Paul Eggert (2023/09/18 13:46 -0700): > On 2023-09-18 02:55, Sébastien Hinderer wrote: > > the project currently has the convention of including > > system headers before its ocnfiguraiotn headers > > Ouch. The Autoconf manual explicitly says "The package should ‘#include’ the > configuration header file before any other header files" > <https://www.gnu.org/savannah-checkouts/gnu/autoconf/manual/autoconf-2.71/html_node/Configuration-Headers.html> Many thanks for the sourced response! > so your project's convention is not supported. That's sad andprobably partly also my responsibility because I was the person in charge of moving from a hand-qritten configure script to one generated from Autoconf and, at the time, I was a completely newbie to these things and the task was so overwhelming that I didn't pay attention enough to this aspect. For sure, we woudlbe in a better situation today regarding this if I had raised the point at that time. I'll still investigate to see whether there is something that can still be done to change this convention. > > One option I am ocnsidering is to use themacro to determine wether the > > macro needs to be defined and then use its result to determine whether > > -D_FILE_OFFSET_BITS=64 should be addeed to the command-line or not. > > Something like that should be a workaround, yes. However, you'll likely need > to add -D options for more than just _FILE_OFFSET_BITS. You are perfectly correct! Many thanks for having raised that. I was pretty sure we were not using this in other places, but I was also perfectly wrong. > I suggest looking > also into the macros _LARGEFILE_SOURCE, _TIME_BITS, _LARGE_FILES, and > __MINGW_USE_VC2005_COMPAT, depending on which platforms you're trying to > port to. I will look into those, too. Thanks. > Also, if you're using any other Autoconf macros you may need to do something > similar. Pretty much any 'undef _*' in your config.hin is suspect. There are > 37 such macros in GNU coreutils, e.g., _GNU_SOURCE, _DARWIN_C_SOURCE, > _REGEX_LARGE_OFFSETS. If GNU coreutils used your project's conventions it'd > likely need to prepare -D options for all of these, and maybe even > more. Just to make sure. Are you sayingthat there should normally not be any undef in the config headers for macros whose name starts with a _ sign? Am I understanding you correctly? Best wishes, Sébastien.
