map(|part| part.span.hi()).max()?;
let bounding_span = Span::with_root_ctxt(lo, hi);
// The different spans might belong to different contexts, if so ignore suggestion.
2020-02-22 16:07:05 +02:00
let lines = sm.span_to_lines(bounding_span).ok()?;
assert!(!lines.lines.is_empty() || bounding_span.is_dummy());
2019-12-22 17:42:04 -05:00
// We can't splice anything if the source is unavailable.
if !sm.ensure_source_file_source_present(&lines.file) {
return None;
}
let mut highlights = vec![];
2017-11-03 16:17:33 +01:00
// To build up the result, we do this for each span:
// - push the line segment trailing the previous span
// (at the beginning a "phantom" span pointing at the start of the line)
// - push lines between the previous and current span (if any)
// - if the previous and current span are not on the same line
// push the line segment leading up to the current span
// - splice in the span substitution
//
// Finally push the trailing line segment of the last span
2020-02-22 16:07:05 +02:00
let sf = &lines.file;
let mut prev_hi = sm.lookup_char_pos(bounding_span.lo());
2017-11-03 16:17:33 +01:00
prev_hi.col = CharPos::from_usize(0);
let mut prev_line =
lines.lines.get(0).and_then(|line0| sf.get_line(line0.line_index));
2017-11-03 16:17:33 +01:00
let mut buf = String::new();
2019-12-22 17:42:04 -05:00
let mut line_highlight = vec![];
// We need to keep track of the difference between the existing code and the added
// or deleted code in order to point at the correct column *after* substitution.
let mut acc = 0;
let mut only_capitalization = false;
for part in &mut substitution.parts {
// If this is a replacement of, e.g. `"a"` into `"ab"`, adjust the
// suggestion and snippet to look as if we just suggested to add
// `"b"`, which is typically much easier for the user to understand.
part.trim_trivial_replacements(sm);
only_capitalization |= is_case_difference(sm, &part.snippet, part.span);
2020-02-22 16:07:05 +02:00
let cur_lo = sm.lookup_char_pos(part.span.lo());
if prev_hi.line == cur_lo.line {
let mut count =
push_trailing(&mut buf, prev_line.as_ref(), &prev_hi, Some(&cur_lo));
while count > 0 {
highlights.push(std::mem::take(&mut line_highlight));
acc = 0;
count -= 1;
}
} else {
2021-08-17 15:46:42 +00:00
acc = 0;
highlights.push(std::mem::take(&mut line_highlight));
let mut count = push_trailing(&mut buf, prev_line.as_ref(), &prev_hi, None);
while count > 0 {
highlights.push(std::mem::take(&mut line_highlight));
count -= 1;
}
// push lines between the previous and current span (if any)
for idx in prev_hi.line..(cur_lo.line - 1) {
2020-02-22 16:07:05 +02:00
if let Some(line) = sf.get_line(idx) {
buf.push_str(line.as_ref());
buf.push('\n');
highlights.push(std::mem::take(&mut line_highlight));
2019-12-22 17:42:04 -05:00
}
}
2020-02-22 16:07:05 +02:00
if let Some(cur_line) = sf.get_line(cur_lo.line - 1) {
2020-04-12 11:36:37 -04:00
let end = match cur_line.char_indices().nth(cur_lo.col.to_usize()) {
Some((i, _)) => i,
None => cur_line.len(),
};
buf.push_str(&cur_line[..end]);
}
}
// Add a whole line highlight per line in the snippet.
2021-08-23 12:42:08 +00:00
let len: isize = part
.snippet
.split('\n')
.next()
.unwrap_or(&part.snippet)
.chars()
.map(|c| match c {
'\t' => 4,
_ => 1,
})
.sum();
if !is_different(sm, &part.snippet, part.span) {
// Account for cases where we are suggesting the same code that's already
// there. This shouldn't happen often, but in some cases for multipart
// suggestions it's much easier to handle it here than in the origin.
} else {
line_highlight.push(SubstitutionHighlight {
start: (cur_lo.col.0 as isize + acc) as usize,
end: (cur_lo.col.0 as isize + acc + len) as usize,
});
}
buf.push_str(&part.snippet);
2021-08-23 12:42:08 +00:00
let cur_hi = sm.lookup_char_pos(part.span.hi());
2023-05-19 00:44:14 +02:00
// Account for the difference between the width of the current code and the
// snippet being suggested, so that the *later* suggestions are correctly
// aligned on the screen. Note that cur_hi and cur_lo can be on different
// lines, so cur_hi.col can be smaller than cur_lo.col
acc += len - (cur_hi.col.0 as isize - cur_lo.col.0 as isize);
2021-08-23 12:42:08 +00:00
prev_hi = cur_hi;
prev_line = sf.get_line(prev_hi.line - 1);
for line in part.snippet.split('\n').skip(1) {
acc = 0;
highlights.push(std::mem::take(&mut line_highlight));
2021-08-23 12:42:08 +00:00
let end: usize = line
.chars()
.map(|c| match c {
'\t' => 4,
_ => 1,
})
.sum();
line_highlight.push(SubstitutionHighlight { start: 0, end });
}
}
highlights.push(std::mem::take(&mut line_highlight));
2017-05-10 13:19:29 +02:00
// if the replacement already ends with a newline, don't print the next line
if !buf.ends_with('\n') {
2017-11-03 16:17:33 +01:00
push_trailing(&mut buf, prev_line.as_ref(), &prev_hi, None);
2017-05-10 13:19:29 +02:00
}
// remove trailing newlines
while buf.ends_with('\n') {
buf.pop();
}
if highlights.iter().all(|parts| parts.is_empty()) {
None
} else {
Some((buf, substitution.parts, highlights, only_capitalization))
}
2017-11-03 16:17:33 +01:00
})
.collect()
}
}
/// Signifies that the compiler died with an explicit call to `.bug`
/// or `.span_bug` rather than a failed assertion, etc.
pub struct ExplicitBug;
/// Signifies that the compiler died due to a delayed bug rather than a failed
/// assertion, etc.
pub struct DelayedBugPanic;
/// A `DiagCtxt` deals with errors and other compiler output.
/// Certain errors (fatal, bug, unimpl) may cause immediate exit,
/// others log errors for later reporting.
2023-12-17 21:48:57 +11:00
pub struct DiagCtxt {
inner: Lock<DiagCtxtInner>,
}
2017-11-20 18:03:20 +00:00
#[derive(Copy, Clone)]
pub struct DiagCtxtHandle<'a> {
dcx: &'a DiagCtxt,
2024-06-27 09:45:26 +00:00
/// Some contexts create `DiagCtxtHandle` with this field set, and thus all
/// errors emitted with it will automatically taint when emitting errors.
tainted_with_errors: Option<&'a Cell<Option<ErrorGuaranteed>>>,
}
impl<'a> std::ops::Deref for DiagCtxtHandle<'a> {
type Target = &'a DiagCtxt;
fn deref(&self) -> &Self::Target {
&self.dcx
}
}
2019-09-23 22:28:14 +02:00
/// This inner struct exists to keep it all behind a single lock;
/// this is done to prevent possible deadlocks in a multi-threaded compiler,
/// as well as inconsistent state observation.
struct DiagCtxtInner {
flags: DiagCtxtFlags,
registry: Registry,
/// The error guarantees from all emitted errors. The length gives the error count.
err_guars: Vec<ErrorGuaranteed>,
/// The error guarantee from all emitted lint errors. The length gives the
/// lint error count.
lint_err_guars: Vec<ErrorGuaranteed>,
/// The delayed bugs and their error guarantees.
delayed_bugs: Vec<(DelayedDiagInner, ErrorGuaranteed)>,
/// The error count shown to the user at the end.
deduplicated_err_count: usize,
/// The warning count shown to the user at the end.
deduplicated_warn_count: usize,
emitter: Box<DynEmitter>,
/// Must we produce a diagnostic to justify the use of the expensive
2024-03-10 16:24:48 +00:00
/// `trimmed_def_paths` function? Backtrace is the location of the call.
must_produce_diag: Option<Backtrace>,
/// Has this diagnostic context printed any diagnostics? (I.e. has
/// `self.emitter.emit_diagnostic()` been called?
has_printed: bool,
/// This flag indicates that an expected diagnostic was emitted and suppressed.
/// This is used for the `must_produce_diag` check.
suppressed_expected_diag: bool,
/// This set contains the code of all emitted diagnostics to avoid
/// emitting the same diagnostic with extended help (`--teach`) twice, which
2021-04-19 15:57:08 +03:00
/// would be unnecessary repetition.
taught_diagnostics: FxHashSet<ErrCode>,
2018-03-03 06:20:26 +01:00
2022-10-08 14:42:38 +02:00
/// Used to suggest rustc --explain `<error code>`
emitted_diagnostic_codes: FxIndexSet<ErrCode>,
/// This set contains a hash of every diagnostic that has been emitted by
/// this `DiagCtxt`. These hashes is used to avoid emitting the same error
/// twice.
emitted_diagnostics: FxHashSet<Hash128>,
2019-09-23 04:45:21 +02:00
/// Stashed diagnostics emitted in one stage of the compiler that may be
Overhaul the handling of errors at the top-level. Currently `emit_stashed_diagnostic` is called from four(!) different places: `print_error_count`, `DiagCtxtInner::drop`, `abort_if_errors`, and `compile_status`. And `flush_delayed` is called from two different places: `DiagCtxtInner::drop` and `Queries`. This is pretty gross! Each one should really be called from a single place, but there's a bunch of entanglements. This commit cleans up this mess. Specifically, it: - Removes all the existing calls to `emit_stashed_diagnostic`, and adds a single new call in `finish_diagnostics`. - Removes the early `flush_delayed` call in `codegen_and_build_linker`, replacing it with a simple early return if delayed bugs are present. - Changes `DiagCtxtInner::drop` and `DiagCtxtInner::flush_delayed` so they both assert that the stashed diagnostics are empty (i.e. processed beforehand). - Changes `interface::run_compiler` so that any errors emitted during `finish_diagnostics` (i.e. late-emitted stashed diagnostics) are counted and cannot be overlooked. This requires adding `ErrorGuaranteed` return values to several functions. - Removes the `stashed_err_count` call in `analysis`. This is possible now that we don't have to worry about calling `flush_delayed` early from `codegen_and_build_linker` when stashed diagnostics are pending. - Changes the `span_bug` case in `handle_tuple_field_pattern_match` to a `delayed_span_bug`, because it now can be reached due to the removal of the `stashed_err_count` call in `analysis`. - Slightly changes the expected output of three tests. If no errors are emitted but there are delayed bugs, the error count is no longer printed. This is because delayed bugs are now always printed after the error count is printed (or not printed, if the error count is zero). There is a lot going on in this commit. It's hard to break into smaller pieces because the existing code is very tangled. It took me a long time and a lot of effort to understand how the different pieces interact, and I think the new code is a lot simpler and easier to understand.
2024-02-19 10:00:19 +11:00
/// stolen and emitted/cancelled by other stages (e.g. to improve them and
/// add more information). All stashed diagnostics must be emitted with
/// `emit_stashed_diagnostics` by the time the `DiagCtxtInner` is dropped,
/// otherwise an assertion failure will occur.
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
stashed_diagnostics: FxIndexMap<(Span, StashKey), (DiagInner, Option<ErrorGuaranteed>)>,
future_breakage_diagnostics: Vec<DiagInner>,
/// expected diagnostic will have the level `Expect` which additionally
/// carries the [`LintExpectationId`] of the expectation that can be
/// marked as fulfilled. This is a collection of all [`LintExpectationId`]s
/// that have been marked as fulfilled this way.
///
2024-09-07 01:34:22 +00:00
/// Emitting expectations after having stolen this field can happen. In particular, an
/// `#[expect(warnings)]` can easily make the `UNFULFILLED_LINT_EXPECTATIONS` lint expect
/// itself. To avoid needless complexity in this corner case, we tolerate failing to track
/// those expectations.
