Conversation

Notices

1. Embed this notice
   Rich Felker (dalias@hachyderm.io)'s status on Saturday, 21-Dec-2024 16:06:57 JST Rich Felker

   LMAO reportedly this broke a bunch of shit in Linux kernel: https://gcc.gnu.org/git/?p=gcc.git;a=commitdiff;h=781634daea8cb788efb33994f4a19df76598542e;hp=600cf1128e49f086522bfddc54490b4e39e30383

   • Embed this notice
     Rob Landley (landley@mstdn.jp)'s status on Saturday, 21-Dec-2024 17:42:01 JST Rob Landley

     in reply to

     @dalias ...?

   • Embed this notice
     Rich Felker (dalias@hachyderm.io)'s status on Saturday, 21-Dec-2024 17:42:01 JST Rich Felker

     in reply to

     • Rob Landley

     @landley They made gcc take advantage of ctz(0) being undefined. Linux uses it.

   • Embed this notice
     Rob Landley (landley@mstdn.jp)'s status on Saturday, 21-Dec-2024 18:12:12 JST Rob Landley

     in reply to

     @dalias If behavior is "undefined", define it to the most obvious thing and stick with it.

     I want a -fdisable-stupid-optimizer for every instance where the plain meaning of the language and the behavior of GCC diverge. (Alas disabling the optimizer entirely loses the old "turbo C for dos did this" stuff like dead code elimination.)

   • Embed this notice
     Rich Felker (dalias@hachyderm.io)'s status on Saturday, 21-Dec-2024 18:12:12 JST Rich Felker

     in reply to

     • Rob Landley

     @landley Both are kinda at fault here. It's always been undefined in the docs, and the insn is even supposedly undefined on some cpu versions (including some x86!) but "happened to produce 32 or 64". GCC making sudden change without realizing it broke Linux and presumably without a chicken switch seems bad tho.