Re: bug#33847: 27.0.50; emacsclient does not find server socket

[Paul Eggert]

On 7/24/21 11:32 PM, Eli Zaretskii wrote:

> > No modules are affected by the --disable-year2038 option on MS-Windows.

It turns out that I was wrong about that. (I don't normally look at the 
MS-Windows part of Gnulib and misunderstood some of the code I was 
reading.) Please see gnulib/m4/year2038.m4 for details. This file is in 
the patches I sent, or you can see it directly here:

https://git.savannah.gnu.org/cgit/gnulib.git/tree/m4/year2038.m4

This code knows about MS-Windows, Mingw, _USE_32BIT_TIME_T, 
__MINGW_USE_VC2005_COMPAT, and so forth, and attempts to do the right 
thing. As near as I can make out it should work for the scenario you 
describe, but I don't use MS-Windows so I could well be wrong. If I'm 
wrong and this code doesn't do what you want, I suggest contacting 
bug-gnulib to alert Bruno Haible, who wrote that part of the code. I'll 
cc bug-gnulib so that Bruno sees this email. (Bruno, this discussion is 
at <https://bugs.gnu.org/33847#161>.)

Here's some more background. There are two Gnulib modules involved.

The largefile module ensures that a program can open/stat/etc. all 
files, by widening types like off_t, dev_t and time_t if necessary. If 
it finds that time_t is narrower than what the system can support, it 
attempts to widen time_t; if this attempt fails it issues a warning but 
continues.

The year2038 module is stricter: it insists that time_t be at least 64 
bits and aborts 'configure' otherwise. (Strictly speaking, it should 
insist only on at least 33 bits (or 32 bits unsigned); I suppose I 
should look into fixing that.)

The Emacs patches that I sent do not use the year2038 module, because I 
expected that you wouldn't want to worry about the year 2038. The 
year2038 module is used by GNU packages like coreutils where Y2038 is a 
problem even now, due to the long lead times and lack of updatability on 
systems that use these other GNU packages.

> So therefore my question seems to be even more important than I
> thought, and I'm still asking which Gnulib modules are affected by
> this, because I'd need to audit them carefully to see whether the
> 32-bit MS-Windows build with mingw.org's MinGW could be affected.

There should be no need to audit, because Gnulib still supports 
platforms that have only 32-bit time_t.

Gnulib is agnostic about time_t width, and is supposed to work even if 
time_t is 40 bits (which it is on a very few mainframes) or any other 
width. We regularly test it only on 32- and 64-bit time_t, though.