///
/// [RFC-2383]: https://rust-lang.github.io/rfcs/2383-lint-reasons.html
2024-08-31 14:32:15 +00:00
fulfilled_expectations: FxIndexSet<LintExpectationId>,
/// The file where the ICE information is stored. This allows delayed_span_bug backtraces to be
/// stored along side the main panic backtrace.
ice_file: Option<PathBuf>,
2019-09-23 04:45:21 +02:00
}
/// A key denoting where from a diagnostic was stashed.
2024-04-05 22:14:10 -04:00
#[derive(Copy, Clone, PartialEq, Eq, Hash, Debug)]
2019-09-23 04:45:21 +02:00
pub enum StashKey {
ItemNoType,
UnderscoreForArrayLengths,
EarlySyntaxWarning,
CallIntoMethod,
/// When an invalid lifetime e.g. `'2` should be reinterpreted
/// as a char literal in the parser
LifetimeIsChar,
/// Maybe there was a typo where a comma was forgotten before
/// FRU syntax
MaybeFruTypo,
CallAssocMethod,
AssociatedTypeSuggestion,
/// Query cycle detected, stashing in favor of a better error.
Cycle,
UndeterminedMacroResolution,
/// Used by `Parser::maybe_recover_trailing_expr`
ExprInPat,
/// If in the parser we detect a field expr with turbofish generic params it's possible that
/// it's a method call without parens. If later on in `hir_typeck` we find out that this is
/// the case we suppress this message and we give a better suggestion.
GenericInFieldExpr,
}
fn default_track_diagnostic<R>(diag: DiagInner, f: &mut dyn FnMut(DiagInner) -> R) -> R {
(*f)(diag)
}
/// Diagnostics emitted by `DiagCtxtInner::emit_diagnostic` are passed through this function. Used
/// for tracking by incremental, to replay diagnostics as necessary.
pub static TRACK_DIAGNOSTIC: AtomicRef<
fn(DiagInner, &mut dyn FnMut(DiagInner) -> Option<ErrorGuaranteed>) -> Option<ErrorGuaranteed>,
> = AtomicRef::new(&(default_track_diagnostic as _));
#[derive(Copy, Clone, Default)]
pub struct DiagCtxtFlags {
/// If false, warning-level lints are suppressed.
/// (rustc: see `--allow warnings` and `--cap-lints`)
2017-11-20 18:03:20 +00:00
pub can_emit_warnings: bool,
/// If Some, the Nth error-level diagnostic is upgraded to bug-level.
/// (rustc: see `-Z treat-err-as-bug`)
2024-01-29 23:59:09 +01:00
pub treat_err_as_bug: Option<NonZero<usize>>,
/// Eagerly emit delayed bugs as errors, so that the compiler debugger may
/// see all of the errors being emitted at once.
pub eagerly_emit_delayed_bugs: bool,
/// Show macro backtraces.
/// (rustc: see `-Z macro-backtrace`)
pub macro_backtrace: bool,
/// If true, identical diagnostics are reported only once.
pub deduplicate_diagnostics: bool,
2022-10-19 00:08:20 +02:00
/// Track where errors are created. Enabled with `-Ztrack-diagnostics`.
pub track_diagnostics: bool,
2017-11-20 18:03:20 +00:00
}
impl Drop for DiagCtxtInner {
fn drop(&mut self) {
// For tools using `interface::run_compiler` (e.g. rustc, rustdoc)
// stashed diagnostics will have already been emitted. But for others
// that don't use `interface::run_compiler` (e.g. rustfmt, some clippy
// lints) this fallback is necessary.
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
//
// Important: it is sound to produce an `ErrorGuaranteed` when stashing
// errors because they are guaranteed to be emitted here or earlier.
self.emit_stashed_diagnostics();
2019-09-23 04:45:21 +02:00
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
// Important: it is sound to produce an `ErrorGuaranteed` when emitting
// delayed bugs because they are guaranteed to be emitted here if
// necessary.
self.flush_delayed();
// Sanity check: did we use some of the expensive `trimmed_def_paths` functions
// unexpectedly, that is, without producing diagnostics? If so, for debugging purposes, we
// suggest where this happened and how to avoid it.
if !self.has_printed && !self.suppressed_expected_diag && !std::thread::panicking() {
2024-03-10 16:24:48 +00:00
if let Some(backtrace) = &self.must_produce_diag {
let suggestion = match backtrace.status() {
BacktraceStatus::Disabled => String::from(
"Backtraces are currently disabled: set `RUST_BACKTRACE=1` and re-run \
to see where it happened.",
),
BacktraceStatus::Captured => format!(
"This happened in the following `must_produce_diag` call's backtrace:\n\
{backtrace}",
),
_ => String::from("(impossible to capture backtrace where this happened)"),
};
panic!(
"`trimmed_def_paths` called, diagnostics were expected but none were emitted. \
Use `with_no_trimmed_paths` for debugging. {suggestion}"
);
}
}
}
}
2023-12-17 21:48:57 +11:00
impl DiagCtxt {
pub fn disable_warnings(mut self) -> Self {
self.inner.get_mut().flags.can_emit_warnings = false;
self
2019-09-23 22:28:14 +02:00
}
pub fn with_flags(mut self, flags: DiagCtxtFlags) -> Self {
self.inner.get_mut().flags = flags;
self
}
pub fn with_ice_file(mut self, ice_file: PathBuf) -> Self {
self.inner.get_mut().ice_file = Some(ice_file);
self
}
pub fn with_registry(mut self, registry: Registry) -> Self {
self.inner.get_mut().registry = registry;
self
}
pub fn new(emitter: Box<DynEmitter>) -> Self {
Self { inner: Lock::new(DiagCtxtInner::new(emitter)) }
}
pub fn make_silent(&self, fatal_note: Option<String>, emit_fatal_diagnostic: bool) {
// An empty type that implements `Emitter` to temporarily swap in place of the real one,
// which will be used in constructing its replacement.
struct FalseEmitter;
impl Emitter for FalseEmitter {
fn emit_diagnostic(&mut self, _: DiagInner, _: &Registry) {
unimplemented!("false emitter must only used during `make_silent`")
}
fn source_map(&self) -> Option<&SourceMap> {
unimplemented!("false emitter must only used during `make_silent`")
}
}
impl translation::Translate for FalseEmitter {
fn fluent_bundle(&self) -> Option<&FluentBundle> {
unimplemented!("false emitter must only used during `make_silent`")
}
fn fallback_fluent_bundle(&self) -> &FluentBundle {
unimplemented!("false emitter must only used during `make_silent`")
}
}
let mut inner = self.inner.borrow_mut();
let mut prev_emitter = Box::new(FalseEmitter) as Box<dyn Emitter + DynSend>;
std::mem::swap(&mut inner.emitter, &mut prev_emitter);
let new_emitter = Box::new(emitter::SilentEmitter {
fatal_emitter: prev_emitter,
fatal_note,
emit_fatal_diagnostic,
});
inner.emitter = new_emitter;
}
pub fn set_emitter(&self, emitter: Box<dyn Emitter + DynSend>) {
self.inner.borrow_mut().emitter = emitter;
}
/// Translate `message` eagerly with `args` to `SubdiagMessage::Eager`.
pub fn eagerly_translate<'a>(
&self,
message: DiagMessage,
args: impl Iterator<Item = DiagArg<'a>>,
) -> SubdiagMessage {
let inner = self.inner.borrow();
inner.eagerly_translate(message, args)
}
/// Translate `message` eagerly with `args` to `String`.
pub fn eagerly_translate_to_string<'a>(
&self,
message: DiagMessage,
args: impl Iterator<Item = DiagArg<'a>>,
) -> String {
let inner = self.inner.borrow();
inner.eagerly_translate_to_string(message, args)
}
// This is here to not allow mutation of flags;
// as of this writing it's used in Session::consider_optimizing and
// in tests in rustc_interface.
pub fn can_emit_warnings(&self) -> bool {
self.inner.borrow_mut().flags.can_emit_warnings
}
/// Resets the diagnostic error count as well as the cached emitted diagnostics.
///
2019-02-08 14:53:55 +01:00
/// NOTE: *do not* call this function from rustc. It is only meant to be called from external
/// tools that want to reuse a `Parser` cleaning the previously emitted diagnostics as well as
/// the overall count of emitted error diagnostics.
pub fn reset_err_count(&self) {
// Use destructuring so that if a field gets added to `DiagCtxtInner`, it's impossible to
// fail to update this method as well.
let mut inner = self.inner.borrow_mut();
let DiagCtxtInner {
flags: _,
registry: _,
err_guars,
lint_err_guars,
delayed_bugs,
deduplicated_err_count,
deduplicated_warn_count,
emitter: _,
must_produce_diag,
has_printed,
suppressed_expected_diag,
taught_diagnostics,
emitted_diagnostic_codes,
emitted_diagnostics,
stashed_diagnostics,
future_breakage_diagnostics,
fulfilled_expectations,
ice_file: _,
} = inner.deref_mut();
// For the `Vec`s and `HashMap`s, we overwrite with an empty container to free the
// underlying memory (which `clear` would not do).
*err_guars = Default::default();
*lint_err_guars = Default::default();
*delayed_bugs = Default::default();
*deduplicated_err_count = 0;
*deduplicated_warn_count = 0;
2024-03-10 16:24:48 +00:00
*must_produce_diag = None;
*has_printed = false;
*suppressed_expected_diag = false;
*taught_diagnostics = Default::default();
*emitted_diagnostic_codes = Default::default();
*emitted_diagnostics = Default::default();
*stashed_diagnostics = Default::default();
*future_breakage_diagnostics = Default::default();
*fulfilled_expectations = Default::default();
2019-09-23 04:45:21 +02:00
}
pub fn handle<'a>(&'a self) -> DiagCtxtHandle<'a> {
DiagCtxtHandle { dcx: self, tainted_with_errors: None }
}
2024-06-27 09:45:26 +00:00
/// Link this to a taintable context so that emitting errors will automatically set
/// the `Option<ErrorGuaranteed>` instead of having to do that manually at every error
/// emission site.
pub fn taintable_handle<'a>(
&'a self,
tainted_with_errors: &'a Cell<Option<ErrorGuaranteed>>,
) -> DiagCtxtHandle<'a> {
DiagCtxtHandle { dcx: self, tainted_with_errors: Some(tainted_with_errors) }
}
}
impl<'a> DiagCtxtHandle<'a> {
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
/// Stashes a diagnostic for possible later improvement in a different,
/// later stage of the compiler. Possible actions depend on the diagnostic
/// level:
/// - Level::Bug, Level:Fatal: not allowed, will trigger a panic.
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
/// - Level::Error: immediately counted as an error that has occurred, because it
/// is guaranteed to be emitted eventually. Can be later accessed with the
/// provided `span` and `key` through
/// [`DiagCtxtHandle::try_steal_modify_and_emit_err`] or
/// [`DiagCtxtHandle::try_steal_replace_and_emit_err`]. These do not allow
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
/// cancellation or downgrading of the error. Returns
/// `Some(ErrorGuaranteed)`.
/// - Level::DelayedBug: this does happen occasionally with errors that are
/// downgraded to delayed bugs. It is not stashed, but immediately
/// emitted as a delayed bug. This is because stashing it would cause it
/// to be counted by `err_count` which we don't want. It doesn't matter
/// that we cannot steal and improve it later, because it's not a
/// user-facing error. Returns `Some(ErrorGuaranteed)` as is normal for
/// delayed bugs.
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
/// - Level::Warning and lower (i.e. !is_error()): can be accessed with the
/// provided `span` and `key` through [`DiagCtxtHandle::steal_non_err()`]. This
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
/// allows cancelling and downgrading of the diagnostic. Returns `None`.
pub fn stash_diagnostic(
&self,
span: Span,
key: StashKey,
diag: DiagInner,
) -> Option<ErrorGuaranteed> {
let guar = match diag.level {
Bug | Fatal => {
self.span_bug(
span,
format!("invalid level in `stash_diagnostic`: {:?}", diag.level),
);
}
// We delay a bug here so that `-Ztreat-err-as-bug -Zeagerly-emit-delayed-bugs`
// can be used to create a backtrace at the stashing site instead of whenever the
// diagnostic context is dropped and thus delayed bugs are emitted.
