On 2016-08-28 00:07, Paul Eggert wrote: > On further thought, I came up with a patch that should be portable without > overly complicating the code or hurting efficiency, so I installed it > (attached). Please give it a try on your platform.
So, I didn't expect that to work, because the C library in question is the system msvcrt, which is compatible with the msvcrt for MSVC 6.0. That msvcrt was released almost 30 years ago. I have a strong suspicion that all system versions of the msvcrt library out there do not support %lld, even if you limit the search to 64-bit Windows... All in all, I was a bit surprised when it actually did work, and started digging. It seems that mingw-w64 includes overrides for the *printf functions, probably because they are so unportable in msvcrt. Anyway, the overrides includes support for the t length modifier, so a better patch might be a configure check to see if %td works, with machinery to set printint/pI to ptrdiff_t/"t" if it does and fall back to long int/"l" if it doesn't. That would be correct even for weird -- but standards conformant -- systems where even long long is too small for some obscure reason. > I am still leery about changing the GNU coding standards, though. In general > we > shouldn't ask developers to jump through these sorts of hoops for platforms > so > far from the typical GNU environment. My main beef with the GCS on this issue is that it suggests that maintainers should not spend any time at all on this issue, as if that would be contrary to some other goal. It would be much more palatable if it stated that maintainers did not need to spend time on the issue if they didn't want to. Cheers, Peter
