Conversation
Occurs most likely when certain g++ optimizations are enabled, so this code needs to be split into parts to make sure we don't run into false-positive Issue reports when some user tries out some compiler optimizations.
|
What compiler optimizations are this? Maybe there is a deeper issue and reordering the strings like this just appears to fix it. I think if a compiler wrongly optimizes adding C++ strings this way, it should be a compiler bug :) |
|
Haha, I have put so many switches on my fork I'm quite unsure which one it was (or maybe some main -O flag). But You know, it occured in GCC 4.8, as well is still occurring in GCC 5.3 (maybe it is a bug in windows mingw only? - unfortunately I lost old commit notes on this due to recreating fork long time ago). |
|
Ah sorry, here it is: "GCC 4.8.5 workaround for x64 windows compilations" - this was my old commit: MrKrzYch00@756aa47 So it seems to be related to compiling x64 version of zopflipng on windows (maybe it was with an additional AVX tuning). GCC for windows is often quite broken, for example x86 crashes with -fipa-cp-clone or -fprofile-generate with pthreads on EncodeTree when march or mcpu is passed. And they are fixing it and breaking with each release. :) |
2 participants
Occurs most likely when certain g++ optimizations are enabled, so this
code needs to be split into parts to make sure we don't run into
false-positive Issue reports when some user tries out some compiler
optimizations.