MC: Support quoted symbol names

[MaskRay]

gas has supported " quoted symbols since 2015:
https://sourceware.org/pipermail/binutils/2015-August/090003.html

We don't handle \ or " , leading to clang -c --save-temps vs clang -c
difference for the following C code:

int x asm("a\"\\b");

Fix #138390

MC/COFF/safeseh.h looks incorrect. \01 in .safeseh "\01foo" is not a
correct escape sequence. Change it to \

[llvmbot]

@llvm/pr-subscribers-mc

Author: Fangrui Song (MaskRay)

Changes

gas has supported " quoted symbols since 2015:
https://sourceware.org/pipermail/binutils/2015-August/090003.html

Close #138390

---

Full diff: https://github.com/llvm/llvm-project/pull/138817.diff

2 Files Affected:

• (modified) llvm/lib/MC/MCContext.cpp (+21)
• (modified) llvm/test/MC/ELF/symbol-names.s (+16-3)

diff --git a/llvm/lib/MC/MCContext.cpp b/llvm/lib/MC/MCContext.cpp
index f70087e14f702..41caf20b331b2 100644
--- a/llvm/lib/MC/MCContext.cpp
+++ b/llvm/lib/MC/MCContext.cpp
@@ -212,6 +212,27 @@ MCDataFragment *MCContext::allocInitialFragment(MCSection &Sec) {
 MCSymbol *MCContext::getOrCreateSymbol(const Twine &Name) {
   SmallString<128> NameSV;
   StringRef NameRef = Name.toStringRef(NameSV);
+  if (NameRef.contains('\\')) {
+    NameSV = NameRef;
+    size_t S = 0;
+    // Support escaped \\ and \" as in GNU Assembler. GAS issues a warning for
+    // other characters following \\, which we do not implement due to code
+    // structure.
+    for (size_t I = 0, E = NameSV.size(); I < E; ++I) {
+      char C = NameSV[I];
+      if (C == '\\') {
+        switch (NameSV[I + 1]) {
+        case '"':
+        case '\\':
+          C = NameSV[++I];
+          break;
+        }
+      }
+      NameSV[S++] = C;
+    }
+    NameSV.resize(S);
+    NameRef = NameSV;
+  }
 
   assert(!NameRef.empty() && "Normal symbols cannot be unnamed!");
 
diff --git a/llvm/test/MC/ELF/symbol-names.s b/llvm/test/MC/ELF/symbol-names.s
index f605c723d4d4d..8c891052ebd9a 100644
--- a/llvm/test/MC/ELF/symbol-names.s
+++ b/llvm/test/MC/ELF/symbol-names.s
@@ -1,12 +1,25 @@
-// RUN: llvm-mc -triple i686-pc-linux -filetype=obj %s -o - | llvm-readobj --symbols - | FileCheck %s
+// RUN: llvm-mc -triple=x86_64 -filetype=obj %s | llvm-objdump -tdr - | FileCheck %s
 
 // MC allows ?'s in symbol names as an extension.
 
+// CHECK-LABEL:SYMBOL TABLE:
+// CHECK-NEXT: 0000000000000001 l     F .text  0000000000000000 a"b\{{$}}
+// CHECK-NEXT: 0000000000000000 g     F .text  0000000000000000 foo?bar
+// CHECK-NEXT: 0000000000000000 *UND*          0000000000000000 a"b\q{{$}}
+// CHECK-EMPTY:
+
 .text
 .globl foo?bar
 .type foo?bar, @function
 foo?bar:
 ret
 
-// CHECK: Symbol
-// CHECK: Name: foo?bar
+// CHECK-LABEL:<a"b\>:
+// CHECK-NEXT:   callq  {{.*}} <a"b\>
+// CHECK-NEXT:   callq  {{.*}}
+// CHECK-NEXT:     R_X86_64_PLT32 a"b\q-0x4
+.type "a\"b\\", @function
+"a\"b\\":
+  call "a\"b\\"
+/// GAS emits a warning for \q
+  call "a\"b\q"

[joker-eph]

This broke the bots it seems: https://lab.llvm.org/buildbot/#/builders/153/builds/31071