Re: Lilypond 2.22.1 fails with GitHub Actions

[Jean Abou Samra]

Le 08/11/2022 à 01:12, Vít Novotný a écrit :
> On Mon, Nov 07, 2022 at 09:20:24PM +0100, Jean Abou Samra wrote:
> > Le 07/11/2022 à 20:59, Vít Novotný a écrit :
> > > Furthermore, the error message indicates that this issue might related
> > > to Ghostscript rather than Lilypond.
> > It does not. The message is
> >
> > warning: g_spawn_sync failed (0): gs: Failed to close file descriptor for
> > child process (Operation not permitted)
> > fatal error: cannot rename `document-tmp-2786319.384671.pdf' to
> > `document.pdf'
> >
> > The warning (g_spawn_sync blabla) is emitted by LilyPond
> > due to an error in GhostScript. The second line, the fatal
> > error, is also emitted by LilyPond and has absolutely nothing
> > to do with GhostScript.
> If Ghostscript is supposed to create `document-tmp-2786319.384671.pdf`
> and fails, then Lilypond's `rename()` will naturally also fail, because
> `document-tmp-2786319.384671.pdf` does not exist. Therefore, even though
> the final error message originates from Lilypond, the primary source of
> the issue would be Ghostscript.



Ah yes, that's right judging from the --verbose log output I got
at some point:

Invoking `gs -dNODISPLAY -dNOSAFER -dNOPAUSE -dBATCH 
-dAutoRotatePages=/None -dPrinted=false /tmp/lilypond-tmp-2672114'...
warning: g_spawn_sync failed (0): gs: Failed to close file descriptor 
for child process (Operation not permitted)

The invocation of GhostScript normally starts like this:

GPL Ghostscript 9.56.1 (2022-04-04)
Copyright (C) 2022 Artifex Software, Inc.  All rights reserved.
This software is supplied under the GNU AGPLv3 and comes with NO WARRANTY:
see the file COPYING for details.


So it sounds like GhostScript doesn't even start.

In any case, the GitHub people gave an answer, which is to use

  --security-opt seccomp=unconfined

for the Docker container, in order to disable sandboxing.

https://github.com/orgs/community/discussions/38510#discussioncomment-4081353

Therefore it is likely that the seccomp sandbox somehow disables
something with subprocesses or the exact way LilyPond does them for
GhostScript.



> It would be neat if Lilypond could check the exit code of Ghostscript
> and fail earlier, but that seems difficult to do if you are sticking to
> ANSI C with its `system()` call that is quite oblivious to exit codes
> and the like.


As you can see from the error message, this call is using g_spawn_sync
from GLib. Emitting a warning and not an error looks intentional to me.
LilyPond tries to follow through in error conditions as much as it can.

Best,
Jean

OpenPGP_signature
Description: OpenPGP digital signature