Bug#987652: surf does not start
Hello Herr Sprickerhof,
First of all thank you for your quick answer. Your last remark
gave me the hint I needed :
It turns out I had a stale libsurf-webext.so in the directory
/usr/local/lib/surf, removing it solved the problem. This is due
to the fact that upstream changed the name of the library from
libsurf-webext.c to webext-surf.c between debian/2.0+git20181009
and debian/2.0+git20201107, which means that a not careful enough
direct use of the Makefile (make install once in each branch, not
separated with a make uninstall) would mean that two versions of
the shared object library (the stale one libsurf-webext.so, and
the new one webext-surf.so) would cohabit in the same directory
/usr/local/lib/surf.
These two shared object libraries would have competing versions of
each symbol, and more specifically two versions of the culprit
symbol webkit_web_initialize_with_user_data() :
- one (libsurf-webext.so) that would have g_variant_get(gv,
"(ii)", , )
- the other (webext-surf.so) that would have g_variant_get(gv,
"i", )
Since in the new version (debian/2.0+git20201107), gv would be
created like this :
gv = g_variant_new("i", spair[1])
surf would work correctly only if the dynamic linking would use
webext-surf.so, and crash if the dynamic linking had chosen
libsurf-webext.so. I have no idea how the choice is made
(alphabetic order of each shared object file maybe ??), but it
seems the dynamic linking systematically chose the stale file
before we removed it.
So the problem was entirely my fault, and a more careful direct
use of the Makefile solved the problem. So I am sorry for having
wasted your time.
However, the stuff I do on my own patched version of surf (that
goes in /usr/local/bin, with the shared object going in
/usr/local/lib/surf/) should have no influence on the Debian
version of surf living in /usr/bin : even when I launched
/usr/bin/surf, the stale library
/usr/local/lib/surf/libsurf-webext.so would be used over
/usr/lib/surf/webext-surf.so ! So a user-compiled shared object
(in /usr/local) would take precedence over a Debian-compiled
version (in /usr), even when the binary is itself in /usr ?
This, in my opinion, is still a bug (albeit a lot less severe and
a lot more subtle, and arguably a different one). What is the next
course of action ?
The symbol webkit_web_initialize_with_user_data() is not called
from surf, but from webkit. So the bug lies not with the surf
package, but probably with the libwebkit2gtk-4.0-37.
So the original bug (surf not starting) can be closed (I have no
idea how to do that). I'll let you tell me whether you agree with
my opinion that webkit should try to resolve
webkit_web_initialize_with_user_data() from /usr/lib/surf and not
from /usr/local/lib/surf when the binary is /usr/bin/surf, and
whether this warrants another bug report (and in which package ) ?
Thank you again for your time reading this long message,
Best regards,
Aymeric Agon-Rambosson
Jochen Sprickerhof writes:
* Aymeric Agon-Rambosson [2021-04-27
03:18]:
$ /usr/bin/surf
output :
(WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.728: the
GVariant format string '(ii)' has a type of '(ii)' but the given
value has a type of 'i'
(WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.728:
g_variant_get: assertion 'valid_format_string (format_string,
TRUE, value)' failed
(WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.729: the
GVariant format string '(ii)' has a type of '(ii)' but the given
value has a type of 'i'
(WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.729:
g_variant_get: assertion 'valid_format_string (format_string,
TRUE, value)' failed
web process terminated: crashed
Also, can you check if you have a custom webext-surf.so in your
~/.surf or elsewhere in your ld path?
Bug#987652: surf does not start
* Aymeric Agon-Rambosson [2021-04-27 03:18]: $ /usr/bin/surf output : (WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.728: the GVariant format string '(ii)' has a type of '(ii)' but the given value has a type of 'i' (WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.728: g_variant_get: assertion 'valid_format_string (format_string, TRUE, value)' failed (WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.729: the GVariant format string '(ii)' has a type of '(ii)' but the given value has a type of 'i' (WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.729: g_variant_get: assertion 'valid_format_string (format_string, TRUE, value)' failed web process terminated: crashed Also, can you check if you have a custom webext-surf.so in your ~/.surf or elsewhere in your ld path? signature.asc Description: PGP signature
Bug#987652: surf does not start
Control: severity -1 normal Control: tags -1 = moreinfo unreproducible Hi, I was not able to reproduce this on testing nor unstable. * Aymeric Agon-Rambosson [2021-04-27 03:18]: $ /usr/bin/surf output : (WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.728: the GVariant format string '(ii)' has a type of '(ii)' but the given value has a type of 'i' (WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.728: g_variant_get: assertion 'valid_format_string (format_string, TRUE, value)' failed (WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.729: the GVariant format string '(ii)' has a type of '(ii)' but the given value has a type of 'i' (WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.729: g_variant_get: assertion 'valid_format_string (format_string, TRUE, value)' failed web process terminated: crashed Can you attach a gdb and send a backtrace? The problem can be traced back to this specific upstream commit : e92fd1aa5f38c399f8fc5d263026fbd9d34ddfbb Which can be found at https://git.suckless.org/surf/commit/e92fd1aa5f38c399f8fc5d263026fbd9d34ddfbb.html That is there since over a year and I'm not aware of any issues with it, probably a red herring. How did you test that? -- Configuration Files: /etc/apparmor.d/usr.bin.surf changed: Probably unrelated but maybe a good idea to reset this to the default. Cheers Jochen signature.asc Description: PGP signature
Bug#987652: surf does not start
Package: surf
Version: 2.0+git20201107-2
Severity: grave
Tags: upstream
Justification: renders package unusable
Dear Maintainer,
surf does not start anymore since version 2.0+git20201107.
