bjoernager/rust
bjoernager/rust
1
Fork 0
Code Issues Pull requests Activity

Make interpolated_to_tokenstream a method on Nonterminal.

Browse source
This commit is contained in:
Nicholas Nethercote 2019-02-18 10:06:26 +11:00
parent f0d8fbd283
commit 82ad4f1f45
4 changed files with 85 additions and 86 deletions
Download patch file Download diff file Expand all files Collapse all files

163
src/libsyntax/parse/token.rs
View file

@ -86,7 +86,7 @@ impl Lit {
}
}
// See comments in `interpolated_to_tokenstream` for why we care about
// See comments in `Nonterminal::to_tokenstream` for why we care about
// *probably* equal here rather than actual equality
fn probably_equal_for_proc_macro(&self, other: &Lit) -> bool {
mem::discriminant(self) == mem::discriminant(other)
@ -502,87 +502,7 @@ impl Token {
}
}
pub fn interpolated_to_tokenstream(sess: &ParseSess, nt: Lrc<Nonterminal>, span: Span)
-> TokenStream {
// An `Interpolated` token means that we have a `Nonterminal`
// which is often a parsed AST item. At this point we now need
// to convert the parsed AST to an actual token stream, e.g.
// un-parse it basically.
//
// Unfortunately there's not really a great way to do that in a
// guaranteed lossless fashion right now. The fallback here is
// to just stringify the AST node and reparse it, but this loses
// all span information.
//
// As a result, some AST nodes are annotated with the token
// stream they came from. Here we attempt to extract these
// lossless token streams before we fall back to the
// stringification.
let tokens = match *nt {
Nonterminal::NtItem(ref item) => {
prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
}
Nonterminal::NtTraitItem(ref item) => {
prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
}
Nonterminal::NtImplItem(ref item) => {
prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
}
Nonterminal::NtIdent(ident, is_raw) => {
let token = Token::Ident(ident, is_raw);
Some(TokenTree::Token(ident.span, token).into())
}
Nonterminal::NtLifetime(ident) => {
let token = Token::Lifetime(ident);
Some(TokenTree::Token(ident.span, token).into())
}
Nonterminal::NtTT(ref tt) => {
Some(tt.clone().into())
}
_ => None,
};
// FIXME(#43081): Avoid this pretty-print + reparse hack
let source = pprust::nonterminal_to_string(&nt);
let filename = FileName::macro_expansion_source_code(&source);
let (tokens_for_real, errors) =
parse_stream_from_source_str(filename, source, sess, Some(span));
emit_unclosed_delims(&errors, &sess.span_diagnostic);
// During early phases of the compiler the AST could get modified
// directly (e.g., attributes added or removed) and the internal cache
// of tokens my not be invalidated or updated. Consequently if the
// "lossless" token stream disagrees with our actual stringification
// (which has historically been much more battle-tested) then we go
// with the lossy stream anyway (losing span information).
//
// Note that the comparison isn't `==` here to avoid comparing spans,
// but it *also* is a "probable" equality which is a pretty weird
// definition. We mostly want to catch actual changes to the AST
// like a `#[cfg]` being processed or some weird `macro_rules!`
// expansion.
//
// What we *don't* want to catch is the fact that a user-defined
// literal like `0xf` is stringified as `15`, causing the cached token
// stream to not be literal `==` token-wise (ignoring spans) to the
// token stream we got from stringification.
//
// Instead the "probably equal" check here is "does each token
// recursively have the same discriminant?" We basically don't look at
// the token values here and assume that such fine grained token stream
// modifications, including adding/removing typically non-semantic
// tokens such as extra braces and commas, don't happen.
if let Some(tokens) = tokens {
if tokens.probably_equal_for_proc_macro(&tokens_for_real) {
return tokens
}
info!("cached tokens found, but they're not \"probably equal\", \
going with stringified version");
}
return tokens_for_real
}
// See comments in `interpolated_to_tokenstream` for why we care about
// See comments in `Nonterminal::to_tokenstream` for why we care about
// *probably* equal here rather than actual equality
crate fn probably_equal_for_proc_macro(&self, other: &Token) -> bool {
if mem::discriminant(self) != mem::discriminant(other) {
@ -714,6 +634,85 @@ impl fmt::Debug for Nonterminal {
}
}
impl Nonterminal {
pub fn to_tokenstream(&self, sess: &ParseSess, span: Span) -> TokenStream {
// A `Nonterminal` is often a parsed AST item. At this point we now
// need to convert the parsed AST to an actual token stream, e.g.
// un-parse it basically.
//
// Unfortunately there's not really a great way to do that in a
// guaranteed lossless fashion right now. The fallback here is to just
// stringify the AST node and reparse it, but this loses all span
// information.
//
// As a result, some AST nodes are annotated with the token stream they
// came from. Here we attempt to extract these lossless token streams
// before we fall back to the stringification.
let tokens = match *self {
Nonterminal::NtItem(ref item) => {
prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
}
Nonterminal::NtTraitItem(ref item) => {
prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
}
Nonterminal::NtImplItem(ref item) => {
prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
}
Nonterminal::NtIdent(ident, is_raw) => {
let token = Token::Ident(ident, is_raw);
Some(TokenTree::Token(ident.span, token).into())
}
Nonterminal::NtLifetime(ident) => {
let token = Token::Lifetime(ident);
Some(TokenTree::Token(ident.span, token).into())
}
Nonterminal::NtTT(ref tt) => {
Some(tt.clone().into())
}
_ => None,
};
// FIXME(#43081): Avoid this pretty-print + reparse hack
let source = pprust::nonterminal_to_string(self);
let filename = FileName::macro_expansion_source_code(&source);
let (tokens_for_real, errors) =
parse_stream_from_source_str(filename, source, sess, Some(span));
emit_unclosed_delims(&errors, &sess.span_diagnostic);
// During early phases of the compiler the AST could get modified
// directly (e.g., attributes added or removed) and the internal cache
// of tokens my not be invalidated or updated. Consequently if the
// "lossless" token stream disagrees with our actual stringification
// (which has historically been much more battle-tested) then we go
// with the lossy stream anyway (losing span information).
//
// Note that the comparison isn't `==` here to avoid comparing spans,
// but it *also* is a "probable" equality which is a pretty weird
// definition. We mostly want to catch actual changes to the AST
// like a `#[cfg]` being processed or some weird `macro_rules!`
// expansion.
//
// What we *don't* want to catch is the fact that a user-defined
// literal like `0xf` is stringified as `15`, causing the cached token
// stream to not be literal `==` token-wise (ignoring spans) to the
// token stream we got from stringification.
//
// Instead the "probably equal" check here is "does each token
// recursively have the same discriminant?" We basically don't look at
// the token values here and assume that such fine grained token stream
// modifications, including adding/removing typically non-semantic
// tokens such as extra braces and commas, don't happen.
if let Some(tokens) = tokens {
if tokens.probably_equal_for_proc_macro(&tokens_for_real) {
return tokens
}
info!("cached tokens found, but they're not \"probably equal\", \
going with stringified version");
}
return tokens_for_real
}
}
crate fn is_op(tok: &Token) -> bool {
match *tok {
OpenDelim(..) | CloseDelim(..) | Literal(..) | DocComment(..) |