2024-04-05 22:14:10 -04:00
Error => Some(self.span_delayed_bug(span, format!("stashing {key:?}"))),
DelayedBug => {
return self.inner.borrow_mut().emit_diagnostic(diag, self.tainted_with_errors);
}
2025-03-18 17:18:54 +00:00
ForceWarning | Warning | Note | OnceNote | Help | OnceHelp | FailureNote | Allow
| Expect => None,
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
};
// FIXME(Centril, #69537): Consider reintroducing panic on overwriting a stashed diagnostic
// if/when we have a more robust macro-friendly replacement for `(span, key)` as a key.
// See the PR for a discussion.
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
let key = (span.with_parent(None), key);
self.inner.borrow_mut().stashed_diagnostics.insert(key, (diag, guar));
guar
2019-09-23 04:45:21 +02:00
}
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
/// Steal a previously stashed non-error diagnostic with the given `Span`
/// and [`StashKey`] as the key. Panics if the found diagnostic is an
/// error.
pub fn steal_non_err(self, span: Span, key: StashKey) -> Option<Diag<'a, ()>> {
let key = (span.with_parent(None), key);
// FIXME(#120456) - is `swap_remove` correct?
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
let (diag, guar) = self.inner.borrow_mut().stashed_diagnostics.swap_remove(&key)?;
assert!(!diag.is_error());
assert!(guar.is_none());
Some(Diag::new_diagnostic(self, diag))
2019-09-23 04:45:21 +02:00
}
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
/// Steals a previously stashed error with the given `Span` and
/// [`StashKey`] as the key, modifies it, and emits it. Returns `None` if
/// no matching diagnostic is found. Panics if the found diagnostic's level
/// isn't `Level::Error`.
pub fn try_steal_modify_and_emit_err<F>(
self,
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
span: Span,
key: StashKey,
mut modify_err: F,
) -> Option<ErrorGuaranteed>
where
F: FnMut(&mut Diag<'_>),
{
let key = (span.with_parent(None), key);
// FIXME(#120456) - is `swap_remove` correct?
let err = self.inner.borrow_mut().stashed_diagnostics.swap_remove(&key);
err.map(|(err, guar)| {
// The use of `::<ErrorGuaranteed>` is safe because level is `Level::Error`.
assert_eq!(err.level, Error);
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
assert!(guar.is_some());
let mut err = Diag::<ErrorGuaranteed>::new_diagnostic(self, err);
modify_err(&mut err);
assert_eq!(err.level, Error);
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
err.emit()
})
}
/// Steals a previously stashed error with the given `Span` and
/// [`StashKey`] as the key, cancels it if found, and emits `new_err`.
/// Panics if the found diagnostic's level isn't `Level::Error`.
pub fn try_steal_replace_and_emit_err(
self,
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
span: Span,
key: StashKey,
new_err: Diag<'_>,
) -> ErrorGuaranteed {
let key = (span.with_parent(None), key);
// FIXME(#120456) - is `swap_remove` correct?
let old_err = self.inner.borrow_mut().stashed_diagnostics.swap_remove(&key);
match old_err {
Some((old_err, guar)) => {
assert_eq!(old_err.level, Error);
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
assert!(guar.is_some());
// Because `old_err` has already been counted, it can only be
// safely cancelled because the `new_err` supplants it.
Diag::<ErrorGuaranteed>::new_diagnostic(self, old_err).cancel();
}
None => {}
};
new_err.emit()
}
pub fn has_stashed_diagnostic(&self, span: Span, key: StashKey) -> bool {
self.inner.borrow().stashed_diagnostics.get(&(span.with_parent(None), key)).is_some()
}
2019-09-23 04:45:21 +02:00
/// Emit all stashed diagnostics.
Overhaul the handling of errors at the top-level. Currently `emit_stashed_diagnostic` is called from four(!) different places: `print_error_count`, `DiagCtxtInner::drop`, `abort_if_errors`, and `compile_status`. And `flush_delayed` is called from two different places: `DiagCtxtInner::drop` and `Queries`. This is pretty gross! Each one should really be called from a single place, but there's a bunch of entanglements. This commit cleans up this mess. Specifically, it: - Removes all the existing calls to `emit_stashed_diagnostic`, and adds a single new call in `finish_diagnostics`. - Removes the early `flush_delayed` call in `codegen_and_build_linker`, replacing it with a simple early return if delayed bugs are present. - Changes `DiagCtxtInner::drop` and `DiagCtxtInner::flush_delayed` so they both assert that the stashed diagnostics are empty (i.e. processed beforehand). - Changes `interface::run_compiler` so that any errors emitted during `finish_diagnostics` (i.e. late-emitted stashed diagnostics) are counted and cannot be overlooked. This requires adding `ErrorGuaranteed` return values to several functions. - Removes the `stashed_err_count` call in `analysis`. This is possible now that we don't have to worry about calling `flush_delayed` early from `codegen_and_build_linker` when stashed diagnostics are pending. - Changes the `span_bug` case in `handle_tuple_field_pattern_match` to a `delayed_span_bug`, because it now can be reached due to the removal of the `stashed_err_count` call in `analysis`. - Slightly changes the expected output of three tests. If no errors are emitted but there are delayed bugs, the error count is no longer printed. This is because delayed bugs are now always printed after the error count is printed (or not printed, if the error count is zero). There is a lot going on in this commit. It's hard to break into smaller pieces because the existing code is very tangled. It took me a long time and a lot of effort to understand how the different pieces interact, and I think the new code is a lot simpler and easier to understand.
2024-02-19 10:00:19 +11:00
pub fn emit_stashed_diagnostics(&self) -> Option<ErrorGuaranteed> {
self.inner.borrow_mut().emit_stashed_diagnostics()
}
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
/// This excludes delayed bugs.
#[inline]
pub fn err_count(&self) -> usize {
let inner = self.inner.borrow();
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
inner.err_guars.len()
+ inner.lint_err_guars.len()
+ inner.stashed_diagnostics.values().filter(|(_diag, guar)| guar.is_some()).count()
}
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
/// This excludes lint errors and delayed bugs. Unless absolutely
/// necessary, prefer `has_errors` to this method.
pub fn has_errors_excluding_lint_errors(&self) -> Option<ErrorGuaranteed> {
self.inner.borrow().has_errors_excluding_lint_errors()
}
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
/// This excludes delayed bugs.
pub fn has_errors(&self) -> Option<ErrorGuaranteed> {
self.inner.borrow().has_errors()
}
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
/// This excludes nothing. Unless absolutely necessary, prefer `has_errors`
/// to this method.
pub fn has_errors_or_delayed_bugs(&self) -> Option<ErrorGuaranteed> {
self.inner.borrow().has_errors_or_delayed_bugs()
}
pub fn print_error_count(&self) {
let mut inner = self.inner.borrow_mut();
Overhaul the handling of errors at the top-level. Currently `emit_stashed_diagnostic` is called from four(!) different places: `print_error_count`, `DiagCtxtInner::drop`, `abort_if_errors`, and `compile_status`. And `flush_delayed` is called from two different places: `DiagCtxtInner::drop` and `Queries`. This is pretty gross! Each one should really be called from a single place, but there's a bunch of entanglements. This commit cleans up this mess. Specifically, it: - Removes all the existing calls to `emit_stashed_diagnostic`, and adds a single new call in `finish_diagnostics`. - Removes the early `flush_delayed` call in `codegen_and_build_linker`, replacing it with a simple early return if delayed bugs are present. - Changes `DiagCtxtInner::drop` and `DiagCtxtInner::flush_delayed` so they both assert that the stashed diagnostics are empty (i.e. processed beforehand). - Changes `interface::run_compiler` so that any errors emitted during `finish_diagnostics` (i.e. late-emitted stashed diagnostics) are counted and cannot be overlooked. This requires adding `ErrorGuaranteed` return values to several functions. - Removes the `stashed_err_count` call in `analysis`. This is possible now that we don't have to worry about calling `flush_delayed` early from `codegen_and_build_linker` when stashed diagnostics are pending. - Changes the `span_bug` case in `handle_tuple_field_pattern_match` to a `delayed_span_bug`, because it now can be reached due to the removal of the `stashed_err_count` call in `analysis`. - Slightly changes the expected output of three tests. If no errors are emitted but there are delayed bugs, the error count is no longer printed. This is because delayed bugs are now always printed after the error count is printed (or not printed, if the error count is zero). There is a lot going on in this commit. It's hard to break into smaller pieces because the existing code is very tangled. It took me a long time and a lot of effort to understand how the different pieces interact, and I think the new code is a lot simpler and easier to understand.
2024-02-19 10:00:19 +11:00
// Any stashed diagnostics should have been handled by
// `emit_stashed_diagnostics` by now.
assert!(inner.stashed_diagnostics.is_empty());
if inner.treat_err_as_bug() {
return;
}
let warnings = match inner.deduplicated_warn_count {
0 => Cow::from(""),
1 => Cow::from("1 warning emitted"),
count => Cow::from(format!("{count} warnings emitted")),
};
let errors = match inner.deduplicated_err_count {
0 => Cow::from(""),
1 => Cow::from("aborting due to 1 previous error"),
count => Cow::from(format!("aborting due to {count} previous errors")),
};
match (errors.len(), warnings.len()) {
(0, 0) => return,
(0, _) => {
// Use `ForceWarning` rather than `Warning` to guarantee emission, e.g. with a
// configuration like `--cap-lints allow --force-warn bare_trait_objects`.
inner.emit_diagnostic(
2025-03-18 17:18:54 +00:00
DiagInner::new(ForceWarning, DiagMessage::Str(warnings)),
None,
);
}
(_, 0) => {
inner.emit_diagnostic(DiagInner::new(Error, errors), self.tainted_with_errors);
}
(_, _) => {
inner.emit_diagnostic(
DiagInner::new(Error, format!("{errors}; {warnings}")),
self.tainted_with_errors,
);
}
}
let can_show_explain = inner.emitter.should_show_explain();
let are_there_diagnostics = !inner.emitted_diagnostic_codes.is_empty();
if can_show_explain && are_there_diagnostics {
let mut error_codes = inner
.emitted_diagnostic_codes
.iter()
.filter_map(|&code| {
if inner.registry.try_find_description(code).is_ok() {
Some(code.to_string())
} else {
None
}
})
.collect::<Vec<_>>();
if !error_codes.is_empty() {
error_codes.sort();
if error_codes.len() > 1 {
let limit = if error_codes.len() > 9 { 9 } else { error_codes.len() };
let msg1 = format!(
"Some errors have detailed explanations: {}{}",
error_codes[..limit].join(", "),
if error_codes.len() > 9 { "..." } else { "." }
);
let msg2 = format!(
"For more information about an error, try `rustc --explain {}`.",
&error_codes[0]
);
inner.emit_diagnostic(DiagInner::new(FailureNote, msg1), None);
inner.emit_diagnostic(DiagInner::new(FailureNote, msg2), None);
} else {
let msg = format!(
"For more information about this error, try `rustc --explain {}`.",
&error_codes[0]
);
inner.emit_diagnostic(DiagInner::new(FailureNote, msg), None);
}
}
}
}
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
/// This excludes delayed bugs. Used for early aborts after errors occurred
/// -- e.g. because continuing in the face of errors is likely to lead to
/// bad results, such as spurious/uninteresting additional errors -- when
/// returning an error `Result` is difficult.
pub fn abort_if_errors(&self) {
if let Some(guar) = self.has_errors() {
guar.raise_fatal();
}
}
/// `true` if we haven't taught a diagnostic with this code already.
/// The caller must then teach the user about such a diagnostic.
