[cfe-commits] Fix for the -mms-bitfields commandline parameter

[Eli Friedman]

On Mon, Oct 8, 2012 at 2:47 PM, Jeremiah Zanin
<Jeremiah.Zanin at volition-inc.com> wrote:
> The "-mms-bitfields" commandline parameter now properly turns on the ms_struct behavior.
>
> I've never submitted a code change before, hopefully I did this correctly. I included the patch and a new test file.
>
> One result of this change is that __attribute__((ms_struct)) on an enum will produce a warning, for example:
>
> enum __attribute__((ms_struct)) bar {
>         BAR_A = 0,
>         BAR_B,
>         BAR_C
> };
>
> ms-bitfields-test.cpp:4:21: warning: 'ms_struct' attribute ignored
> enum __attribute__((ms_struct)) bar {
>
> Is this desired?

Yes.  Please include a testcase for that.

Please include testcases inside a patch (use "svn add").

This patch would be much simpler if you just implemented isMsStruct as:

bool RecordDecl::isMsStruct(ASTContext &Ctx) {
  return RD->hasAttr<MsStructAttr>() ||
         Ctx.getLangOpts().MSBitfields == 1;
}

(Also, your current patch doesn't properly handle serialization of
RecordDecls; no change is necessary if you don't add a new bit.)

+  if (RecordDecl *RD = dyn_cast<RecordDecl>(D)) {
+    RD->addAttr(::new (S.Context) MsStructAttr(Attr.getRange(), S.Context));
+    RD->setMsStruct(true);
+  }
   else
     S.Diag(Attr.getLoc(), diag::warn_attribute_ignored) << Attr.getName();

We generally prefer brace on the same line as else.

-Eli