[Demangle][Rust] Parse dot suffix

[tmiasko]

Summary

Allow mangled names to include an arbitrary dot suffix, akin to vendor
specific suffix in Itanium mangling.

Primary motivation is a support for symbols renamed during ThinLTO
import / promotion (ThinLTO is the default configuration for optimized
builds in rustc).

[dblaikie]

llvm/lib/Demangle/RustDemangle.cpp

123–130

Is this similar to the Itanium implementation? Or would it be possible/reasonable/simpler (maybe not, I don't know) to strip the '.' before doing any parsing at all? (or are there cases where the rust mangling can include '.' and so finding the trailing '.' must be a context-sensitive operation)

[tmiasko]

• Strip dot suffix before parsing
• Include the dot in demangled output

[tmiasko]

llvm/lib/Demangle/RustDemangle.cpp

123–130

There is some variation in existing implementations. LLVM Itanium demangler prints _Z1fv.llvm.1234 as f() (.llvm.1234). The c++filt prints it as f() [clone .llvm.1234]. The rustc-demangle crate, which is used by rustc, omits .llvm.* suffixes from output, but prints others verbatim. The Rust demangler in c++filt rejects symbols with suffixes altogether.

The Rust mangling does not use . character by itself. So yes, suffix can be stripped ahead of time, which seems simpler than it was before.

[dblaikie]

Looks good, thanks!

llvm/lib/Demangle/RustDemangle.cpp

bool Demangler::demangle(StringView Mangled) {
   Print = true;
   RecursionLevel = 0;
   BoundLifetimes = 0;
 
   if (!Mangled.consumeFront("_R")) {
     Error = true;
     return false;
   }
-  Input = Mangled;
+  size_t Dot = Mangled.find('.');
+  Input = Mangled.substr(0, Dot);
+  StringView Suffix = Mangled.dropFront(Dot);
 
   demanglePath(rust_demangle::InType::No);
 
   if (Position != Input.size()) {
     SwapAndRestore<bool> SavePrint(Print, false);
     demanglePath(InType::No);
   }
 
   if (Position != Input.size())
     Error = true;
 
+  if (!Suffix.empty()) {
+    print(" (");
+    print(Suffix);
+    print(")");
+  }
+
   return !Error;
 }
 
 // Demangles a path. InType indicates whether a path is inside a type. When
 // LeaveOpen is true, a closing `>` after generic arguments is omitted from the
 // output. Return value indicates whether generics arguments have been left
 // open.
 //

llvm/test/Demangle/rust.test

CHECK: _RNvB_1a
        _RNvB_1a
 
 CHECK: _RIC7backrefSB9_E
        _RIC7backrefSB9_E
 
 CHECK: _RIC7backrefKBa_E
        _RIC7backrefKBa_E
 
+; Dot suffix
+
+CHECK: dot (.llvm.1234)
+       _RC3dot.llvm.1234
+
+CHECK: dot (.llvm.6789)
+       _RC3dotC5crate.llvm.6789
+
 ; Invalid mangled characters
 
 CHECK: _RNvC2a.1c
        _RNvC2a.1c
 
 CHECK: _RNvC2a$1c
        _RNvC2a$1c