[ping] RE: [PATCH] review for MC ELF header e_flags setting

[Robinson, Paul]

Sorry for coming late to the party; I've only recently realized that
something I'm doing requires me to set an e_flags bit and that this
stuff might be helpful.

One comment on the changes:

>  void MipsAsmPrinter::EmitEndOfAsmFile(Module &M) {
> 
> +  if (OutStreamer.hasRawTextSupport()) return;
> +
>    // Emit Mips ELF register info
>    Subtarget->getMReginfo().emitMipsReginfoSectionCG(
>               OutStreamer, getObjFileLowering(), *Subtarget);
> +  MipsELFStreamer & MES = static_cast<MipsELFStreamer &>(OutStreamer);
> +  MES.emitELFHeaderFlagsCG(*Subtarget);
>  }

Okay, for writing a .s file you skip the flags stuff, and for writing
a .o file you set the flags in the assembler object.  Then the ELF
streamer just uses the flags it was told to use.

I didn't see a change to MipsAsmParser.cpp that would set the flags
in the standalone assembler?  It looked like you did have a change of
that nature for ARMAsmParser.


Granted it's a little late to suggest other approaches, but wouldn't
it be more natural to deal with this in the target-specific assembler
backend?  Now that you have ELFObjectWriter::WriteHeader getting a
handle on MCAssembler, it can easily call a new hook in the
(target-specific) assembler backend to return the (target-specific)
e_flags word.

If there's info you need that the assembler backend doesn't have
(which may be why you don't have a MipsAsmParser change?) that makes
things trickier of course; maybe you'd need to invent new assembler
directives to pass that from AsmPrinter to the backend.  This could
require more tweaks, like making MCELFStreamer::EmitAssemblerFlag call
the assembler backend, the way MCMachOStreamer already does, but then
the assembler backend can remember the flags.

Apologies if you've already worked through that idea and abandoned it;
like I said, I didn't realize this could be useful to me until lately.

Thanks,
--paulr