///
/// Used to suppress emitting the same error multiple times with extended explanation when
/// calling `-Zteach`.
pub fn must_teach(&self, code: ErrCode) -> bool {
self.inner.borrow_mut().taught_diagnostics.insert(code)
}
pub fn emit_diagnostic(&self, diagnostic: DiagInner) -> Option<ErrorGuaranteed> {
self.inner.borrow_mut().emit_diagnostic(diagnostic, self.tainted_with_errors)
}
pub fn emit_artifact_notification(&self, path: &Path, artifact_type: &str) {
self.inner.borrow_mut().emitter.emit_artifact_notification(path, artifact_type);
}
pub fn emit_future_breakage_report(&self) {
let inner = &mut *self.inner.borrow_mut();
let diags = std::mem::take(&mut inner.future_breakage_diagnostics);
if !diags.is_empty() {
inner.emitter.emit_future_breakage_report(diags, &inner.registry);
}
}
pub fn emit_unused_externs(
&self,
lint_level: rustc_lint_defs::Level,
loud: bool,
unused_externs: &[&str],
) {
let mut inner = self.inner.borrow_mut();
// This "error" is an odd duck.
// - It's only produce with JSON output.
// - It's not emitted the usual way, via `emit_diagnostic`.
// - The `$message_type` field is "unused_externs" rather than the usual
// "diagnosic".
//
// We count it as a lint error because it has a lint level. The value
// of `loud` (which comes from "unused-externs" or
// "unused-externs-silent"), also affects whether it's treated like a
// hard error or not.
if loud && lint_level.is_error() {
// This `unchecked_error_guaranteed` is valid. It is where the
// `ErrorGuaranteed` for unused_extern errors originates.
#[allow(deprecated)]
inner.lint_err_guars.push(ErrorGuaranteed::unchecked_error_guaranteed());
inner.panic_if_treat_err_as_bug();
}
inner.emitter.emit_unused_externs(lint_level, unused_externs)
}
/// This methods steals all [`LintExpectationId`]s that are stored inside
/// [`DiagCtxtInner`] and indicate that the linked expectation has been fulfilled.
#[must_use]
2024-08-31 14:32:15 +00:00
pub fn steal_fulfilled_expectation_ids(&self) -> FxIndexSet<LintExpectationId> {
std::mem::take(&mut self.inner.borrow_mut().fulfilled_expectations)
}
pub fn flush_delayed(&self) {
self.inner.borrow_mut().flush_delayed();
}
/// Used when trimmed_def_paths is called and we must produce a diagnostic
/// to justify its cost.
2024-03-10 16:24:48 +00:00
#[track_caller]
pub fn set_must_produce_diag(&self) {
2024-03-10 16:24:48 +00:00
assert!(
self.inner.borrow().must_produce_diag.is_none(),
"should only need to collect a backtrace once"
);
self.inner.borrow_mut().must_produce_diag = Some(Backtrace::capture());
}
}
// This `impl` block contains only the public diagnostic creation/emission API.
//
// Functions beginning with `struct_`/`create_` create a diagnostic. Other
// functions create and emit a diagnostic all in one go.
impl<'a> DiagCtxtHandle<'a> {
// No `#[rustc_lint_diagnostics]` and no `impl Into<DiagMessage>` because bug messages aren't
// user-facing.
#[track_caller]
pub fn struct_bug(self, msg: impl Into<Cow<'static, str>>) -> Diag<'a, BugAbort> {
Diag::new(self, Bug, msg.into())
}
// No `#[rustc_lint_diagnostics]` and no `impl Into<DiagMessage>` because bug messages aren't
// user-facing.
#[track_caller]
pub fn bug(self, msg: impl Into<Cow<'static, str>>) -> ! {
self.struct_bug(msg).emit()
}
// No `#[rustc_lint_diagnostics]` and no `impl Into<DiagMessage>` because bug messages aren't
// user-facing.
#[track_caller]
pub fn struct_span_bug(
self,
span: impl Into<MultiSpan>,
msg: impl Into<Cow<'static, str>>,
) -> Diag<'a, BugAbort> {
self.struct_bug(msg).with_span(span)
}
// No `#[rustc_lint_diagnostics]` and no `impl Into<DiagMessage>` because bug messages aren't
// user-facing.
#[track_caller]
pub fn span_bug(self, span: impl Into<MultiSpan>, msg: impl Into<Cow<'static, str>>) -> ! {
self.struct_span_bug(span, msg.into()).emit()
}
#[track_caller]
pub fn create_bug(self, bug: impl Diagnostic<'a, BugAbort>) -> Diag<'a, BugAbort> {
bug.into_diag(self, Bug)
}
#[track_caller]
pub fn emit_bug(self, bug: impl Diagnostic<'a, BugAbort>) -> ! {
self.create_bug(bug).emit()
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn struct_fatal(self, msg: impl Into<DiagMessage>) -> Diag<'a, FatalAbort> {
Diag::new(self, Fatal, msg)
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn fatal(self, msg: impl Into<DiagMessage>) -> ! {
self.struct_fatal(msg).emit()
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn struct_span_fatal(
self,
span: impl Into<MultiSpan>,
msg: impl Into<DiagMessage>,
) -> Diag<'a, FatalAbort> {
self.struct_fatal(msg).with_span(span)
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn span_fatal(self, span: impl Into<MultiSpan>, msg: impl Into<DiagMessage>) -> ! {
self.struct_span_fatal(span, msg).emit()
}
#[track_caller]
pub fn create_fatal(self, fatal: impl Diagnostic<'a, FatalAbort>) -> Diag<'a, FatalAbort> {
fatal.into_diag(self, Fatal)
}
#[track_caller]
pub fn emit_fatal(self, fatal: impl Diagnostic<'a, FatalAbort>) -> ! {
self.create_fatal(fatal).emit()
}
#[track_caller]
pub fn create_almost_fatal(
self,
fatal: impl Diagnostic<'a, FatalError>,
) -> Diag<'a, FatalError> {
fatal.into_diag(self, Fatal)
}
#[track_caller]
pub fn emit_almost_fatal(self, fatal: impl Diagnostic<'a, FatalError>) -> FatalError {
self.create_almost_fatal(fatal).emit()
}
// FIXME: This method should be removed (every error should have an associated error code).
#[rustc_lint_diagnostics]
#[track_caller]
pub fn struct_err(self, msg: impl Into<DiagMessage>) -> Diag<'a> {
Diag::new(self, Error, msg)
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn err(self, msg: impl Into<DiagMessage>) -> ErrorGuaranteed {
self.struct_err(msg).emit()
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn struct_span_err(
self,
span: impl Into<MultiSpan>,
msg: impl Into<DiagMessage>,
) -> Diag<'a> {
self.struct_err(msg).with_span(span)
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn span_err(
self,
span: impl Into<MultiSpan>,
msg: impl Into<DiagMessage>,
) -> ErrorGuaranteed {
self.struct_span_err(span, msg).emit()
}
#[track_caller]
pub fn create_err(self, err: impl Diagnostic<'a>) -> Diag<'a> {
err.into_diag(self, Error)
}
#[track_caller]
pub fn emit_err(self, err: impl Diagnostic<'a>) -> ErrorGuaranteed {
self.create_err(err).emit()
}
/// Ensures that an error is printed. See `Level::DelayedBug`.
//
// No `#[rustc_lint_diagnostics]` and no `impl Into<DiagMessage>` because bug messages aren't
// user-facing.
#[track_caller]
pub fn delayed_bug(self, msg: impl Into<Cow<'static, str>>) -> ErrorGuaranteed {
Diag::<ErrorGuaranteed>::new(self, DelayedBug, msg.into()).emit()
}
2024-12-10 23:19:45 +01:00
/// Ensures that an error is printed. See [`Level::DelayedBug`].
///
/// Note: this function used to be called `delay_span_bug`. It was renamed
/// to match similar functions like `span_err`, `span_warn`, etc.
//
// No `#[rustc_lint_diagnostics]` and no `impl Into<DiagMessage>` because bug messages aren't
// user-facing.
#[track_caller]
pub fn span_delayed_bug(
self,
sp: impl Into<MultiSpan>,
msg: impl Into<Cow<'static, str>>,
) -> ErrorGuaranteed {
Diag::<ErrorGuaranteed>::new(self, DelayedBug, msg.into()).with_span(sp).emit()
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn struct_warn(self, msg: impl Into<DiagMessage>) -> Diag<'a, ()> {
Diag::new(self, Warning, msg)
2020-06-30 18:58:15 +02:00
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn warn(self, msg: impl Into<DiagMessage>) {
self.struct_warn(msg).emit()
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn struct_span_warn(
self,
span: impl Into<MultiSpan>,
msg: impl Into<DiagMessage>,
) -> Diag<'a, ()> {
self.struct_warn(msg).with_span(span)
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn span_warn(self, span: impl Into<MultiSpan>, msg: impl Into<DiagMessage>) {
self.struct_span_warn(span, msg).emit()
}
#[track_caller]
pub fn create_warn(self, warning: impl Diagnostic<'a, ()>) -> Diag<'a, ()> {
warning.into_diag(self, Warning)
}
#[track_caller]
pub fn emit_warn(self, warning: impl Diagnostic<'a, ()>) {
self.create_warn(warning).emit()
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn struct_note(self, msg: impl Into<DiagMessage>) -> Diag<'a, ()> {
Diag::new(self, Note, msg)
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn note(&self, msg: impl Into<DiagMessage>) {
self.struct_note(msg).emit()
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn struct_span_note(
self,
span: impl Into<MultiSpan>,
msg: impl Into<DiagMessage>,
) -> Diag<'a, ()> {
self.struct_note(msg).with_span(span)
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn span_note(self, span: impl Into<MultiSpan>, msg: impl Into<DiagMessage>) {
self.struct_span_note(span, msg).emit()
}
#[track_caller]
pub fn create_note(self, note: impl Diagnostic<'a, ()>) -> Diag<'a, ()> {
note.into_diag(self, Note)
}
#[track_caller]
pub fn emit_note(self, note: impl Diagnostic<'a, ()>) {
self.create_note(note).emit()
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn struct_help(self, msg: impl Into<DiagMessage>) -> Diag<'a, ()> {
Diag::new(self, Help, msg)
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn struct_failure_note(self, msg: impl Into<DiagMessage>) -> Diag<'a, ()> {
Diag::new(self, FailureNote, msg)
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn struct_allow(self, msg: impl Into<DiagMessage>) -> Diag<'a, ()> {
Diag::new(self, Allow, msg)
}
#[rustc_lint_diagnostics]
#[track_caller]
pub fn struct_expect(self, msg: impl Into<DiagMessage>, id: LintExpectationId) -> Diag<'a, ()> {
2025-03-18 17:18:54 +00:00
Diag::new(self, Expect, msg).with_lint_id(id)
}
}
2023-12-17 21:48:57 +11:00
// Note: we prefer implementing operations on `DiagCtxt`, rather than
// `DiagCtxtInner`, whenever possible. This minimizes functions where
2023-12-17 21:48:57 +11:00
// `DiagCtxt::foo()` just borrows `inner` and forwards a call to
2024-01-05 17:19:37 +11:00
// `DiagCtxtInner::foo`.
impl DiagCtxtInner {
fn new(emitter: Box<DynEmitter>) -> Self {
Self {
flags: DiagCtxtFlags { can_emit_warnings: true, ..Default::default() },
registry: Registry::new(&[]),
err_guars: Vec::new(),
lint_err_guars: Vec::new(),
delayed_bugs: Vec::new(),
deduplicated_err_count: 0,
deduplicated_warn_count: 0,
emitter,
2024-03-10 16:24:48 +00:00
must_produce_diag: None,
has_printed: false,
suppressed_expected_diag: false,
taught_diagnostics: Default::default(),
emitted_diagnostic_codes: Default::default(),
emitted_diagnostics: Default::default(),
stashed_diagnostics: Default::default(),
future_breakage_diagnostics: Vec::new(),
fulfilled_expectations: Default::default(),
ice_file: None,
}
}
2019-09-23 04:45:21 +02:00
/// Emit all stashed diagnostics.
