https://bugzilla.novell.com/show_bug.cgi?id=828751 https://bugzilla.novell.com/show_bug.cgi?id=828751#c1 --- Comment #1 from Dmitriy Perlow 2013-07-09 20:53:00 UTC --- Some relevant info and considerations could be found in this bugreports: https://bugzilla.redhat.com/show_bug.cgi?id=977528 https://bugs.gentoo.org/show_bug.cgi?id=476234 -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=828751 https://bugzilla.novell.com/show_bug.cgi?id=828751#c3 Johannes Meixner changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |jsmeix@suse.com --- Comment #3 from Johannes Meixner 2013-07-11 10:00:39 CEST --- I am really not at all an expert regarding lcms2 therefore I can only guess here:

From what there is in your attachment#547477

(gdb) run Starting program: /usr/bin/leechcraft [Thread debugging using libthread_db enabled] Using host libthread_db library "/lib64/libthread_db.so.1". [New Thread 0x7fffef96d700 (LWP 27865)] [New Thread 0x7fffe4f10700 (LWP 27902)] undefined -21 undefined -21 [New Thread 0x7fffc563e700 (LWP 27970)] Program received signal SIGSEGV, Segmentation fault. gs_lcms2_malloc (id=0x0, size=3744) at base/gsicc_lcms2.c:48 --------------------------------------------------------------------------- my guess is the following: It seems /usr/bin/leechcraft starts some threads but it seems LCMS2 needs multi-threading fixes according to https://bugs.launchpad.net/ubuntu/+source/lcms2/+bug/1126427 "LCMS2 needs multi-threading fixes to work with the new Ghostscript 9.07" -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=828751 https://bugzilla.novell.com/show_bug.cgi?id=828751#c5 --- Comment #5 from Johannes Meixner 2013-07-11 11:02:14 CEST --- What puzzles me is that neither /usr/bin/gs nor /usr/lib/libgs.so.9.07 is linked with liblcms2 in openSUSE. At least not on my openSUSE 12.3 with the newest Ghostscript 9.07 from the OBS project "Printing" installed neither "ldd /usr/bin/gs | grep -i cms" nor "ldd /usr/lib/libgs.so.9.07 | grep -i cms" shows anything. Dmitriy Perlow, is this perhaps different on your openSUSE Factory system? FYI: It is correct that Ghostscript in openSUSE is not linked with liblcms2 because I build Ghostscript in full compliance with upstream, see bnc#735824 in particular see https://bugzilla.novell.com/show_bug.cgi?id=735824#c36 This means /usr/bin/gs and /usr/lib/libgs.so.9.07 are built by using Ghostscript's own included lcms2 sources that were specifically changed according to what Ghostscript upstream needs. If openSUSE would build Ghostscript by using the system's liblcms2, then Ghostscript's cms may no longer work at all or it would at least no longer work in compliance with what Ghostscript upstream needs. In contrast e.g. Ubuntu builds Ghostscript by using the system's liblcms2 and therefore their bug report https://bugs.launchpad.net/ubuntu/+source/lcms2/+bug/1126427 makes sense for me. Curently I do not understand what actually goes wrong here. How could "Downgrading to lcms2 ver. 2.4 fixes the issue" when Ghostscript is not linked with the system's liblcms2? -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=828751 