webp-pixbuf-loader 0.2.0 regression test failure

Question

webp-pixbuf-loader 0.2.0 regression test failure

Closed this issue 2 years ago · 15 comments

$ meson test
ninja: Entering directory '/home/gv74/work/rpm/BUILD/webp-pixbuf-loader-0.2.0/x86_64-redhat-linux-gnu' ninja: no work to do.
1/5 get file info OK 0.01s
2/5 load 1x1 image FAIL 0.10s killed by signal 11 SIGSEGV
>>> MALLOC_PERTURB_=189 TEST_FILE=/home/gv74/work/rpm/BUILD/webp-pixbuf-loader-0.2.0/tests/t1.webp GDK_PIXBUF_MODULE_FILE=/home/gv74/work/rpm/BUILD/webp-pixbuf-loader-0.2.0/x86_64-redhat-linux-gnu/tests/loaders.cache /home/gv74/work/rpm/BUILD/webp-pixbuf-loader-0.2.0/x86_64-redhat-linux-gnu/tests/t1

3/5 load animation FAIL 0.11s killed by signal 11 SIGSEGV
>>> GDK_PIXBUF_MODULE_FILE=/home/gv74/work/rpm/BUILD/webp-pixbuf-loader-0.2.0/x86_64-redhat-linux-gnu/tests/loaders.cache MALLOC_PERTURB_=116 TEST_FILE=/home/gv74/work/rpm/BUILD/webp-pixbuf-loader-0.2.0/tests/t3.webp /home/gv74/work/rpm/BUILD/webp-pixbuf-loader-0.2.0/x86_64-redhat-linux-gnu/tests/t3

4/5 load 200x200 image FAIL 0.12s killed by signal 11 SIGSEGV
>>> TEST_FILE=/home/gv74/work/rpm/BUILD/webp-pixbuf-loader-0.2.0/tests/t2.webp MALLOC_PERTURB_=203 GDK_PIXBUF_MODULE_FILE=/home/gv74/work/rpm/BUILD/webp-pixbuf-loader-0.2.0/x86_64-redhat-linux-gnu/tests/loaders.cache /home/gv74/work/rpm/BUILD/webp-pixbuf-loader-0.2.0/x86_64-redhat-linux-gnu/tests/t2

5/5 save data FAIL 0.13s killed by signal 11 SIGSEGV
>>> TEST_FILE=/home/gv74/work/rpm/BUILD/webp-pixbuf-loader-0.2.0/tests/t2.webp MALLOC_PERTURB_=34 GDK_PIXBUF_MODULE_FILE=/home/gv74/work/rpm/BUILD/webp-pixbuf-loader-0.2.0/x86_64-redhat-linux-gnu/tests/loaders.cache /home/gv74/work/rpm/BUILD/webp-pixbuf-loader-0.2.0/x86_64-redhat-linux-gnu/tests/t_save

Ok: 1
Expected Fail: 0
Fail: 4
Unexpected Pass: 0
Skipped: 0
Timeout: 0

Full log written to /home/gv74/work/rpm/BUILD/webp-pixbuf-loader-0.2.0/x86_64-redhat-linux-gnu/meson-logs/testlog.txt

Relevant /var/log/messages entries:
[ 257.777098] t2[1846]: segfault at 0 ip 0000000000000000 sp 00007ffedb79abd8 error 14 in t2[5558e836f000+1000]
[ 257.777144] Code: Unable to access opcode bytes at RIP 0xffffffffffffffd6.
[ 257.777484] t1[1845]: segfault at 0 ip 0000000000000000 sp 00007fff40543788 error 14 in t1[558e4a902000+1000]
[ 257.777508] Code: Unable to access opcode bytes at RIP 0xffffffffffffffd6.
[ 257.777622] t3[1847]: segfault at 0 ip 0000000000000000 sp 00007ffd3fd49188 error 14 in t3[558a9db33000+2000]
[ 257.777649] Code: Unable to access opcode bytes at RIP 0xffffffffffffffd6.
[ 257.790624] t_save[1852]: segfault at 0 ip 0000000000000000 sp 00007ffe0d84ed08 error 14 in t_save[5608e125f000+2000]
[ 257.790669] Code: Unable to access opcode bytes at RIP 0xffffffffffffffd6.

