Re: Assertion error when building in Debug mode with MSVC

[Julien Marrec]

hello,

Thanks for the answers, I'm trying to keep up / follow.

> Are the programs in question using the fclose module, either directly or indirectly?

Well, simply doing `m4.exe --version` does exhibit the issue. Not sure if I'm answering the question correctly or not.

I'm looking at the close-out.c, which calls close-stream.c

In close_stream, shouldn't the stdout FILE * be set to NULL if the close worked, so that in the event it's called a second time you can first check if stream == NULL?

Something like this:

diff --git a/lib/close-stream.c b/lib/close-stream.c
index 86f6d6e1d..432907a0c 100644
--- a/lib/close-stream.c
+++ b/lib/close-stream.c
@@ -55,6 +55,7 @@
 int
 close_stream (FILE *stream)
 {
+  if (stream != NULL) {
   const bool some_pending = (__fpending (stream) != 0);
   const bool prev_fail = (ferror (stream) != 0);
   const bool fclose_fail = (fclose (stream) != 0);
@@ -69,10 +70,12 @@ close_stream (FILE *stream)

   if (prev_fail || (fclose_fail && (some_pending || errno != EBADF)))
     {
-      if (! fclose_fail)
+      if (! fclose_fail) {
         errno = 0;
+        stream = NULL;
+      }
       return EOF;
     }

+  }
   return 0;
 }

Best,

Julien

--
Julien Marrec, EBCP, BPI MFBA
Owner at EffiBEM
T: +33 6 95 14 42 13

LinkedIn (en) | (fr) :

Le mar. 25 janv. 2022 à 22:33, Paul Eggert <eggert@cs.ucla.edu> a écrit :

On 1/25/22 08:13, Eric Blake wrote:
> it something that gnulib should work around by
> overriding fclose() to be guaranteed that it has an open fd?

This is a bug that Gnulib's fclose module is already supposed to be
working around. Are the programs in question using the fclose module,
either directly or indirectly? If so, we should investigate why the
fclose module isn't working around the MSVC bug. If not, perhaps we
should change the close-stream module to start depending on the fclose
module.