6:24 a.m.
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
Summary: segfault in FTC_CMapCache_Lookup()
https://bugzilla.redhat.com/show_bug.cgi?id=513582
Summary: segfault in FTC_CMapCache_Lookup()
Product: Fedora
Version: 11
Platform: All
OS/Version: Linux
Status: NEW
Severity: medium
Priority: low
Component: freetype
AssignedTo: besfahbo(a)redhat.com
ReportedBy: fabrice(a)bellet.info
QAContact: extras-qa(a)fedoraproject.org
CC: fabrice(a)bellet.info, besfahbo(a)redhat.com,
kevin(a)tigcc.ticalc.org,
fedora-fonts-bugs-list(a)redhat.com
Classification: Fedora
I have a VTK application that segfaults in libfreetype, when text in rendered.
I tried to look for details with gdb, and the problem seems to be in the
inlined code of FTC_CACHE_LOOKUP_CMP, called just before ftccmap.c:382. The
crash occurs when optimization is enabled. In this case, node is stored in a
register, and for an unknown reason, node is reset to zero before being
dereferenced.
The code following the _Ok label, in the macro definition of
FTC_CACHE_LOOKUP_CMP
_pnode = (FTC_Node*)(void*)&(node);
*_pnode = _node;
seems equivalent to this simpler version :
node = _node;
And the app no longer crashes with this fix.
I'll attach the simple VTK app that generates the crash for me. Compile it,
with vtk-devel installed, with :
gcc -c AppliPlanes.cxx -I/usr/include/vtk
gcc -o AppliPlanes AppliPlanes.o -lvtkRendering
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
6:26 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #1 from Fabrice Bellet <fabrice(a)bellet.info> 2009-07-24 07:26:07 EDT
---
Created an attachment (id=355001)
--> (https://bugzilla.redhat.com/attachment.cgi?id=355001)
sample vtk app trigerring the crash
AppliPlanes.cxx
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
6:27 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #2 from Fabrice Bellet <fabrice(a)bellet.info> 2009-07-24 07:27:05 EDT
---
Created an attachment (id=355002)
--> (https://bugzilla.redhat.com/attachment.cgi?id=355002)
sample vtk app trigerring the crash (header)
AppliPlanes.h
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
6:29 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #3 from Fabrice Bellet <fabrice(a)bellet.info> 2009-07-24 07:29:03 EDT
---
Created an attachment (id=355003)
--> (https://bugzilla.redhat.com/attachment.cgi?id=355003)
workaround patch
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
8:29 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
Kevin Kofler <kevin(a)tigcc.ticalc.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Flag| |needinfo?(besfahbo(a)redhat.c
| |om)
--- Comment #4 from Kevin Kofler <kevin(a)tigcc.ticalc.org> 2009-07-24 09:29:08 EDT
---
Yeah, a classic aliasing violation.
Behdad: ping?
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
3:06 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
Behdad Esfahbod <besfahbo(a)redhat.com> changed:
What |Removed |Added
----------------------------------------------------------------------------
Flag|needinfo?(besfahbo(a)redhat.c |
|om) |
--- Comment #5 from Behdad Esfahbod <besfahbo(a)redhat.com> 2009-07-25 16:06:42 EDT
---
Thanks. Building.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
3:24 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #6 from Fedora Update System <updates(a)fedoraproject.org> 2009-07-25
16:24:14 EDT ---
freetype-2.3.9-5.fc11 has been submitted as an update for Fedora 11.
http://admin.fedoraproject.org/updates/freetype-2.3.9-5.fc11
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
4:59 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #7 from Kevin Kofler <kevin(a)tigcc.ticalc.org> 2009-07-25 17:59:05 EDT
---
What about F12? And F10?
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
10:31 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #8 from Behdad Esfahbod <besfahbo(a)redhat.com> 2009-07-25 23:31:02 EDT
---
I reported the issue upstream, so I'm hoping to get a new release with the fix
for F12. Feel free to build for F10.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
4:27 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #9 from Fedora Update System <updates(a)fedoraproject.org> 2009-07-27
17:27:27 EDT ---
freetype-2.3.9-5.fc11 has been pushed to the Fedora 11 stable repository. If
problems still persist, please make note of it in this bug report.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
4:27 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
Fedora Update System <updates(a)fedoraproject.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |CLOSED
Fixed In Version| |2.3.9-5.fc11
Resolution| |ERRATA
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
4:49 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
Werner Lemberg <wl(a)gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |wl(a)gnu.org
--- Comment #10 from Werner Lemberg <wl(a)gnu.org> 2009-07-29 05:49:04 EDT ---
Applied to FreeType git, thanks.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
11:05 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
Adam Williamson <awilliam(a)redhat.com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |awilliam(a)redhat.com
--- Comment #11 from Adam Williamson <awilliam(a)redhat.com> 2009-07-29 12:05:28 EDT
---
Bah, you mean you have to post bugs with _patches_ to get any interest around
here? :)
Werner lets me know that https://bugzilla.redhat.com/show_bug.cgi?id=506840 is
likely the same as this. I will test and confirm.
