Don't generate crash report when inline asm can't be parsed

[steven_wu]

Summary

Crash report is currently generated when inline asm can't be parsed
and no DiagnosticHandler is registered. However, clang doesn't set up
DiagnosticHandle for inline asm when the input is bitcode and
compiling invalid asm from bitcode shouldn't be a compiler crash.

[echristo]

Not sure I agree with this - it changes the behavior (crash or not) based on whether or not there's bitcode as the input. There are lots of places that the behavior will continue as crashing. I think what should happen if you don't want this type of behavior is to avoid setting a global error handler via clang.

Thoughts? More explanation on your logic?

-eric

[steven_wu]

I think my patch only disable the crash report when Parser->run returns error in EmitInlineAsm. It is not related to what input type is. Is there any circumstance when it is a compiler crash when Parser->run returns error?
I thought about putting a error handler in clang but I don’t think clang can emit any diagnostics that is more helpful than backend.

lib/CodeGen/AsmPrinter/AsmPrinterInlineAsm.cpp

void AsmPrinter::EmitInlineAsm(StringRef Str, const MDNode *LocMDNode,
   }
 
   emitInlineAsmStart(STIOrig);
   // Don't implicitly switch to the text section before the asm.
   int Res = Parser->Run(/*NoInitialTextSection*/ true,
                         /*NoFinalize*/ true);
   emitInlineAsmEnd(STIOrig, STI.get());
   if (Res && !HasDiagHandler)
-    report_fatal_error("Error parsing inline asm\n");
+    report_fatal_error("Error parsing inline asm\n", false);
 }
 
 static void EmitMSInlineAsmStr(const char *AsmStr, const MachineInstr *MI,
                                MachineModuleInfo *MMI, int InlineAsmVariant,
                                AsmPrinter *AP, unsigned LocCookie,
                                raw_ostream &OS) {
   // Switch to the inline assembly variant.
   OS << "\t.intel_syntax\n\t";