Overhaul the handling of errors at the top-level. Currently `emit_stashed_diagnostic` is called from four(!) different places: `print_error_count`, `DiagCtxtInner::drop`, `abort_if_errors`, and `compile_status`. And `flush_delayed` is called from two different places: `DiagCtxtInner::drop` and `Queries`. This is pretty gross! Each one should really be called from a single place, but there's a bunch of entanglements. This commit cleans up this mess. Specifically, it: - Removes all the existing calls to `emit_stashed_diagnostic`, and adds a single new call in `finish_diagnostics`. - Removes the early `flush_delayed` call in `codegen_and_build_linker`, replacing it with a simple early return if delayed bugs are present. - Changes `DiagCtxtInner::drop` and `DiagCtxtInner::flush_delayed` so they both assert that the stashed diagnostics are empty (i.e. processed beforehand). - Changes `interface::run_compiler` so that any errors emitted during `finish_diagnostics` (i.e. late-emitted stashed diagnostics) are counted and cannot be overlooked. This requires adding `ErrorGuaranteed` return values to several functions. - Removes the `stashed_err_count` call in `analysis`. This is possible now that we don't have to worry about calling `flush_delayed` early from `codegen_and_build_linker` when stashed diagnostics are pending. - Changes the `span_bug` case in `handle_tuple_field_pattern_match` to a `delayed_span_bug`, because it now can be reached due to the removal of the `stashed_err_count` call in `analysis`. - Slightly changes the expected output of three tests. If no errors are emitted but there are delayed bugs, the error count is no longer printed. This is because delayed bugs are now always printed after the error count is printed (or not printed, if the error count is zero). There is a lot going on in this commit. It's hard to break into smaller pieces because the existing code is very tangled. It took me a long time and a lot of effort to understand how the different pieces interact, and I think the new code is a lot simpler and easier to understand.
2024-02-19 10:00:19 +11:00
fn emit_stashed_diagnostics(&mut self) -> Option<ErrorGuaranteed> {
let mut guar = None;
let has_errors = !self.err_guars.is_empty();
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
for (_, (diag, _guar)) in std::mem::take(&mut self.stashed_diagnostics).into_iter() {
if !diag.is_error() {
// Unless they're forced, don't flush stashed warnings when
// there are errors, to avoid causing warning overload. The
// stash would've been stolen already if it were important.
if !diag.is_force_warn() && has_errors {
continue;
2022-08-17 06:07:33 -05:00
}
}
guar = guar.or(self.emit_diagnostic(diag, None));
}
Overhaul the handling of errors at the top-level. Currently `emit_stashed_diagnostic` is called from four(!) different places: `print_error_count`, `DiagCtxtInner::drop`, `abort_if_errors`, and `compile_status`. And `flush_delayed` is called from two different places: `DiagCtxtInner::drop` and `Queries`. This is pretty gross! Each one should really be called from a single place, but there's a bunch of entanglements. This commit cleans up this mess. Specifically, it: - Removes all the existing calls to `emit_stashed_diagnostic`, and adds a single new call in `finish_diagnostics`. - Removes the early `flush_delayed` call in `codegen_and_build_linker`, replacing it with a simple early return if delayed bugs are present. - Changes `DiagCtxtInner::drop` and `DiagCtxtInner::flush_delayed` so they both assert that the stashed diagnostics are empty (i.e. processed beforehand). - Changes `interface::run_compiler` so that any errors emitted during `finish_diagnostics` (i.e. late-emitted stashed diagnostics) are counted and cannot be overlooked. This requires adding `ErrorGuaranteed` return values to several functions. - Removes the `stashed_err_count` call in `analysis`. This is possible now that we don't have to worry about calling `flush_delayed` early from `codegen_and_build_linker` when stashed diagnostics are pending. - Changes the `span_bug` case in `handle_tuple_field_pattern_match` to a `delayed_span_bug`, because it now can be reached due to the removal of the `stashed_err_count` call in `analysis`. - Slightly changes the expected output of three tests. If no errors are emitted but there are delayed bugs, the error count is no longer printed. This is because delayed bugs are now always printed after the error count is printed (or not printed, if the error count is zero). There is a lot going on in this commit. It's hard to break into smaller pieces because the existing code is very tangled. It took me a long time and a lot of effort to understand how the different pieces interact, and I think the new code is a lot simpler and easier to understand.
2024-02-19 10:00:19 +11:00
guar
2019-09-23 04:45:21 +02:00
}
// Return value is only `Some` if the level is `Error` or `DelayedBug`.
fn emit_diagnostic(
&mut self,
mut diagnostic: DiagInner,
taint: Option<&Cell<Option<ErrorGuaranteed>>>,
) -> Option<ErrorGuaranteed> {
if diagnostic.has_future_breakage() {
// Future breakages aren't emitted if they're `Level::Allow` or
// `Level::Expect`, but they still need to be constructed and
// stashed below, so they'll trigger the must_produce_diag check.
2025-03-18 17:18:54 +00:00
assert_matches!(diagnostic.level, Error | Warning | Allow | Expect);
self.future_breakage_diagnostics.push(diagnostic.clone());
}
// We call TRACK_DIAGNOSTIC with an empty closure for the cases that
// return early *and* have some kind of side-effect, except where
// noted.
match diagnostic.level {
Bug => {}
Fatal | Error => {
if self.treat_next_err_as_bug() {
// `Fatal` and `Error` can be promoted to `Bug`.
diagnostic.level = Bug;
}
}
DelayedBug => {
// Note that because we check these conditions first,
// `-Zeagerly-emit-delayed-bugs` and `-Ztreat-err-as-bug`
// continue to work even after we've issued an error and
// stopped recording new delayed bugs.
if self.flags.eagerly_emit_delayed_bugs {
// `DelayedBug` can be promoted to `Error` or `Bug`.
if self.treat_next_err_as_bug() {
diagnostic.level = Bug;
} else {
diagnostic.level = Error;
}
} else {
// If we have already emitted at least one error, we don't need
// to record the delayed bug, because it'll never be used.
return if let Some(guar) = self.has_errors() {
Some(guar)
} else {
// No `TRACK_DIAGNOSTIC` call is needed, because the
// incremental session is deleted if there is a delayed
// bug. This also saves us from cloning the diagnostic.
let backtrace = std::backtrace::Backtrace::capture();
// This `unchecked_error_guaranteed` is valid. It is where the
// `ErrorGuaranteed` for delayed bugs originates. See
// `DiagCtxtInner::drop`.
#[allow(deprecated)]
let guar = ErrorGuaranteed::unchecked_error_guaranteed();
self.delayed_bugs
.push((DelayedDiagInner::with_backtrace(diagnostic, backtrace), guar));
Some(guar)
};
}
}
2025-03-18 17:18:54 +00:00
ForceWarning if diagnostic.lint_id.is_none() => {} // `ForceWarning(Some(...))` is below, with `Expect`
Warning => {
if !self.flags.can_emit_warnings {
// We are not emitting warnings.
if diagnostic.has_future_breakage() {
// The side-effect is at the top of this method.
TRACK_DIAGNOSTIC(diagnostic, &mut |_| None);
}
return None;
}
}
Note | Help | FailureNote => {}
OnceNote | OnceHelp => panic!("bad level: {:?}", diagnostic.level),
Allow => {
// Nothing emitted for allowed lints.
if diagnostic.has_future_breakage() {
// The side-effect is at the top of this method.
TRACK_DIAGNOSTIC(diagnostic, &mut |_| None);
self.suppressed_expected_diag = true;
}
return None;
}
2025-03-18 17:18:54 +00:00
Expect | ForceWarning => {
self.fulfilled_expectations.insert(diagnostic.lint_id.unwrap());
if let Expect = diagnostic.level {
// Nothing emitted here for expected lints.
TRACK_DIAGNOSTIC(diagnostic, &mut |_| None);
self.suppressed_expected_diag = true;
return None;
}
}
}
TRACK_DIAGNOSTIC(diagnostic, &mut |mut diagnostic| {
if let Some(code) = diagnostic.code {
self.emitted_diagnostic_codes.insert(code);
}
2023-10-04 18:34:50 +00:00
let already_emitted = {
2022-03-20 20:02:18 +01:00
let mut hasher = StableHasher::new();
diagnostic.hash(&mut hasher);
2022-03-20 20:02:18 +01:00
let diagnostic_hash = hasher.finish();
2023-10-04 18:34:50 +00:00
!self.emitted_diagnostics.insert(diagnostic_hash)
2022-03-20 20:02:18 +01:00
};
let is_error = diagnostic.is_error();
let is_lint = diagnostic.is_lint.is_some();
// Only emit the diagnostic if we've been asked to deduplicate or
// haven't already emitted an equivalent diagnostic.
2023-10-04 18:34:50 +00:00
if !(self.flags.deduplicate_diagnostics && already_emitted) {
debug!(?diagnostic);
debug!(?self.emitted_diagnostics);
let not_yet_emitted = |sub: &mut Subdiag| {
debug!(?sub);
if sub.level != OnceNote && sub.level != OnceHelp {
return true;
}
let mut hasher = StableHasher::new();
sub.hash(&mut hasher);
let diagnostic_hash = hasher.finish();
debug!(?diagnostic_hash);
self.emitted_diagnostics.insert(diagnostic_hash)
};
diagnostic.children.retain_mut(not_yet_emitted);
2023-10-04 18:34:50 +00:00
if already_emitted {
Reduce capabilities of `Diagnostic`. Currently many diagnostic modifier methods are available on both `Diagnostic` and `DiagnosticBuilder`. This commit removes most of them from `Diagnostic`. To minimize the diff size, it keeps them within `diagnostic.rs` but changes the surrounding `impl Diagnostic` block to `impl DiagnosticBuilder`. (I intend to move things around later, to give a more sensible code layout.) `Diagnostic` keeps a few methods that it still needs, like `sub`, `arg`, and `replace_args`. The `forward!` macro, which defined two additional methods per call (e.g. `note` and `with_note`), is replaced by the `with_fn!` macro, which defines one additional method per call (e.g. `with_note`). It's now also only used when necessary -- not all modifier methods currently need a `with_*` form. (New ones can be easily added as necessary.) All this also requires changing `trait AddToDiagnostic` so its methods take `DiagnosticBuilder` instead of `Diagnostic`, which leads to many mechanical changes. `SubdiagnosticMessageOp` gains a type parameter `G`. There are three subdiagnostics -- `DelayedAtWithoutNewline`, `DelayedAtWithNewline`, and `InvalidFlushedDelayedDiagnosticLevel` -- that are created within the diagnostics machinery and appended to external diagnostics. These are handled at the `Diagnostic` level, which means it's now hard to construct them via `derive(Diagnostic)`, so instead we construct them by hand. This has no effect on what they look like when printed. There are lots of new `allow` markers for `untranslatable_diagnostics` and `diagnostics_outside_of_impl`. This is because `#[rustc_lint_diagnostics]` annotations were present on the `Diagnostic` modifier methods, but missing from the `DiagnosticBuilder` modifier methods. They're now present.
2024-02-06 16:44:30 +11:00
let msg = "duplicate diagnostic emitted due to `-Z deduplicate-diagnostics=no`";
diagnostic.sub(Note, msg, MultiSpan::new());
2023-10-04 18:34:50 +00:00
}
if is_error {
self.deduplicated_err_count += 1;
2025-03-18 17:18:54 +00:00
} else if matches!(diagnostic.level, ForceWarning | Warning) {
self.deduplicated_warn_count += 1;
}
self.has_printed = true;
self.emitter.emit_diagnostic(diagnostic, &self.registry);
}
if is_error {
// If we have any delayed bugs recorded, we can discard them
// because they won't be used. (This should only occur if there
// have been no errors previously emitted, because we don't add
// new delayed bugs once the first error is emitted.)
if !self.delayed_bugs.is_empty() {
assert_eq!(self.lint_err_guars.len() + self.err_guars.len(), 0);
self.delayed_bugs.clear();
self.delayed_bugs.shrink_to_fit();
}
// This `unchecked_error_guaranteed` is valid. It is where the
// `ErrorGuaranteed` for errors and lint errors originates.
