Re: [Patch] Windows relation extension failure at 2GB and 4GB

On Thu, Nov 6, 2025 at 10:20 PM Thomas Munro <thomas.munro@gmail.com> wrote:
> The only system left with narrow off_t is Windows.

By the way, that's not always true: Meson + MinGW have a 64-bit off_t
on Windows, because meson decides to set -D_FIILE_OFFSET_BITS=64 for
all C compilers except msvc[1] (only other exclusion is macOS, but
that is 64-bit only these days; there are other systems like FreeBSD
where sizeof(off_t) is always 8 but it doesn't seem to know about that
or bother to check), and MinGW's headers react to that.  I suspect
autoconf's AC_SYS_LARGEFILE would do that too with MinGW, IDK, but
configure.ac doesn't call it for win32 by special condition.  That
creates a strange difference between meson and autoconf builds IMHO,
but if we resolve that in the only direction possible AFAICS we'd
still have a strange difference between MSVC and MinGW.

Observing that mess, I kinda wonder what would happen if we just used
a big hammer to redefine off_t to be __int64 ourselves.  On the face
of it, it sounds like an inherently bad idea that could bite you when
interacting with libraries whose headers use off_t.  On the other
hand, the world of open source libraries we care about might already
be resistant to that chaos, if libraries are being built with and
without -D_FIILE_OFFSET_BITS=64 willy-nilly, or they actually can't
deal with large files at all in which case that's something we'd have
to deal with whatever we do.  I don't know, it's just a thought that
occurred to me while contemplating how unpleasant it is to splatter
pgoff_t all over our tree, and yet *still* have to tread very
carefully with the boundaries of external libraries that might be
using off_t, researching each case...

[1]
https://github.com/mesonbuild/meson/blob/97a1c567c9813176e4bec40f6055f228b2121609/mesonbuild/compilers/compilers.py#L1144