https://bugzilla.novell.com/show_bug.cgi?id=828751#c6 --- Comment #6 from Dmitriy Perlow 2013-07-11 09:43:23 UTC --- Seemingly this patch is for lcms 2.4 (but this version doesn't cause any segfaults), and moreover it's a patch for Debian's patches, thus neither me (an openSUSE user) nor LC developer (a Gentoo user) could easily apply it... I cann't look directly at the system because I am at work, heh. But as I see neither Factory's Ghostscript not mine (oS 12.2) requires any lcms libs: https://build.opensuse.org/package/binary/openSUSE:Factory/ghostscript?arch=x86_64&filename=ghostscript-x11-9.07-2.3.x86_64.rpm&repository=standard https://build.opensuse.org/package/binary/openSUSE:Factory/ghostscript?arch=x86_64&filename=ghostscript-9.07-2.3.x86_64.rpm&repository=standard https://build.opensuse.org/package/binary/home:DarkSS:stuff/ghostscript?arch=x86_64&filename=ghostscript-9.07-41.1.x86_64.rpm&repository=openSUSE_12.2 https://build.opensuse.org/package/binary/home:DarkSS:stuff/ghostscript?arch=x86_64&filename=ghostscript-x11-9.07-41.1.x86_64.rpm&repository=openSUSE_12.2 Though the LC dev says on his system ghostscript is linked to lcms: % ldd /usr/lib/libgs.so.9.06 | grep lcms liblcms2.so.2 => /usr/lib64/liblcms2.so.2 (0x00007f1d379c5000) And we get the same segfault. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=828751 https://bugzilla.novell.com/show_bug.cgi?id=828751#c8 Johannes Meixner changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEW |NEEDINFO InfoProvider| |dap@open.by --- Comment #8 from Johannes Meixner 2013-07-12 10:47:03 CEST --- In the current Ghostscript sources at http://git.ghostscript.com/ghostpdl.git there have been some more changes in Ghostscript's own included lcms2 sources. I built Ghostscript with a lcms2-upgrade-to-current-git.diff that upgrades lcms2 in Ghostscript 9.07 to the one in the current upstream Ghostscript Git repository. I built the "ghostscript" package in the openSUSE build service project "home:jsmeix:branches:Printing" for various openSUSE versions and you can download the RPMs it from http://download.opensuse.org/repositories/home:/jsmeix:/branches:/Printing/ Dmitriy Perlow, coud you install Ghostscript 9.07 from there (use the RPMs from the sub-directory that matches exactly your system) and try out if this works for you? -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=828751 https://bugzilla.novell.com/show_bug.cgi?id=828751#c10 --- Comment #10 from Dmitriy Perlow 2013-07-13 10:10:51 UTC --- Yes, as I've already mentioned there are no problems with lcms2 <= 2.4. I installed lcms2 ver. 2.5 and your ghostscript packages (ghostscript-x11-9.07-42.2.x86_64.rpm & ghostscript-9.07-42.2.x86_64.rpm) and got the same segmentation fault as with Factory's default ones. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=828751 https://bugzilla.novell.com/show_bug.cgi?id=828751#c12 --- Comment #12 from Dmitriy Perlow 2013-07-13 13:57:32 UTC --- The repo: https://build.opensuse.org/project/monitor/home:DarkSS:stuff :) -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

