MCP #705: Provide the option -Csymbol-mangling-version=hashed -Z unstable-options to shorten symbol names by replacing them with a digest.

Enrich test cases

compiler/rustc_symbol_mangling/src/hashed.rs

@@ -0,0 +1,43 @@
+use crate::v0;
+use rustc_data_structures::stable_hasher::{Hash64, HashStable, StableHasher};
+use rustc_hir::def_id::CrateNum;
+use rustc_middle::ty::{Instance, TyCtxt};
+
+use std::fmt::Write;
+
+pub(super) fn mangle<'tcx>(
+    tcx: TyCtxt<'tcx>,
+    instance: Instance<'tcx>,
+    instantiating_crate: Option<CrateNum>,
+    full_mangling_name: impl FnOnce() -> String,
+) -> String {
+    // The symbol of a generic function may be scattered in multiple downstream dylibs.
+    // If the symbol of a generic function still contains `crate name`, hash conflicts between the
+    // generic funcion and other symbols of the same `crate` cannot be detected in time during
+    // construction. This symbol conflict is left over until it occurs during run time.
+    // In this case, `instantiating-crate name` is used to replace `crate name` can completely
+    // eliminate the risk of the preceding potential hash conflict.
+    let crate_num =
+        if let Some(krate) = instantiating_crate { krate } else { instance.def_id().krate };
+
+    let mut symbol = "_RNxC".to_string();
+    v0::push_ident(tcx.crate_name(crate_num).as_str(), &mut symbol);
+
+    let hash = tcx.with_stable_hashing_context(|mut hcx| {
+        let mut hasher = StableHasher::new();
+        full_mangling_name().hash_stable(&mut hcx, &mut hasher);
+        hasher.finish::<Hash64>().as_u64()
+    });
+
+    push_hash64(hash, &mut symbol);
+
+    symbol
+}
+
+// The hash is encoded based on `base-62` and the final terminator `_` is removed because it does
+// not help prevent hash collisions
+fn push_hash64(hash: u64, output: &mut String) {
+    let hash = v0::encode_integer_62(hash);
+    let hash_len = hash.len();
+    let _ = write!(output, "{hash_len}H{}", &hash[..hash_len - 1]);
+}

compiler/rustc_symbol_mangling/src/lib.rs

@@ -111,6 +111,7 @@ use rustc_middle::query::Providers;
 use rustc_middle::ty::{self, Instance, TyCtxt};
 use rustc_session::config::SymbolManglingVersion;
 
+mod hashed;
 mod legacy;
 mod v0;
 
@@ -265,6 +266,9 @@ fn compute_symbol_name<'tcx>(
     let symbol = match mangling_version {
         SymbolManglingVersion::Legacy => legacy::mangle(tcx, instance, instantiating_crate),
         SymbolManglingVersion::V0 => v0::mangle(tcx, instance, instantiating_crate),
+        SymbolManglingVersion::Hashed => hashed::mangle(tcx, instance, instantiating_crate, || {
+            v0::mangle(tcx, instance, instantiating_crate)
+        }),
     };
 
     debug_assert!(

compiler/rustc_symbol_mangling/src/v0.rs

@@ -116,10 +116,7 @@ impl<'tcx> SymbolMangler<'tcx> {
     /// * `x > 0` is encoded as `x - 1` in base 62, followed by `"_"`,
     ///   e.g. `1` becomes `"0_"`, `62` becomes `"Z_"`, etc.
     fn push_integer_62(&mut self, x: u64) {
-        if let Some(x) = x.checked_sub(1) {
-            base_n::push_str(x as u128, 62, &mut self.out);
-        }
-        self.push("_");
+        push_integer_62(x, &mut self.out)
     }
 