--
Fedora Bugzappers volunteer triage team
https://fedoraproject.org/wiki/BugZappers
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
11:54 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #12 from Kevin Kofler <kevin(a)tigcc.ticalc.org> 2009-07-29 12:54:55 EDT
---
*** Bug 506840 has been marked as a duplicate of this bug. ***
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
12:29 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #13 from Werner Lemberg <wl(a)gnu.org> 2009-07-29 13:29:50 EDT ---
Note that the fix proposed above makes compilation fail with C++. Any idea how
to correct that?
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
12:51 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #14 from Kevin Kofler <kevin(a)tigcc.ticalc.org> 2009-07-29 13:51:42 EDT
---
Is ftccache.h a public header? If it's not, why does it have to be buildable as
C++?
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
1:09 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #15 from Werner Lemberg <wl(a)gnu.org> 2009-07-29 14:09:38 EDT ---
Some platforms use the C++ compiler even for compiling C code. We always try
to make FreeType compile without warnings (and errors) for both C and C++.
Nelson Beebe, who is maintaining a real zoo of different Unix machines,
sometimes provides reports like this:
http://lists.gnu.org/archive/html/freetype-devel/2003-04/msg00054.html
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
1:40 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #16 from Adam Williamson <awilliam(a)redhat.com> 2009-07-29 14:40:42 EDT
---
confirming the fix/workaround patch in this report fixes my bug 506840. Thanks,
Fabrice.
--
Fedora Bugzappers volunteer triage team
https://fedoraproject.org/wiki/BugZappers
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
2:14 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #17 from Kevin Kofler <kevin(a)tigcc.ticalc.org> 2009-07-29 15:14:23 EDT
---
Hmmm, the problem is that even with that aliasing issue fixed, you're still
casting pointers of different types, which is both a likely violation of
aliasing rules and invalid C++.
You have a structure like this:
/* the cmap cache node */
typedef struct FTC_CMapNodeRec_
{
FTC_NodeRec node;
FTC_FaceID face_id;
FT_UInt cmap_index;
FT_UInt32 first; /* first character in node */
FT_UInt16 indices[FTC_CMAP_INDICES_MAX]; /* array of glyph indices */
} FTC_CMapNodeRec, *FTC_CMapNode;
and then you're trying to use an FTC_CMapNode (an FTC_CMapNodeRec *) as an
FTC_Node (an FTC_NodeRec *). The GCC C frontend has special rules to allow this
type of "poor man's inheritance", in C++ you can't do that without
actual
inheritance.
You should be using &(node->node) rather than just node where you're passing
your FTC_CMapNode to functions which expect an FTC_Node.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
2:15 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #18 from Kevin Kofler <kevin(a)tigcc.ticalc.org> 2009-07-29 15:15:42 EDT
---
Hmmm, &(node->node) isn't going to work in that context because you're
using it
as an lvalue.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
2:21 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #19 from Kevin Kofler <kevin(a)tigcc.ticalc.org> 2009-07-29 15:21:42 EDT
---
I think the only fix which is likely to work with C++ is to edit
FTC_CMapCache_Lookup in ftccmap.c to add a variable:
FTC_Node ftcnode;
and to change the FTC_CACHE_LOOKUP_CMP call to:
FTC_CACHE_LOOKUP_CMP( cache, ftc_cmap_node_compare, hash, &query,
ftcnode, error );
node = (FTC_CMapNode) ftcnode;
If FTC_CACHE_LOOKUP_CMP is also used in other places, they'll need the same
type of fix.
This is still a somehow unsafe cast, mind you, but at least this one should
compile and not trigger an aliasing violation in GCC.
(I still think the idea of writing code which compiles as both C and C++ is
insane though. ;-) )
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
11:38 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #20 from Adam Williamson <awilliam(a)redhat.com> 2009-07-30 12:38:54 EDT
---
Would it be too much to ask if this could be patched in Rawhide while we wait
for a new upstream release? It would help me get my navit package review done,
because that app fails to run entirely unless this bug is fixed, so my review
for it will be stuck in limbo until this bug is fixed. thanks!
