Hi Willy, Thank you for looking into this. Please let me know if any additional information or testing from my side would be helpful.
Best, Yongqiang Tian Assistant Professor at Monash University, Australia On Thu, 20 Aug 2026 at 13:22, Willy Tarreau <[email protected]> wrote: > Hi! > > On Thu, Aug 20, 2026 at 09:27:13AM +1000, Yongqiang Tian wrote: > > Commit 9752f802b3869c96446f7e84e43d89360695c0f8 introduced > > eliminate_tail_calls to turn the final call of a function into a jump. > > Commit 431b6b1cd92a54f4a53b66a5ff27a64381ff9a23 then applied it to > > complain() and complain_with_dbg() so that these wrappers would not > appear > > in diagnostic backtraces. > > > > The GCC expansion enables sibling-call optimization. The Clang expansion > > currently uses disable_tail_calls, which has the opposite meaning: Clang > > documents it as preventing tail-call optimization inside the attributed > > function. > > Interesting, because I remember that when I found this attribute, it was > mentioned what you said, but it didn't be have as described for me and I > concluded that the explanation was either unclear or just mistaken. > > the fact that you also had to pass noinline to the callee might explain > why I observed a different behavior: since the compiler is usually free > to inline what it wants, it's possible that the differences I observed > were in fact due to its decision to inline the callee or not. > > I will retest with your change on different machines and merge it if it > looks OK. > > Many thanks for testing and reporting this! > Willy >
