Re: lib/56323: fvwrite assertion failures during install

[RVP <rvp%SDF.ORG@localhost>]

The following reply was made to PR lib/56323; it has been noted by GNATS.

From: RVP <rvp%SDF.ORG@localhost>
To: gnats-bugs%netbsd.org@localhost
Cc: Andreas Gustafsson <gson%gson.org@localhost>
Subject: Re: lib/56323: fvwrite assertion failures during install
Date: Thu, 22 Jul 2021 22:01:22 +0000 (UTC)

 On Thu, 22 Jul 2021, Andreas Gustafsson wrote:
 
 > When installing NetBSD-current/i386 using sysinst, multiple error
 > messages like this one are printed to the terminal during the set
 > extraction phase:
 >
 >  assertion "fp->_w >= 0" failed: file "/tmp/build/2021.07.22.04.46.29-i386-laptop/src/lib/libc/stdio/fvwrite.c", line 60, function "flush_adj"
 >
 > The installation succeeds nonetheless.
 >
 > The problem started around the time of this commit:
 >
 >  2021.07.16.12.34.10 christos src/lib/libc/stdio/fvwrite.c 1.28
 >
 
 That's one of my assertions. Andreas, can you tell more about the
 test setup where this is being triggered? On a real machine (Intel
 Core i3 with 4GB RAM) one of the tests I ran before asking Christos
 to make that commit is this one (and I didn't see this assertion):
 
 ---
 #!/bin/sh
 
 set -eu -o pipefail
 for f
 do      mkdir X
          echo "$(basename "$f") (1)"
          progress -zf $f tar -C X -xf -
          echo "$(basename "$f") (2)"
          progress -f $f tar -C X -xzf -
          sudo rm -rf X
 done
 ---
 
 In any case, until this is sorted out, I've requested Christos to
 revert the fwrite changes back to the NetBSD originals. Even in
 -HEAD there is no reason to subject users to not-fully-completed
 changes.
 
 -RVP