#[allow(deprecated)]
let guar = ErrorGuaranteed::unchecked_error_guaranteed();
if is_lint {
self.lint_err_guars.push(guar);
} else {
if let Some(taint) = taint {
taint.set(Some(guar));
}
self.err_guars.push(guar);
}
self.panic_if_treat_err_as_bug();
Some(guar)
} else {
None
}
})
}
fn treat_err_as_bug(&self) -> bool {
self.flags
.treat_err_as_bug
.is_some_and(|c| self.err_guars.len() + self.lint_err_guars.len() >= c.get())
}
// Use this one before incrementing `err_count`.
fn treat_next_err_as_bug(&self) -> bool {
Don't consider delayed bugs for `-Ztreat-err-as-bug`. `-Ztreat-err-as-bug` treats normal errors and delayed bugs equally, which can lead to some really surprising results. This commit changes `-Ztreat-err-as-bug` so it ignores delayed bugs, unless they get promoted to proper bugs and are printed. This feels to me much simpler and more logical. And it simplifies the implementation: - The `-Ztreat-err-as-bug` check is removed from in `DiagCtxt::{delayed_bug,span_delayed_bug}`. - `treat_err_as_bug` doesn't need to count delayed bugs. - The `-Ztreat-err-as-bug` panic message is simpler, because it doesn't have to mention delayed bugs. Output of delayed bugs is now more consistent. They're always printed the same way. Previously when they triggered `-Ztreat-err-as-bug` they would be printed slightly differently, via `span_bug` in `span_delayed_bug` or `delayed_bug`. A minor behaviour change: the "no errors encountered even though `span_delayed_bug` issued" printed before delayed bugs is now a note rather than a bug. This is done so it doesn't get counted as an error that might trigger `-Ztreat-err-as-bug`, which would be silly. This means that if you use `-Ztreat-err-as-bug=1` and there are no normal errors but there are delayed bugs, the first delayed bug will be shown (and the panic will happen after it's printed). Also, I have added a second note saying "those delayed bugs will now be shown as internal compiler errors". I think this makes it clearer what is happening, because the whole concept of delayed bugs is non-obvious. There are some test changes. - equality-in-canonical-query.rs: Minor output changes, and the error count reduces by one because the "no errors encountered even though `span_delayed_bug` issued" message is no longer counted as an error. - rpit_tait_equality_in_canonical_query.rs: Ditto. - storage-live.rs: The query stack disappears because these delayed bugs are now printed at the end, rather than when they are created. - storage-return.rs, span_delayed_bug.rs: now need `-Zeagerly-emit-delayed-bugs` because they need the delayed bugs emitted immediately to preserve behaviour.
2024-01-12 07:38:42 +11:00
self.flags
.treat_err_as_bug
.is_some_and(|c| self.err_guars.len() + self.lint_err_guars.len() + 1 >= c.get())
}
fn has_errors_excluding_lint_errors(&self) -> Option<ErrorGuaranteed> {
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
self.err_guars.get(0).copied().or_else(|| {
if let Some((_diag, guar)) = self
.stashed_diagnostics
.values()
.find(|(diag, guar)| guar.is_some() && diag.is_lint.is_none())
{
*guar
} else {
None
}
})
}
fn has_errors(&self) -> Option<ErrorGuaranteed> {
Overhaul how stashed diagnostics work, again. Stashed errors used to be counted as errors, but could then be cancelled, leading to `ErrorGuaranteed` soundness holes. #120828 changed that, closing the soundness hole. But it introduced other difficulties because you sometimes have to account for pending stashed errors when making decisions about whether errors have occured/will occur and it's easy to overlook these. This commit aims for a middle ground. - Stashed errors (not warnings) are counted immediately as emitted errors, avoiding the possibility of forgetting to consider them. - The ability to cancel (or downgrade) stashed errors is eliminated, by disallowing the use of `steal_diagnostic` with errors, and introducing the more restrictive methods `try_steal_{modify,replace}_and_emit_err` that can be used instead. Other things: - `DiagnosticBuilder::stash` and `DiagCtxt::stash_diagnostic` now both return `Option<ErrorGuaranteed>`, which enables the removal of two `delayed_bug` calls and one `Ty::new_error_with_message` call. This is possible because we store error guarantees in `DiagCtxt::stashed_diagnostics`. - Storing the guarantees also saves us having to maintain a counter. - Calls to the `stashed_err_count` method are no longer necessary alongside calls to `has_errors`, which is a nice simplification, and eliminates two more `span_delayed_bug` calls and one FIXME comment. - Tests are added for three of the four fixed PRs mentioned below. - `issue-121108.rs`'s output improved slightly, omitting a non-useful error message. Fixes #121451. Fixes #121477. Fixes #121504. Fixes #121508.
2024-02-26 15:21:01 +11:00
self.err_guars.get(0).copied().or_else(|| self.lint_err_guars.get(0).copied()).or_else(
|| {
if let Some((_diag, guar)) =
self.stashed_diagnostics.values().find(|(_diag, guar)| guar.is_some())
{
*guar
} else {
None
}
},
)
}
fn has_errors_or_delayed_bugs(&self) -> Option<ErrorGuaranteed> {
self.has_errors().or_else(|| self.delayed_bugs.get(0).map(|(_, guar)| guar).copied())
}
/// Translate `message` eagerly with `args` to `SubdiagMessage::Eager`.
fn eagerly_translate<'a>(
&self,
message: DiagMessage,
args: impl Iterator<Item = DiagArg<'a>>,
) -> SubdiagMessage {
SubdiagMessage::Translated(Cow::from(self.eagerly_translate_to_string(message, args)))
}
/// Translate `message` eagerly with `args` to `String`.
fn eagerly_translate_to_string<'a>(
&self,
message: DiagMessage,
args: impl Iterator<Item = DiagArg<'a>>,
) -> String {
let args = crate::translation::to_fluent_args(args);
self.emitter.translate_message(&message, &args).map_err(Report::new).unwrap().to_string()
}
Reduce capabilities of `Diagnostic`. Currently many diagnostic modifier methods are available on both `Diagnostic` and `DiagnosticBuilder`. This commit removes most of them from `Diagnostic`. To minimize the diff size, it keeps them within `diagnostic.rs` but changes the surrounding `impl Diagnostic` block to `impl DiagnosticBuilder`. (I intend to move things around later, to give a more sensible code layout.) `Diagnostic` keeps a few methods that it still needs, like `sub`, `arg`, and `replace_args`. The `forward!` macro, which defined two additional methods per call (e.g. `note` and `with_note`), is replaced by the `with_fn!` macro, which defines one additional method per call (e.g. `with_note`). It's now also only used when necessary -- not all modifier methods currently need a `with_*` form. (New ones can be easily added as necessary.) All this also requires changing `trait AddToDiagnostic` so its methods take `DiagnosticBuilder` instead of `Diagnostic`, which leads to many mechanical changes. `SubdiagnosticMessageOp` gains a type parameter `G`. There are three subdiagnostics -- `DelayedAtWithoutNewline`, `DelayedAtWithNewline`, and `InvalidFlushedDelayedDiagnosticLevel` -- that are created within the diagnostics machinery and appended to external diagnostics. These are handled at the `Diagnostic` level, which means it's now hard to construct them via `derive(Diagnostic)`, so instead we construct them by hand. This has no effect on what they look like when printed. There are lots of new `allow` markers for `untranslatable_diagnostics` and `diagnostics_outside_of_impl`. This is because `#[rustc_lint_diagnostics]` annotations were present on the `Diagnostic` modifier methods, but missing from the `DiagnosticBuilder` modifier methods. They're now present.
2024-02-06 16:44:30 +11:00
fn eagerly_translate_for_subdiag(
&self,
diag: &DiagInner,
msg: impl Into<SubdiagMessage>,
) -> SubdiagMessage {
Reduce capabilities of `Diagnostic`. Currently many diagnostic modifier methods are available on both `Diagnostic` and `DiagnosticBuilder`. This commit removes most of them from `Diagnostic`. To minimize the diff size, it keeps them within `diagnostic.rs` but changes the surrounding `impl Diagnostic` block to `impl DiagnosticBuilder`. (I intend to move things around later, to give a more sensible code layout.) `Diagnostic` keeps a few methods that it still needs, like `sub`, `arg`, and `replace_args`. The `forward!` macro, which defined two additional methods per call (e.g. `note` and `with_note`), is replaced by the `with_fn!` macro, which defines one additional method per call (e.g. `with_note`). It's now also only used when necessary -- not all modifier methods currently need a `with_*` form. (New ones can be easily added as necessary.) All this also requires changing `trait AddToDiagnostic` so its methods take `DiagnosticBuilder` instead of `Diagnostic`, which leads to many mechanical changes. `SubdiagnosticMessageOp` gains a type parameter `G`. There are three subdiagnostics -- `DelayedAtWithoutNewline`, `DelayedAtWithNewline`, and `InvalidFlushedDelayedDiagnosticLevel` -- that are created within the diagnostics machinery and appended to external diagnostics. These are handled at the `Diagnostic` level, which means it's now hard to construct them via `derive(Diagnostic)`, so instead we construct them by hand. This has no effect on what they look like when printed. There are lots of new `allow` markers for `untranslatable_diagnostics` and `diagnostics_outside_of_impl`. This is because `#[rustc_lint_diagnostics]` annotations were present on the `Diagnostic` modifier methods, but missing from the `DiagnosticBuilder` modifier methods. They're now present.
2024-02-06 16:44:30 +11:00
let msg = diag.subdiagnostic_message_to_diagnostic_message(msg);
self.eagerly_translate(msg, diag.args.iter())
Reduce capabilities of `Diagnostic`. Currently many diagnostic modifier methods are available on both `Diagnostic` and `DiagnosticBuilder`. This commit removes most of them from `Diagnostic`. To minimize the diff size, it keeps them within `diagnostic.rs` but changes the surrounding `impl Diagnostic` block to `impl DiagnosticBuilder`. (I intend to move things around later, to give a more sensible code layout.) `Diagnostic` keeps a few methods that it still needs, like `sub`, `arg`, and `replace_args`. The `forward!` macro, which defined two additional methods per call (e.g. `note` and `with_note`), is replaced by the `with_fn!` macro, which defines one additional method per call (e.g. `with_note`). It's now also only used when necessary -- not all modifier methods currently need a `with_*` form. (New ones can be easily added as necessary.) All this also requires changing `trait AddToDiagnostic` so its methods take `DiagnosticBuilder` instead of `Diagnostic`, which leads to many mechanical changes. `SubdiagnosticMessageOp` gains a type parameter `G`. There are three subdiagnostics -- `DelayedAtWithoutNewline`, `DelayedAtWithNewline`, and `InvalidFlushedDelayedDiagnosticLevel` -- that are created within the diagnostics machinery and appended to external diagnostics. These are handled at the `Diagnostic` level, which means it's now hard to construct them via `derive(Diagnostic)`, so instead we construct them by hand. This has no effect on what they look like when printed. There are lots of new `allow` markers for `untranslatable_diagnostics` and `diagnostics_outside_of_impl`. This is because `#[rustc_lint_diagnostics]` annotations were present on the `Diagnostic` modifier methods, but missing from the `DiagnosticBuilder` modifier methods. They're now present.
2024-02-06 16:44:30 +11:00
}
fn flush_delayed(&mut self) {
Overhaul the handling of errors at the top-level. Currently `emit_stashed_diagnostic` is called from four(!) different places: `print_error_count`, `DiagCtxtInner::drop`, `abort_if_errors`, and `compile_status`. And `flush_delayed` is called from two different places: `DiagCtxtInner::drop` and `Queries`. This is pretty gross! Each one should really be called from a single place, but there's a bunch of entanglements. This commit cleans up this mess. Specifically, it: - Removes all the existing calls to `emit_stashed_diagnostic`, and adds a single new call in `finish_diagnostics`. - Removes the early `flush_delayed` call in `codegen_and_build_linker`, replacing it with a simple early return if delayed bugs are present. - Changes `DiagCtxtInner::drop` and `DiagCtxtInner::flush_delayed` so they both assert that the stashed diagnostics are empty (i.e. processed beforehand). - Changes `interface::run_compiler` so that any errors emitted during `finish_diagnostics` (i.e. late-emitted stashed diagnostics) are counted and cannot be overlooked. This requires adding `ErrorGuaranteed` return values to several functions. - Removes the `stashed_err_count` call in `analysis`. This is possible now that we don't have to worry about calling `flush_delayed` early from `codegen_and_build_linker` when stashed diagnostics are pending. - Changes the `span_bug` case in `handle_tuple_field_pattern_match` to a `delayed_span_bug`, because it now can be reached due to the removal of the `stashed_err_count` call in `analysis`. - Slightly changes the expected output of three tests. If no errors are emitted but there are delayed bugs, the error count is no longer printed. This is because delayed bugs are now always printed after the error count is printed (or not printed, if the error count is zero). There is a lot going on in this commit. It's hard to break into smaller pieces because the existing code is very tangled. It took me a long time and a lot of effort to understand how the different pieces interact, and I think the new code is a lot simpler and easier to understand.