     /// Push a `tag`-prefixed base 62 integer, when larger than `0`, that is:
@@ -138,45 +135,7 @@ impl<'tcx> SymbolMangler<'tcx> {
     }
 
     fn push_ident(&mut self, ident: &str) {
-        let mut use_punycode = false;
-        for b in ident.bytes() {
-            match b {
-                b'_' | b'a'..=b'z' | b'A'..=b'Z' | b'0'..=b'9' => {}
-                0x80..=0xff => use_punycode = true,
-                _ => bug!("symbol_names: bad byte {} in ident {:?}", b, ident),
-            }
-        }
-
-        let punycode_string;
-        let ident = if use_punycode {
-            self.push("u");
-
-            // FIXME(eddyb) we should probably roll our own punycode implementation.
-            let mut punycode_bytes = match punycode::encode(ident) {
-                Ok(s) => s.into_bytes(),
-                Err(()) => bug!("symbol_names: punycode encoding failed for ident {:?}", ident),
-            };
-
-            // Replace `-` with `_`.
-            if let Some(c) = punycode_bytes.iter_mut().rfind(|&&mut c| c == b'-') {
-                *c = b'_';
-            }
-
-            // FIXME(eddyb) avoid rechecking UTF-8 validity.
-            punycode_string = String::from_utf8(punycode_bytes).unwrap();
-            &punycode_string
-        } else {
-            ident
-        };
-
-        let _ = write!(self.out, "{}", ident.len());
-
-        // Write a separating `_` if necessary (leading digit or `_`).
-        if let Some('_' | '0'..='9') = ident.chars().next() {
-            self.push("_");
-        }
-
-        self.push(ident);
+        push_ident(ident, &mut self.out)
     }
 
     fn path_append_ns(
@@ -836,3 +795,62 @@ impl<'tcx> Printer<'tcx> for SymbolMangler<'tcx> {
         Ok(())
     }
 }
+/// Push a `_`-terminated base 62 integer, using the format
+/// specified in the RFC as `<base-62-number>`, that is:
+/// * `x = 0` is encoded as just the `"_"` terminator
+/// * `x > 0` is encoded as `x - 1` in base 62, followed by `"_"`,
+///   e.g. `1` becomes `"0_"`, `62` becomes `"Z_"`, etc.
+pub(crate) fn push_integer_62(x: u64, output: &mut String) {
+    if let Some(x) = x.checked_sub(1) {
+        base_n::push_str(x as u128, 62, output);
+    }
+    output.push('_');
+}
+
+pub(crate) fn encode_integer_62(x: u64) -> String {
+    let mut output = String::new();
+    push_integer_62(x, &mut output);
+    output
+}
+
+pub(crate) fn push_ident(ident: &str, output: &mut String) {
+    let mut use_punycode = false;
+    for b in ident.bytes() {
+        match b {
+            b'_' | b'a'..=b'z' | b'A'..=b'Z' | b'0'..=b'9' => {}
+            0x80..=0xff => use_punycode = true,
+            _ => bug!("symbol_names: bad byte {} in ident {:?}", b, ident),
+        }
+    }
+
+    let punycode_string;
+    let ident = if use_punycode {
+        output.push('u');
+
+        // FIXME(eddyb) we should probably roll our own punycode implementation.
+        let mut punycode_bytes = match punycode::encode(ident) {
+            Ok(s) => s.into_bytes(),
+            Err(()) => bug!("symbol_names: punycode encoding failed for ident {:?}", ident),
+        };
+
+        // Replace `-` with `_`.
+        if let Some(c) = punycode_bytes.iter_mut().rfind(|&&mut c| c == b'-') {
+            *c = b'_';
+        }
+
+        // FIXME(eddyb) avoid rechecking UTF-8 validity.
+        punycode_string = String::from_utf8(punycode_bytes).unwrap();
+        &punycode_string
+    } else {
+        ident
+    };
+
+    let _ = write!(output, "{}", ident.len());
+
+    // Write a separating `_` if necessary (leading digit or `_`).
+    if let Some('_' | '0'..='9') = ident.chars().next() {
+        output.push('_');
+    }
+
+    output.push_str(ident);
+}