> The only problem I see with this email is that you are restricting > the > scope to much to haiku itself and you are addressing a standards body > that may dismiss this based on the current relevancy of Haiku (as > much > as we like it, it is probably not relevant from their point of view). > of Well indeed they might just say "we don't care, just fix your OS" as I am (too often) replied when I ask for a simple fix in some places... This just depends on their openness. We really need to stress out that a) we must stay compatible with BeOS, and BeOS was there before them, and b) we really want to comply to standards. We might want to check for other OSes that potentially use negative errors... Didn't win32 do that ? (ok they aren't POSIX errors anyway, and mingw/ cygwin converts them). Maybe Syllabe, SkyOS or others ? I thought QNX did that but it seems not. VxWorks doesn't either: http://www.slac.stanford.edu/exp/glast/flight/sw/vxdocs/vxworks/errno/errnoNumList.html There must be someone else in this universe ! (that uses <0 errors...) > course programs that refer to error codes as defines would only need > to > be recompiled (in other platforms, I mean), but although it is a > really That's not the problem, they often use return -EFOO, which in BeOS makes it positive back (and sometimes overflows), making it not an error anymore. > bad practice, can we completely rule out the possibility of programs > checking error codes directly by number? They are already known to not be the same on every platform, see: http://article.gmane.org/gmane.comp.video.ffmpeg.devel/55405/match=einval+e2big+eintr François.