[PATCH] D139986: [clang][TypePrinter] Teach isSubstitutedDefaultArgument about integral types

[Adrian Prantl via Phabricator via cfe-commits]

aprantl added inline comments.


================
Comment at: clang/test/CodeGenObjCXX/encode.mm:93-94
   // FIXME: This difference is due to D76801. It was probably an unintentional change. Maybe we want to undo it?
-  // CHECKCXX98: @_ZN11rdar93574002ggE ={{.*}} constant [49 x i8] c"{vector<float, rdar9357400::fixed<4, -1> >=[4f]}\00"
-  // CHECKCXX20: @_ZN11rdar93574002ggE ={{.*}} constant [48 x i8] c"{vector<float, rdar9357400::fixed<4, -1>>=[4f]}\00"
+  // CHECKCXX98: @_ZN11rdar93574002ggE ={{.*}} constant [45 x i8] c"{vector<float, rdar9357400::fixed<4> >=[4f]}\00"
+  // CHECKCXX20: @_ZN11rdar93574002ggE ={{.*}} constant [44 x i8] c"{vector<float, rdar9357400::fixed<4>>=[4f]}\00"
   extern const char gg[] = @encode(vector4f);
----------------
dblaikie wrote:
> aprantl wrote:
> > aprantl wrote:
> > > dblaikie wrote:
> > > > aprantl wrote:
> > > > > Michael137 wrote:
> > > > > > Michael137 wrote:
> > > > > > > dblaikie wrote:
> > > > > > > > @aprantl any idea if this is good/OK? (I guess it probably is - but maybe these strings were never meant to ignore/suppress default arguments of any kind? or maybe this is an ABI sort of thing where it suppressing some but not others is now unchangeable?)
> > > > > > > Good point. There was a thread on the cfe mailing list a while ago about the last time this broke: https://lists.llvm.org/pipermail/cfe-dev/2020-November/067194.html
> > > > > > > 
> > > > > > > This was @rsmith's stance:
> > > > > > > ```
> > > > > > > I think some of the other recent TypePrinter changes might also risk
> > > > > > > changing the @encode output. Generally it seems unwise for @encode to be
> > > > > > > using the type pretty-printer if it wants to be ABI-stable; I don't think
> > > > > > > it's reasonable to expect any guarantees as to the stability of
> > > > > > > pretty-printed type names. I think USR generation suffers from similar
> > > > > > > problems; it too uses the type pretty-printer to generate
> > > > > > > supposedly-ABI-stable keys in at least some cases.
> > > > > > > ```
> > > > > > see https://reviews.llvm.org/D90622
> > > > > To me it really looks like the intention of the feature is to not substitute default parameters. But if we stop doing this now it will likely result in a surprising code size increase, that may not be considered worth it compared to the risk of breaking ABI by changing a default template parameter.
> > > > > 
> > > > > As far as this patch is concerned, it's neutral to this decision (which may not have been a conscious one).
> > > > > It's certainly not good that every type printer change is an ABI break.
> > > > Awesome, thanks for tracking down that context @Michael137.
> > > > 
> > > > Not quite sure I'm following you @aprantl, but I think you're saying this change is OK/seems consistent with other changes?
> > > The current code (for @encode and USRs) seems to assume that TypePrinter output is stable.
> > > 
> > > I (personally) think that assumption ought to be wrong, because otherwise we'd never be able to make improvements such as this patch.
> > > 
> > > Aside from my personal preferences, practically this patch causes a problem for shipping an Objective-C compiler, since this is an ABI-breaking change. That's why I'd like to hear from someone with more insight into how `@encoding` is used in Objective-C wether this is something we need to be concerned about. If it is a concern we may need to add a TypePrinter configuration optimized for stability that preserves the current output format in eternity.
> > To summarize an offline conversation about this: Because there are no system frameworks that vend Objective-C++ types we are not concerned by a potential ABI break caused by this patch.
> > 
> > LGTM!
> Awesome - thanks for getting that info, @aprantl - any idea if we could write this down somewhere? (in comments in tests that verify `@encoding` if there aren't too many of them?) So it's easy/clear next time.
@Michael137 You could add a comment to this very test here, stating that the fact that the @encoding for C++ is effectively dependent on the TypePrinter implementation is a known bug.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D139986/new/

https://reviews.llvm.org/D139986