Bug#1134097: webkit2gtk: Please refresh no-must-tail.patch to address FTBFS on alpha and sh4

[Adrian Bunk]

On Fri, Apr 17, 2026 at 01:55:24PM +0200, John Paul Adrian Glaubitz wrote:
> Hi Adrian,
> 
> On Fri, 2026-04-17 at 14:49 +0300, Adrian Bunk wrote:
> > On Thu, Apr 16, 2026 at 02:06:16PM +0200, Alberto Garcia wrote:
> > > Control: tags -1 pending
> > > 
> > > On Thu, Apr 16, 2026 at 12:29:00PM +0200, John Paul Adrian Glaubitz wrote:
> > > > the patch no-must-tail.patch needs an update to fix the FTBFS on
> > > > alpha and sh4.
> > > 
> > > Applied, thanks
> > > ...
> > 
> > I wonder whether whitelisting the architectures where it works might be 
> > upstreamable?
> > 
> > Source/ThirdParty/skia/modules/skcms/src/skcms_internals.h says:
> >         // [[clang::musttail]] is great for performance, but it's not well supported and we run into
> >         // a variety of problems when we use it. Fortunately, it's an optional feature that doesn't
> >         // affect correctness
> > 
> > The problem is that gcc implements it as a "must" since there are 
> > other use cases where musttail is relevant for correctness.
> > 
> > The same issue was fixed in protobuf with
> >   https://github.com/protocolbuffers/protobuf/commit/719f3037032b2e952afe7fc49152cc4be38fa7a3
> 
> Yeah, I wanted to have this properly fixed in Skia for a while now, but I didn't have the time yet.
> 
> Does it work on x86_64 and arm64 only? If yes, I would go ahead and use that.

No, based on the ifdefs in skia it seems to work fine e.g. on sparc64.

This is a tradeoff between maintainability and a potential marginal
performance improvement, with the latter being mostly irrelevant on
architectures where no one really cares about the performance.

In practice the whitelisted architectures should be a superset of the 
architectures where it works and upstream cares about (and tests).

> Adrian

cu
Adrian