Messages in this thread | | | From | Roland McGrath <> | Subject | Re: [PATCH 0/3] tracepoints: delay argument evaluation | Date | Wed, 20 May 2009 02:18:17 -0700 (PDT) |
| |
> But I think we should be asking GCC to get fixed. Since inline functions > don't act like a compiler barrier anyway, they might as well not force > argument evaluation either.
I quite agree. I'm all for better compilation. However, usual practice in the kernel has been to complain about introducing C code sequences that add a few instructions when compiled by compilers over 5 years old, let alone the very latest one. I think it was the expectation of that sort of sensitive reaction to the hot-path impact of unused tracepoints that motivated Jason's attempt at microoptimization.
Thanks, Roland
| |