Conversation

Notices

1. Embed this notice
   Andy Wingo (wingo@mastodon.social)'s status on Thursday, 15-Jun-2023 06:27:35 JST Andy Wingo

   the people require __attribute__((__musttail__)) in gcc. we require it

   • clacke likes this.
   • Embed this notice
     Laurence Tratt (ltratt@mastodon.social)'s status on Thursday, 15-Jun-2023 06:27:39 JST Laurence Tratt

     in reply to

     • Stephen Kell
     • CF Bolz-Tereick

     @stephenrkell @cfbolz @wingo Many years ago I wrote a blog post on the downsides of tail calls on debugging https://tratt.net/laurie/blog/2004/tail_call_optimization.html only to discover later that at least one paper (Smith and Ungar, cited at the end in an update) had also alluded to the problem. I've sometimes wondered: how soon after tail calls were introduced did the debugging downsides become known?

     clacke likes this.
   • Embed this notice
     Stephen Kell (stephenrkell@recurse.social)'s status on Thursday, 15-Jun-2023 06:27:44 JST Stephen Kell

     in reply to

     • CF Bolz-Tereick

     @cfbolz @wingo Tangential, but one of the tricky things I sometimes think about is how to reconcile tail-call optimisation with debuggability. Do you have a take?

     E.g. with tail recursion, even supposing you can reconstruct the stack depth and locals from ambient other state, it's not expressible within DWARF to 'synthesise n frames, computed like so'... and it'd probably be a misfeature even if it were. What should happen?

     Sibling tail-call opt is even trickier... I often tell it to -f off.

   • Embed this notice
     CF Bolz-Tereick (cfbolz@mastodon.social)'s status on Thursday, 15-Jun-2023 06:27:47 JST CF Bolz-Tereick

     in reply to

     @wingo oh yes, that would be excellent

   • Embed this notice
     Andy Wingo (wingo@mastodon.social)'s status on Thursday, 15-Jun-2023 06:27:49 JST Andy Wingo

     in reply to

     • Laurence Tratt
     • Stephen Kell
     • CF Bolz-Tereick

     @ltratt @stephenrkell @cfbolz i think in this concrete case there is no issue because it's something the programmer intends, rather than an unexpected optimization. would be weird to see a stack frame when you explicitly asked not to have one :)

     clacke likes this.
   • Embed this notice
     Andy Wingo (wingo@mastodon.social)'s status on Thursday, 15-Jun-2023 06:27:51 JST Andy Wingo

     in reply to

     • Laurence Tratt
     • Stephen Kell
     • CF Bolz-Tereick

     @ltratt @stephenrkell @cfbolz ok gotta be pedantic here, what is being opted into is proper tail calls, not tail call *optimization*; clang::musttail means tail calls even when -O0

     clacke likes this.
   • Embed this notice
     Laurence Tratt (ltratt@mastodon.social)'s status on Thursday, 15-Jun-2023 06:27:52 JST Laurence Tratt

     in reply to

     • Stephen Kell
     • CF Bolz-Tereick

     @wingo @stephenrkell @cfbolz I agree -- when one opts in to TCO, one implicitly accepts the lack of stack traces.

   • Embed this notice
     Stephen Kell (stephenrkell@recurse.social)'s status on Thursday, 15-Jun-2023 06:27:54 JST Stephen Kell

     in reply to

     • Laurence Tratt
     • CF Bolz-Tereick

     @wingo @ltratt @cfbolz Time for my pedantry... let's not mix 'stack traces' (a debugging feature) with 'stack frame' (a memory allocation). Stack traces already include 'frames' that do not correspond to any physical frame, thanks to inlining

     (That is another example of something that can be done at -O0, via always_inline, and straddles 'correctness' vs 'optimisation'. For TC[O] it depends whether you classify asymptotic space usage as functional or extrafunctional... the use case matters.)

     clacke likes this.
   • Embed this notice
     clacke (clacke@libranet.de)'s status on Thursday, 15-Jun-2023 06:27:54 JST clacke

     in reply to

     • James Henstridge
     @jamesh @wingo Being a normal person is overrated.
   • Embed this notice
     James Henstridge (jamesh@aus.social)'s status on Thursday, 15-Jun-2023 06:28:00 JST James Henstridge

     in reply to

     @wingo Or you could just learn to use while loops like a normal person 😃