2024-02-19 10:00:19 +11:00
// Stashed diagnostics must be emitted before delayed bugs are flushed.
// Otherwise, we might ICE prematurely when errors would have
// eventually happened.
assert!(self.stashed_diagnostics.is_empty());
if !self.err_guars.is_empty() {
// If an error happened already. We shouldn't expose delayed bugs.
return;
}
if self.delayed_bugs.is_empty() {
// Nothing to do.
return;
}
let bugs: Vec<_> =
std::mem::take(&mut self.delayed_bugs).into_iter().map(|(b, _)| b).collect();
let backtrace = std::env::var_os("RUST_BACKTRACE").as_deref() != Some(OsStr::new("0"));
let decorate = backtrace || self.ice_file.is_none();
let mut out = self
.ice_file
.as_ref()
.and_then(|file| std::fs::File::options().create(true).append(true).open(file).ok());
// Put the overall explanation before the `DelayedBug`s, to frame them
// better (e.g. separate warnings from them). Also, use notes, which
// don't count as errors, to avoid possibly triggering
// `-Ztreat-err-as-bug`, which we don't want.
let note1 = "no errors encountered even though delayed bugs were created";
let note2 = "those delayed bugs will now be shown as internal compiler errors";
self.emit_diagnostic(DiagInner::new(Note, note1), None);
self.emit_diagnostic(DiagInner::new(Note, note2), None);
for bug in bugs {
if let Some(out) = &mut out {
_ = write!(
out,
"delayed bug: {}\n{}\n",
bug.inner
.messages
.iter()
.filter_map(|(msg, _)| msg.as_str())
.collect::<String>(),
&bug.note
);
}
let mut bug = if decorate { bug.decorate(self) } else { bug.inner };
// "Undelay" the delayed bugs into plain bugs.
if bug.level != DelayedBug {
// NOTE(eddyb) not panicking here because we're already producing
// an ICE, and the more information the merrier.
Reduce capabilities of `Diagnostic`. Currently many diagnostic modifier methods are available on both `Diagnostic` and `DiagnosticBuilder`. This commit removes most of them from `Diagnostic`. To minimize the diff size, it keeps them within `diagnostic.rs` but changes the surrounding `impl Diagnostic` block to `impl DiagnosticBuilder`. (I intend to move things around later, to give a more sensible code layout.) `Diagnostic` keeps a few methods that it still needs, like `sub`, `arg`, and `replace_args`. The `forward!` macro, which defined two additional methods per call (e.g. `note` and `with_note`), is replaced by the `with_fn!` macro, which defines one additional method per call (e.g. `with_note`). It's now also only used when necessary -- not all modifier methods currently need a `with_*` form. (New ones can be easily added as necessary.) All this also requires changing `trait AddToDiagnostic` so its methods take `DiagnosticBuilder` instead of `Diagnostic`, which leads to many mechanical changes. `SubdiagnosticMessageOp` gains a type parameter `G`. There are three subdiagnostics -- `DelayedAtWithoutNewline`, `DelayedAtWithNewline`, and `InvalidFlushedDelayedDiagnosticLevel` -- that are created within the diagnostics machinery and appended to external diagnostics. These are handled at the `Diagnostic` level, which means it's now hard to construct them via `derive(Diagnostic)`, so instead we construct them by hand. This has no effect on what they look like when printed. There are lots of new `allow` markers for `untranslatable_diagnostics` and `diagnostics_outside_of_impl`. This is because `#[rustc_lint_diagnostics]` annotations were present on the `Diagnostic` modifier methods, but missing from the `DiagnosticBuilder` modifier methods. They're now present.
2024-02-06 16:44:30 +11:00
//
// We are at the `DiagInner`/`DiagCtxtInner` level rather than
// the usual `Diag`/`DiagCtxt` level, so we must augment `bug`
// in a lower-level fashion.
Reduce capabilities of `Diagnostic`. Currently many diagnostic modifier methods are available on both `Diagnostic` and `DiagnosticBuilder`. This commit removes most of them from `Diagnostic`. To minimize the diff size, it keeps them within `diagnostic.rs` but changes the surrounding `impl Diagnostic` block to `impl DiagnosticBuilder`. (I intend to move things around later, to give a more sensible code layout.) `Diagnostic` keeps a few methods that it still needs, like `sub`, `arg`, and `replace_args`. The `forward!` macro, which defined two additional methods per call (e.g. `note` and `with_note`), is replaced by the `with_fn!` macro, which defines one additional method per call (e.g. `with_note`). It's now also only used when necessary -- not all modifier methods currently need a `with_*` form. (New ones can be easily added as necessary.) All this also requires changing `trait AddToDiagnostic` so its methods take `DiagnosticBuilder` instead of `Diagnostic`, which leads to many mechanical changes. `SubdiagnosticMessageOp` gains a type parameter `G`. There are three subdiagnostics -- `DelayedAtWithoutNewline`, `DelayedAtWithNewline`, and `InvalidFlushedDelayedDiagnosticLevel` -- that are created within the diagnostics machinery and appended to external diagnostics. These are handled at the `Diagnostic` level, which means it's now hard to construct them via `derive(Diagnostic)`, so instead we construct them by hand. This has no effect on what they look like when printed. There are lots of new `allow` markers for `untranslatable_diagnostics` and `diagnostics_outside_of_impl`. This is because `#[rustc_lint_diagnostics]` annotations were present on the `Diagnostic` modifier methods, but missing from the `DiagnosticBuilder` modifier methods. They're now present.
2024-02-06 16:44:30 +11:00
bug.arg("level", bug.level);
let msg = crate::fluent_generated::errors_invalid_flushed_delayed_diagnostic_level;
let msg = self.eagerly_translate_for_subdiag(&bug, msg); // after the `arg` call
bug.sub(Note, msg, bug.span.primary_span().unwrap().into());
}
bug.level = Bug;
self.emit_diagnostic(bug, None);
}
// Panic with `DelayedBugPanic` to avoid "unexpected panic" messages.
panic::panic_any(DelayedBugPanic);
}
fn panic_if_treat_err_as_bug(&self) {
if self.treat_err_as_bug() {
Don't consider delayed bugs for `-Ztreat-err-as-bug`. `-Ztreat-err-as-bug` treats normal errors and delayed bugs equally, which can lead to some really surprising results. This commit changes `-Ztreat-err-as-bug` so it ignores delayed bugs, unless they get promoted to proper bugs and are printed. This feels to me much simpler and more logical. And it simplifies the implementation: - The `-Ztreat-err-as-bug` check is removed from in `DiagCtxt::{delayed_bug,span_delayed_bug}`. - `treat_err_as_bug` doesn't need to count delayed bugs. - The `-Ztreat-err-as-bug` panic message is simpler, because it doesn't have to mention delayed bugs. Output of delayed bugs is now more consistent. They're always printed the same way. Previously when they triggered `-Ztreat-err-as-bug` they would be printed slightly differently, via `span_bug` in `span_delayed_bug` or `delayed_bug`. A minor behaviour change: the "no errors encountered even though `span_delayed_bug` issued" printed before delayed bugs is now a note rather than a bug. This is done so it doesn't get counted as an error that might trigger `-Ztreat-err-as-bug`, which would be silly. This means that if you use `-Ztreat-err-as-bug=1` and there are no normal errors but there are delayed bugs, the first delayed bug will be shown (and the panic will happen after it's printed). Also, I have added a second note saying "those delayed bugs will now be shown as internal compiler errors". I think this makes it clearer what is happening, because the whole concept of delayed bugs is non-obvious. There are some test changes. - equality-in-canonical-query.rs: Minor output changes, and the error count reduces by one because the "no errors encountered even though `span_delayed_bug` issued" message is no longer counted as an error. - rpit_tait_equality_in_canonical_query.rs: Ditto. - storage-live.rs: The query stack disappears because these delayed bugs are now printed at the end, rather than when they are created. - storage-return.rs, span_delayed_bug.rs: now need `-Zeagerly-emit-delayed-bugs` because they need the delayed bugs emitted immediately to preserve behaviour.
2024-01-12 07:38:42 +11:00
let n = self.flags.treat_err_as_bug.map(|c| c.get()).unwrap();
assert_eq!(n, self.err_guars.len() + self.lint_err_guars.len());
Don't consider delayed bugs for `-Ztreat-err-as-bug`. `-Ztreat-err-as-bug` treats normal errors and delayed bugs equally, which can lead to some really surprising results. This commit changes `-Ztreat-err-as-bug` so it ignores delayed bugs, unless they get promoted to proper bugs and are printed. This feels to me much simpler and more logical. And it simplifies the implementation: - The `-Ztreat-err-as-bug` check is removed from in `DiagCtxt::{delayed_bug,span_delayed_bug}`. - `treat_err_as_bug` doesn't need to count delayed bugs. - The `-Ztreat-err-as-bug` panic message is simpler, because it doesn't have to mention delayed bugs. Output of delayed bugs is now more consistent. They're always printed the same way. Previously when they triggered `-Ztreat-err-as-bug` they would be printed slightly differently, via `span_bug` in `span_delayed_bug` or `delayed_bug`. A minor behaviour change: the "no errors encountered even though `span_delayed_bug` issued" printed before delayed bugs is now a note rather than a bug. This is done so it doesn't get counted as an error that might trigger `-Ztreat-err-as-bug`, which would be silly. This means that if you use `-Ztreat-err-as-bug=1` and there are no normal errors but there are delayed bugs, the first delayed bug will be shown (and the panic will happen after it's printed). Also, I have added a second note saying "those delayed bugs will now be shown as internal compiler errors". I think this makes it clearer what is happening, because the whole concept of delayed bugs is non-obvious. There are some test changes. - equality-in-canonical-query.rs: Minor output changes, and the error count reduces by one because the "no errors encountered even though `span_delayed_bug` issued" message is no longer counted as an error. - rpit_tait_equality_in_canonical_query.rs: Ditto. - storage-live.rs: The query stack disappears because these delayed bugs are now printed at the end, rather than when they are created. - storage-return.rs, span_delayed_bug.rs: now need `-Zeagerly-emit-delayed-bugs` because they need the delayed bugs emitted immediately to preserve behaviour.
2024-01-12 07:38:42 +11:00
if n == 1 {
panic!("aborting due to `-Z treat-err-as-bug=1`");
} else {
panic!("aborting after {n} errors due to `-Z treat-err-as-bug={n}`");
}
}
}
}
struct DelayedDiagInner {
inner: DiagInner,
note: Backtrace,
}
impl DelayedDiagInner {
fn with_backtrace(diagnostic: DiagInner, backtrace: Backtrace) -> Self {
DelayedDiagInner { inner: diagnostic, note: backtrace }
}
fn decorate(self, dcx: &DiagCtxtInner) -> DiagInner {
// We are at the `DiagInner`/`DiagCtxtInner` level rather than the
// usual `Diag`/`DiagCtxt` level, so we must construct `diag` in a
// lower-level fashion.
