Conversation

Notices

1. Embed this notice
   Max (goldstein@im-in.space)'s status on Friday, 01-Nov-2024 10:15:21 JST Max

   TIL that neither GCC (in C or C++ mode) nor Clang (in C++ mode) can compile code like

   struct f01 { int a, b; };
   struct f02 { f01 a, b; };
   struct f03 { f02 a, b; };
   /* repeat a couple dozen times */
   
   int main() {
   f36 funny;
   memset(&funny, 0, sizeof(funny));
   }
   

   in reasonable time (and compilation time grows exponentially with the number of nesting layers)

   • Embed this notice
     pinskia (pinskia@hachyderm.io)'s status on Friday, 01-Nov-2024 10:15:19 JST pinskia

     in reply to

     @goldstein GCC used to be able to handle that in a reasonible time. I wonder where the slowness is now. I had sped up the front-end back in early 2000s to handle lookups with a lot fields too. If I get some time next week I will see what is going on with GCC. It either could be a front-end issue or middle-end one.

   • Embed this notice
     pinskia (pinskia@hachyderm.io)'s status on Saturday, 02-Nov-2024 02:25:23 JST pinskia

     in reply to

     @goldstein -fsyntax-only for GCC does a little bit more than the front-end (silly internal reason is that some error message happen during gimplification; been trying to fix that for 20 years now).

   • Embed this notice
     Max (goldstein@im-in.space)'s status on Saturday, 02-Nov-2024 02:25:24 JST Max

     in reply to

     • pinskia

     @pinskia I tried that with -fsyntax-only, so that’s probably a frontend issue? I’m not sure where syntax-only compilation stops.

   • Embed this notice
     pinskia (pinskia@hachyderm.io)'s status on Saturday, 02-Nov-2024 05:39:06 JST pinskia

     in reply to

     • feedable

     @feedable @goldstein So this is a regression too from GCC 5 which didn't have check_flexarrays. I will try to get to a full bug report Tomorrow or Sunday. But check_flexarrays is only the C++ front-end one. The C front-end issue has to be different but maybe a similar issue. I will get one for each then.

   • Embed this notice
     feedable (feedable@im-in.space)'s status on Saturday, 02-Nov-2024 05:39:10 JST feedable

     in reply to

     • pinskia

     @goldstein @pinskia so the issue is that during finish_struct a check for flexible array members is performed (via check_flexarrays), and that function performs a recursive search of all the fields, which is what causes the exponential growth in compile times. Should be an easy fix, though, since I think we can just store flexmems_t somewhere in the tree for the struct type

   • Embed this notice
     Max (goldstein@im-in.space)'s status on Saturday, 02-Nov-2024 05:39:11 JST Max

     in reply to

     • pinskia
     • feedable

     @pinskia a friend (@feedable) tinkered with it for a bit and found that it’s stuck in find_flexarrays, presumably because this computation isn’t cached

   • Embed this notice
     pinskia (pinskia@hachyderm.io)'s status on Friday, 08-Nov-2024 01:28:25 JST pinskia

     in reply to

     • feedable

     @feedable @goldstein thanks for the reminder. I will try to get to it today.

   • Embed this notice
     feedable (feedable@im-in.space)'s status on Friday, 08-Nov-2024 01:28:26 JST feedable

     in reply to

     • pinskia

     @goldstein @pinskia Do you the bug reports now?

   • Embed this notice
     pinskia (pinskia@hachyderm.io)'s status on Sunday, 10-Nov-2024 03:06:35 JST pinskia

     in reply to

     • feedable

     @feedable @goldstein sorry about this. I have not forgot about it; I was just fixing a different bug yesterday and didn't get time to create the shell script to create the sources for the test. I will get some time today to do it.

   • Embed this notice
     feedable (feedable@im-in.space)'s status on Sunday, 10-Nov-2024 03:06:36 JST feedable

     in reply to

     • pinskia

     @pinskia @goldstein This is getting kinda awkward... Should I just try to file these myself?

   • Embed this notice
     pinskia (pinskia@hachyderm.io)'s status on Sunday, 10-Nov-2024 06:52:06 JST pinskia

     in reply to

     • feedable

     @feedable @goldstein Sorry it took so long to file the bug. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117516