Expected behaviour : surf should start.
Steps to reproduce :
$ /usr/bin/surf
output :
(WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.728: the GVariant format
string '(ii)' has a type of '(ii)' but the given value has a type of 'i'
(WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.728: g_variant_get:
assertion 'valid_format_string (format_string, TRUE, value)' failed
(WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.729: the GVariant format
string '(ii)' has a type of '(ii)' but the given value has a type of 'i'
(WebKitWebProcess:94294): GLib-CRITICAL **: 02:49:52.729: g_variant_get:
assertion 'valid_format_string (format_string, TRUE, value)' failed
web process terminated: crashed
And surf crashes.
The problem can be traced back to this specific upstream commit :
e92fd1aa5f38c399f8fc5d263026fbd9d34ddfbb
Which can be found at
https://git.suckless.org/surf/commit/e92fd1aa5f38c399f8fc5d263026fbd9d34ddfbb.html
One possible fix/workaround is the following :
diff --git a/surf.c b/surf.c
index ac832ff..e84a538 100644
--- a/surf.c
+++ b/surf.c
@@ -1269,7 +1269,7 @@ initwebextensions(WebKitWebContext *wc, Client *c)
if (spair[1] < 0)
return;
- gv = g_variant_new("i", spair[1]);
+ gv = g_variant_new("(ii)", spair[1]);
webkit_web_context_set_web_extensions_initialization_user_data(wc, gv);
webkit_web_context_set_web_extensions_directory(wc, WEBEXTDIR);
diff --git a/webext-surf.c b/webext-surf.c
index d087219..da16ddf 100644
--- a/webext-surf.c
+++ b/webext-surf.c
@@ -95,7 +95,7 @@
webkit_web_extension_initialize_with_user_data(WebKitWebExtension *e,
webext = e;
- g_variant_get(gv, "i", );
+ g_variant_get(gv, "(ii)", );
gchansock = g_io_channel_unix_new(sock);
g_io_channel_set_encoding(gchansock, NULL, NULL);
But this workaround seems wrong when we look at the semantic of g_variant_new()
in
gvariant.c :
- sock and spair[1] are ints, and should work with "i".
- Similarly, "(ii)" should mean two extra arguments after the format, but only
works.
And I don't know whether this fix breaks surf in another way.
One other way would be to revert to 2.0+git20181009-4, or some other version
inbetween.
Thank you in advance for your time,
Aymeric Agon-Rambosson
-- System Information:
Debian Release: bullseye/sid
APT prefers testing-security
APT policy: (500, 'testing-security'), (500, 'testing'), (500, 'stable')
Architecture: amd64 (x86_64)
Foreign Architectures: i386
Kernel: Linux 5.10.0-6-amd64 (SMP w/16 CPU threads)
Locale: LANG=fr_FR.UTF-8, LC_CTYPE=fr_FR.UTF-8 (charmap=UTF-8), LANGUAGE not set
Shell: /bin/sh linked to /usr/bin/dash
Init: systemd (via /run/systemd/system)
LSM: AppArmor: enabled
Versions of packages surf depends on:
ii libc62.31-11
ii libgcr-base-3-1 3.38.1-2
ii libgcr-ui-3-13.38.1-2
ii libglib2.0-0 2.66.8-1
ii libgtk-3-0 3.24.24-3
ii libjavascriptcoregtk-4.0-18 2.30.6-1
ii libwebkit2gtk-4.0-37 2.30.6-1
ii libx11-6 2:1.7.0-2
Versions of packages surf recommends:
ii curl 7.74.0-1.2
ii stterm [x-terminal-emulator] 0.8.4-1
ii suckless-tools46-1
ii x11-utils 7.7+5
Versions of packages surf suggests:
ii apparmor 2.13.6-10
-- Configuration Files:
/etc/apparmor.d/usr.bin.surf changed:
/usr/bin/surf flags=(complain) {
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
owner @{HOME}/.surf/ w,
owner @{HOME}/.surf/** rwkl,
owner @{HOME}/.cache/ rw,
@{PROC}/@{pid}/cmdline r,
@{PROC}/@{pid}/fd/ r,
@{PROC}/@{pid}/smaps r,
/dev/ r,
/sys/devices/pci[0-9]*/** r,
/sys/devices/platform/soc/soc:gpu/* r,
/usr/share/glib-2.0/schemas/gschemas.compiled r,
/usr/share/doc/** r,
# WebKit
/usr/lib/@{multiarch}/webkit2gtk-4.0/WebKit*Process ix,
/{dev,run}/shm/WK2SharedMemory.* rw,
/var/tmp/WebKit-Media-* rw,
/usr/share/publicsuffix/public_suffix_list.{dat,dafsa} r,
owner @{HOME}/.local/share/webkitgtk/ w,
owner @{HOME}/.local/share/webkitgtk/** rw,
owner @{HOME}/.cache/webkitgtk/ w,
owner @{HOME}/.cache/webkitgtk/** rwk,
# fontconfig
/usr/share/fontconfig/conf.avail/ r,
# dconf
owner @{HOME}/.cache/dconf/user rw,
owner /run/user/*/dconf/user rw,
/usr/bin/surf ix,
/{usr/,}bin/dash ix,
/{usr/,}bin/sed ix,
/usr/bin/dmenu ix,
/usr/bin/printf ix,
/usr/bin/xargs ix,
/usr/bin/xprop ix,
# for downloading files
/dev/ptmx rw,
/dev/pts/* rw,
/usr/bin/st ix,
# unconfined because it is called in (and downloading to) the cwd
/usr/bin/curl Ux,
#