Reduce capabilities of `Diagnostic`. Currently many diagnostic modifier methods are available on both `Diagnostic` and `DiagnosticBuilder`. This commit removes most of them from `Diagnostic`. To minimize the diff size, it keeps them within `diagnostic.rs` but changes the surrounding `impl Diagnostic` block to `impl DiagnosticBuilder`. (I intend to move things around later, to give a more sensible code layout.) `Diagnostic` keeps a few methods that it still needs, like `sub`, `arg`, and `replace_args`. The `forward!` macro, which defined two additional methods per call (e.g. `note` and `with_note`), is replaced by the `with_fn!` macro, which defines one additional method per call (e.g. `with_note`). It's now also only used when necessary -- not all modifier methods currently need a `with_*` form. (New ones can be easily added as necessary.) All this also requires changing `trait AddToDiagnostic` so its methods take `DiagnosticBuilder` instead of `Diagnostic`, which leads to many mechanical changes. `SubdiagnosticMessageOp` gains a type parameter `G`. There are three subdiagnostics -- `DelayedAtWithoutNewline`, `DelayedAtWithNewline`, and `InvalidFlushedDelayedDiagnosticLevel` -- that are created within the diagnostics machinery and appended to external diagnostics. These are handled at the `Diagnostic` level, which means it's now hard to construct them via `derive(Diagnostic)`, so instead we construct them by hand. This has no effect on what they look like when printed. There are lots of new `allow` markers for `untranslatable_diagnostics` and `diagnostics_outside_of_impl`. This is because `#[rustc_lint_diagnostics]` annotations were present on the `Diagnostic` modifier methods, but missing from the `DiagnosticBuilder` modifier methods. They're now present.
2024-02-06 16:44:30 +11:00
let mut diag = self.inner;
let msg = match self.note.status() {
BacktraceStatus::Captured => crate::fluent_generated::errors_delayed_at_with_newline,
// Avoid the needless newline when no backtrace has been captured,
// the display impl should just be a single line.
Reduce capabilities of `Diagnostic`. Currently many diagnostic modifier methods are available on both `Diagnostic` and `DiagnosticBuilder`. This commit removes most of them from `Diagnostic`. To minimize the diff size, it keeps them within `diagnostic.rs` but changes the surrounding `impl Diagnostic` block to `impl DiagnosticBuilder`. (I intend to move things around later, to give a more sensible code layout.) `Diagnostic` keeps a few methods that it still needs, like `sub`, `arg`, and `replace_args`. The `forward!` macro, which defined two additional methods per call (e.g. `note` and `with_note`), is replaced by the `with_fn!` macro, which defines one additional method per call (e.g. `with_note`). It's now also only used when necessary -- not all modifier methods currently need a `with_*` form. (New ones can be easily added as necessary.) All this also requires changing `trait AddToDiagnostic` so its methods take `DiagnosticBuilder` instead of `Diagnostic`, which leads to many mechanical changes. `SubdiagnosticMessageOp` gains a type parameter `G`. There are three subdiagnostics -- `DelayedAtWithoutNewline`, `DelayedAtWithNewline`, and `InvalidFlushedDelayedDiagnosticLevel` -- that are created within the diagnostics machinery and appended to external diagnostics. These are handled at the `Diagnostic` level, which means it's now hard to construct them via `derive(Diagnostic)`, so instead we construct them by hand. This has no effect on what they look like when printed. There are lots of new `allow` markers for `untranslatable_diagnostics` and `diagnostics_outside_of_impl`. This is because `#[rustc_lint_diagnostics]` annotations were present on the `Diagnostic` modifier methods, but missing from the `DiagnosticBuilder` modifier methods. They're now present.
2024-02-06 16:44:30 +11:00
_ => crate::fluent_generated::errors_delayed_at_without_newline,
};
diag.arg("emitted_at", diag.emitted_at.clone());
diag.arg("note", self.note);
let msg = dcx.eagerly_translate_for_subdiag(&diag, msg); // after the `arg` calls
diag.sub(Note, msg, diag.span.primary_span().unwrap_or(DUMMY_SP).into());
Reduce capabilities of `Diagnostic`. Currently many diagnostic modifier methods are available on both `Diagnostic` and `DiagnosticBuilder`. This commit removes most of them from `Diagnostic`. To minimize the diff size, it keeps them within `diagnostic.rs` but changes the surrounding `impl Diagnostic` block to `impl DiagnosticBuilder`. (I intend to move things around later, to give a more sensible code layout.) `Diagnostic` keeps a few methods that it still needs, like `sub`, `arg`, and `replace_args`. The `forward!` macro, which defined two additional methods per call (e.g. `note` and `with_note`), is replaced by the `with_fn!` macro, which defines one additional method per call (e.g. `with_note`). It's now also only used when necessary -- not all modifier methods currently need a `with_*` form. (New ones can be easily added as necessary.) All this also requires changing `trait AddToDiagnostic` so its methods take `DiagnosticBuilder` instead of `Diagnostic`, which leads to many mechanical changes. `SubdiagnosticMessageOp` gains a type parameter `G`. There are three subdiagnostics -- `DelayedAtWithoutNewline`, `DelayedAtWithNewline`, and `InvalidFlushedDelayedDiagnosticLevel` -- that are created within the diagnostics machinery and appended to external diagnostics. These are handled at the `Diagnostic` level, which means it's now hard to construct them via `derive(Diagnostic)`, so instead we construct them by hand. This has no effect on what they look like when printed. There are lots of new `allow` markers for `untranslatable_diagnostics` and `diagnostics_outside_of_impl`. This is because `#[rustc_lint_diagnostics]` annotations were present on the `Diagnostic` modifier methods, but missing from the `DiagnosticBuilder` modifier methods. They're now present.
2024-02-06 16:44:30 +11:00
diag
}
}
/// | Level | is_error | EmissionGuarantee | Top-level | Sub | Used in lints?
/// | ----- | -------- | ----------------- | --------- | --- | --------------
/// | Bug | yes | BugAbort | yes | - | -
/// | Fatal | yes | FatalAbort/FatalError[^star] | yes | - | -
/// | Error | yes | ErrorGuaranteed | yes | - | yes
/// | DelayedBug | yes | ErrorGuaranteed | yes | - | -
/// | ForceWarning | - | () | yes | - | lint-only
/// | Warning | - | () | yes | yes | yes
/// | Note | - | () | rare | yes | -
/// | OnceNote | - | () | - | yes | lint-only
/// | Help | - | () | rare | yes | -
/// | OnceHelp | - | () | - | yes | lint-only
/// | FailureNote | - | () | rare | - | -
/// | Allow | - | () | yes | - | lint-only
/// | Expect | - | () | yes | - | lint-only
///
/// [^star]: `FatalAbort` normally, `FatalError` in the non-aborting "almost fatal" case that is
/// occasionally used.
///
#[derive(Copy, PartialEq, Eq, Clone, Hash, Debug, Encodable, Decodable)]
pub enum Level {
/// For bugs in the compiler. Manifests as an ICE (internal compiler error) panic.
Bug,
/// An error that causes an immediate abort. Used for things like configuration errors,
/// internal overflows, some file operation errors.
Fatal,
/// An error in the code being compiled, which prevents compilation from finishing. This is the
/// most common case.
Error,
/// This is a strange one: lets you register an error without emitting it. If compilation ends
/// without any other errors occurring, this will be emitted as a bug. Otherwise, it will be
/// silently dropped. I.e. "expect other errors are emitted" semantics. Useful on code paths
/// that should only be reached when compiling erroneous code.
DelayedBug,
/// A `force-warn` lint warning about the code being compiled. Does not prevent compilation
/// from finishing.
///
2025-03-18 17:18:54 +00:00
/// Requires a [`LintExpectationId`] for expected lint diagnostics. In all other cases this
/// should be `None`.
2025-03-18 17:18:54 +00:00
ForceWarning,
/// A warning about the code being compiled. Does not prevent compilation from finishing.
/// Will be skipped if `can_emit_warnings` is false.
Warning,
/// A message giving additional context.
Note,
/// A note that is only emitted once.
2022-03-20 20:02:18 +01:00
OnceNote,
/// A message suggesting how to fix something.
Help,
/// A help that is only emitted once.
OnceHelp,
/// Similar to `Note`, but used in cases where compilation has failed. When printed for human
/// consumption, it doesn't have any kind of `note:` label.
2018-02-28 16:17:44 +01:00
FailureNote,
/// Only used for lints.
Allow,
2025-03-18 17:18:54 +00:00
/// Only used for lints. Requires a [`LintExpectationId`] for silencing the lints.
Expect,
}
impl fmt::Display for Level {
2019-02-07 03:53:01 +09:00
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
self.to_str().fmt(f)
}
}
impl Level {
fn color(self) -> ColorSpec {
let mut spec = ColorSpec::new();
match self {
Bug | Fatal | Error | DelayedBug => {
spec.set_fg(Some(Color::Red)).set_intense(true);
}
2025-03-18 17:18:54 +00:00
ForceWarning | Warning => {
spec.set_fg(Some(Color::Yellow)).set_intense(cfg!(windows));
}
2022-03-20 20:02:18 +01:00
Note | OnceNote => {
spec.set_fg(Some(Color::Green)).set_intense(true);
}
Help | OnceHelp => {
spec.set_fg(Some(Color::Cyan)).set_intense(true);
2016-10-18 23:13:02 +05:30
}
2018-02-28 16:17:44 +01:00
FailureNote => {}
2025-03-18 17:18:54 +00:00
Allow | Expect => unreachable!(),
}
2018-02-28 16:17:44 +01:00
spec
}
pub fn to_str(self) -> &'static str {
match self {
Bug | DelayedBug => "error: internal compiler error",
Fatal | Error => "error",
2025-03-18 17:18:54 +00:00
ForceWarning | Warning => "warning",
2022-03-20 20:02:18 +01:00
Note | OnceNote => "note",
Help | OnceHelp => "help",
FailureNote => "failure-note",
2025-03-18 17:18:54 +00:00
Allow | Expect => unreachable!(),
}
}
2018-02-28 16:17:44 +01:00
pub fn is_failure_note(&self) -> bool {
2020-10-26 21:02:48 -04:00
matches!(*self, FailureNote)
2018-02-28 16:17:44 +01:00
}
// Can this level be used in a subdiagnostic message?
fn can_be_subdiag(&self) -> bool {
match self {
2025-03-18 17:18:54 +00:00
Bug | DelayedBug | Fatal | Error | ForceWarning | FailureNote | Allow | Expect => false,
Warning | Note | Help | OnceNote | OnceHelp => true,
}
}
}
// FIXME(eddyb) this doesn't belong here AFAICT, should be moved to callsite.
pub fn elided_lifetime_in_path_suggestion(
source_map: &SourceMap,
n: usize,
path_span: Span,
incl_angl_brckt: bool,
insertion_span: Span,
) -> ElidedLifetimeInPathSubdiag {
let expected = ExpectedLifetimeParameter { span: path_span, count: n };
// Do not try to suggest anything if generated by a proc-macro.
let indicate = source_map.is_span_accessible(insertion_span).then(|| {
let anon_lts = vec!["'_"; n].join(", ");
let suggestion =
if incl_angl_brckt { format!("<{anon_lts}>") } else { format!("{anon_lts}, ") };
IndicateAnonymousLifetime { span: insertion_span.shrink_to_hi(), count: n, suggestion }
});
ElidedLifetimeInPathSubdiag { expected, indicate }
}
2019-11-24 16:59:56 -05:00
pub fn report_ambiguity_error<'a, G: EmissionGuarantee>(
diag: &mut Diag<'a, G>,
ambiguity: rustc_lint_defs::AmbiguityErrorDiag,
) {
diag.span_label(ambiguity.label_span, ambiguity.label_msg);
diag.note(ambiguity.note_msg);
diag.span_note(ambiguity.b1_span, ambiguity.b1_note_msg);
for help_msg in ambiguity.b1_help_msgs {
diag.help(help_msg);
}
diag.span_note(ambiguity.b2_span, ambiguity.b2_note_msg);
for help_msg in ambiguity.b2_help_msgs {
diag.help(help_msg);
}
}
/// Grammatical tool for displaying messages to end users in a nice form.
///
/// Returns "an" if the given string starts with a vowel, and "a" otherwise.
pub fn a_or_an(s: &str) -> &'static str {
let mut chars = s.chars();
let Some(mut first_alpha_char) = chars.next() else {
return "a";
};
if first_alpha_char == '`' {
let Some(next) = chars.next() else {
return "a";
};
first_alpha_char = next;
}
if ["a", "e", "i", "o", "u", "&"].contains(&&first_alpha_char.to_lowercase().to_string()[..]) {
"an"
} else {
"a"
}
}
#[derive(Clone, Copy, PartialEq, Hash, Debug)]
pub enum TerminalUrl {
No,
Yes,
Auto,
}