Answer 1 · 2023-02-24T13:14:47.000Z

FWIW I can't reproduce this on Fedora Rawhide.. perhaps share some more details of your setup

Answer 2 · 2023-02-24T13:22:20.000Z

It seems that needs ninja before running tests. That is to say test subcommand compiles test programs only, but not the loader...

Answer 3 · 2023-02-24T13:55:42.000Z

OS: Rocky Linux 8.7 (aka RHEL 8.7)
Build command:
rpmbuild -ba webp-pixbuf-loader.spec 2>&1 | tee build.log
build.log

Answer 4 · 2023-02-24T14:32:37.000Z

OS: Rocky Linux 8.7 (aka RHEL 8.7) Build command: rpmbuild -ba webp-pixbuf-loader.spec 2>&1 | tee build.log build.log

We need the testlog too

Answer 5 · 2023-02-24T14:33:24.000Z

@mikelolasagasti are you having issues in EL8 too?

Answer 6 · 2023-02-24T14:41:50.000Z

We need the testlog too
testlog.txt

Answer 7 · 2023-02-24T17:08:40.000Z

This is really odd, gdk-pixbuf passes size_func and update_func as null... I am inclined to say this is a gdk-pixbuf bug in RHEL8.

Breakpoint 1, begin_load (size_func=0x0, prepare_func=0x7f0f71e9f2c0 <prepared_notify>, update_func=0x0, user_data=0x7ffcbbed2048, error=0x0) at ../io-webp.c:23
23	  WebPContext *context  = g_new0 (WebPContext, 1);
(gdb)

Answer 8 · 2023-02-24T17:18:40.000Z

This smells like the compiler over optimized things by removing those empty functions and their address point to NULL now.

This is a gdk pixbuf bug in RHEL 8.

static void
noop_size_notify (gint     *width,
		  gint     *height,
		  gpointer  data)
{
}

static void
prepared_notify (GdkPixbuf *pixbuf, 
                 GdkPixbufAnimation *anim, 
                 gpointer user_data)
{
        if (pixbuf != NULL)
                g_object_ref (pixbuf);
        *((GdkPixbuf **)user_data) = pixbuf;
}

static void
noop_updated_notify (GdkPixbuf *pixbuf,
		     int        x,
		     int        y,
		     int        width,
		     int        height,
		     gpointer   user_data)
{
}

static GdkPixbuf *
generic_load_incrementally (GdkPixbufModule *module, FILE *f, GError **error)
{
        GdkPixbuf *pixbuf = NULL;
	gpointer context;

	context = module->begin_load (noop_size_notify, prepared_notify, noop_updated_notify, &pixbuf, error);

Answer 9 · 2023-02-24T17:35:42.000Z

I have submitted a bug to RHEL8 about this:

https://bugzilla.redhat.com/show_bug.cgi?id=2173198

Answer 10 · 2023-03-06T14:06:00.000Z

I have submitted a bug to RHEL8 about this:

https://bugzilla.redhat.com/show_bug.cgi?id=2173198

This is pointless. The generic_load_incrementally() from gdk-pixbuf2 2.36.12 specifically call module->begin_load() as this:
context = module->begin_load (NULL, prepared_notify, NULL, &pixbuf, error);
There are no noop_size_notify and noop_updated_notify functions.

The attached patch declare them in webp-pixbuf-loader and now everything is working for me.

patch.zip

Answer 11 · 2023-03-06T14:21:01.000Z

Can you try 0.2.2? I check if those functions are NULL before calling them now.

Answer 12 · 2023-03-06T14:39:18.000Z

The patch is for 0.2.2.

Answer 13 · 2023-03-06T14:43:04.000Z

Can you reproduce the crash in 0.2.2 though?

Answer 14 · 2023-03-06T14:47:40.000Z

No. You are right. 0.2.2 works fine.
I created the patch for 0.2.0 and assumed that 0.2.2 fail too. This is not the case.
Thank you.

Answer 15 · 2023-03-06T14:51:15.000Z

Phew, I was going crazy wondering how this could happen.

Great, thanks!