From my non-expert point of view the fact that downgrading

https://bugzilla.novell.com/show_bug.cgi?id=828751 https://bugzilla.novell.com/show_bug.cgi?id=828751#c14 --- Comment #14 from Johannes Meixner 2013-07-16 10:47:11 CEST --- Regarding comment#10 "there are no problems with lcms2 <= 2.4": I am really not an expert in linking but I think that using lcms2 2.4 only avoids the segfault but this does not mean that all runs correctly. I think when in one same address space (i.e. in one process) the same set of functions gets included two times (once via compiled-in sources and once via a linked library) and those duplicate set of functions is used simultaneously by several threads, but those functions are not thread-safe, then any kind of misbehaviour could happen. liblcms2 to lcms2 2.4 avoids the segfault indicates that somehow the lcms2 functions in Ghostscript and in liblcms2.so get used in a mixed up way. If Ghostscript would use only its built-in lcms2 functions and the other thread would use only the lcms2 functions from liblcms2.so so that both set of functions are used strictly separated, then there should be no segfault regardless what lcms2 version is used in Ghostscript and by the other thread.

From my non-expert point of view the actual bug lies deeper.

I think it is perfectly valid when one free software project incorporates source code from another free software project. For example see Gnulib http://www.gnu.org/software/gnulib/ ----------------------------------------------------------------- Its components are intended to be shared at the source level, rather than being a library that gets built, installed, and linked against. ----------------------------------------------------------------- Therefore I think it is o.k. when Ghostscript incorporates lcms2 at the source level. Furthermore I think it is o.k. when a library is not thread-safe provided it is sufficiently documented so that developers know that they cannot "just use" it in multi-threaded programs. Finally I think it belongs to the developers of a program to ensure that their program runs correctly. In the end this means (from my non-expert point of view): - Ghostscript is o.k. to incorporate lcms2 at the source level. - liblcms2.so is o.k. to be not thread-safe. - Any multi-threaded program that likes to run Ghostscript and liblcms2.so in several threads, must ensure that it runs correctly. Therefore I think that such multi-threaded programs like Evince and LeechCraft Monocle would need to be fixed. For example a multi-threaded program may no longer run Ghostscript and liblcms2.so in several threads simultaneously but only one after the other to get the different lcms2 usages strictly separated. Alternatively a multi-threaded program could get the different lcms2 usages strictly separated by running Ghostscript and liblcms2.so in separated processes via fork() or it may use the clone() system call in a more appropriate way to get the threads better separated. But again: I am really neither an expert in lcms2 nor in linking nor in multi-threading environments so that all I can do here is basically only guesswork. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=828751 https://bugzilla.novell.com/show_bug.cgi?id=828751#c16 Johannes Meixner changed: What |Removed |Added ---------------------------------------------------------------------------- Priority|P5 - None |P3 - Medium Status|NEW |ASSIGNED CC| |mrdocs@opensuse.org AssignedTo|mrdocs@opensuse.org |jsmeix@suse.com --- Comment #16 from Johannes Meixner 2013-07-18 15:06:20 CEST --- Meanwhile I think the root cause is in Ghostscript, see the "Can Ghostscript be run in a thread?" mail thread on gs-devel@ghostscript.com on http://ghostscript.com/pipermail/gs-devel/2013-July/thread.html Currently I think the root cause is: libgs.so provides too many dynamic symbols see in particular http://ghostscript.com/pipermail/gs-devel/2013-July/009475.html What I still do not understand is how could "Downgrading to lcms2 ver. 2.4 fixes the issue" when Ghostscript is not linked with the system's liblcms2? When downgrading the system's liblcms2 to version 2.4 avoids a segfault in gs_lcms2_malloc it means as far as I imagine that somehow Ghostscript does not use its own built-in lcms2 functions because if Ghostscript would use its own built-in lcms2 functions the version of the system's liblcms2 should not make a difference. Therefore it seems as far as I imagine that Ghostscript somehow uses the system's liblcms2 functions regardless that Ghostscript is not linked with the system's liblcms2. Strange... Regardless of what exactly goes wrong here I think if leechcraft would fork(2) a separated process that runs Ghostscript, then it would "just work". In particular it does not work if leechcraft spawns several threads that run (directly or indirectly) Ghostscript simultaneously, see http://ghostscript.com/pipermail/gs-devel/2013-July/009474.html -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=828751 https://bugzilla.novell.com/show_bug.cgi?id=828751#c18 --- Comment #18 from Dmitriy Perlow 2013-07-27 14:17:20 UTC --- Here I am again. Unfortunately, seems like there is a little misunderstanding here. LeechCraft doesn't spawn any threads at all neither for poppler nor for ghostscript nor for anything that could call them. Thus the fork(2) solution isn't really a solution since there is no such problem. Everything just happens in the main GUI thread. -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.

https://bugzilla.novell.com/show_bug.cgi?id=828751 https://bugzilla.novell.com/show_bug.cgi?id=828751#c20 --- Comment #20 from Dmitriy Perlow 2013-07-30 11:21:18 UTC ---

Usually Ghostscript would initiate its built-in liblcms2 with its gs_lcms2_malloc plugin to tell it to use certain Ghostscript specific callbacks for memory allocation.

Currently I do not understand how the memory allocator plugin from Ghostscript was set in the liblcms2 which is called by Poppler.

Sure, but if Ghostscript hasn't changed naming of lcms2 symbols, then built-in lcms2 symbols would clash with system-wide lcms' ones, thus system-wide lcms2 would also use that same memory allocator.

could you check if after downgrading to lcms2 ver. 2.4 Poppler still calls gs_lcms2_malloc?

Should I just search for "malloc" in gdb output with lcms2.4? If no explain please what should I do? -- Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are on the CC list for the bug.