--
Fedora Bugzappers volunteer triage team
https://fedoraproject.org/wiki/BugZappers
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
1:05 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #21 from Behdad Esfahbod <besfahbo(a)redhat.com> 2009-07-30 14:05:21 EDT
---
Building.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
5:43 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #22 from Adam Williamson <awilliam(a)redhat.com> 2009-07-30 18:43:47 EDT
---
Thanks a bunch, Behdad - I really appreciate it.
--
Fedora Bugzappers volunteer triage team
https://fedoraproject.org/wiki/BugZappers
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
12:39 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
mpsuzuki(a)hiroshima-u.ac.jp changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |mpsuzuki(a)hiroshima-u.ac.jp
--- Comment #23 from mpsuzuki(a)hiroshima-u.ac.jp 2009-09-25 01:39:16 EDT ---
As Kevin mentioned already, proposed fix has a side effect
that GCC finds a cast between incompatible pointers.
Although current git head of FreeType2 includes it,
I want to revert it.
In my investigation, the crashing behaviour was supposed to
be introduced by Richard Guenther's inline optimizer for
GCC-4.4 branch, committed on 2008-08-09.
For detail, see https://savannah.nongnu.org/bugs/index.php?27441
I propose another fix to pass the problem, aslike:
diff --git a/src/cache/ftccache.h b/src/cache/ftccache.h
index 2082bc4..5e932b7 100644
--- a/src/cache/ftccache.h
+++ b/src/cache/ftccache.h
@@ -206,7 +206,7 @@ FT_BEGIN_HEADER
\
\
error = 0; \
- node = NULL; \
+ /* node = NULL; */ \
_idx = _hash & _cache->mask; \
if ( _idx < _cache->p ) \
_idx = _hash & ( _cache->mask*2 + 1 ); \
@@ -246,7 +246,8 @@ FT_BEGIN_HEADER
error = FTC_Cache_NewNode( _cache, _hash, query, &_node ); \
\
_Ok: \
- node = _node; \
+ _pnode = (FTC_Node*)(void*)&(node); \
+ *_pnode = _node; \
FT_END_STMNT
#else /* !FTC_INLINE */
I want to revert the fix that current FC11 uses and apply
this patch. Anybody can test if this fix works well?
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
12:43 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #24 from Adam Williamson <awilliam(a)redhat.com> 2009-09-25 01:43:41 EDT
---
if you throw out a scratch build, I will check it.
--
Fedora Bugzappers volunteer triage team
https://fedoraproject.org/wiki/BugZappers
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
1:07 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #25 from mpsuzuki(a)hiroshima-u.ac.jp 2009-09-25 02:07:54 EDT ---
Created an attachment (id=362615)
--> (https://bugzilla.redhat.com/attachment.cgi?id=362615)
source rpm including the patch proposed by mpsuzuki
Here is the source rpm including the patch proposed in
https://bugzilla.redhat.com/show_bug.cgi?id=513582#c23
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
1:06 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #26 from Adam Williamson <awilliam(a)redhat.com> 2009-09-25 14:06:17 EDT
---
my incarnation of this problem (navit failing to run) does not happen with the
build from comment #25, so I have no problem with this change.
--
Fedora Bugzappers volunteer triage team
https://fedoraproject.org/wiki/BugZappers
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
1:34 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #27 from mpsuzuki(a)hiroshima-u.ac.jp 2009-09-25 14:34:00 EDT ---
Thank you very much for testing!
I will replace previous fix for ftccache.h by new one.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
9:21 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #28 from Behdad Esfahbod <besfahbo(a)redhat.com> 2009-09-25 22:21:01 EDT
---
As far as I understand, the proposed fix still violates C aliasing rules, even
if gcc currently is not smart enough for it. One fix I can imagine is to
introduce a union that has both pointer types as members, the assing to one,
read from the other.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
1:23 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #29 from Kevin Kofler <kevin(a)tigcc.ticalc.org> 2009-09-26 02:23:48 EDT
---
Behdad is right, mpsuzuki's "fix" is NOT correct, please DO NOT use this!
Every
"fix" using that pnode hack will always be an aliasing rule violation. Even if
it happens to work with the current GCC, it may break at any point in the
future. Please read comment #19 if you really want to be able to compile this C
code with C++ (which I don't give a darn about, and which shouldn't take
priority over making it actually VALID C – that aliasing violation is not valid
C++ either, by the way; you'd use inheritance to do this properly in C++, if
this were actually C++ code, which it isn't).
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
1:44 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #30 from mpsuzuki(a)hiroshima-u.ac.jp 2009-09-26 02:44:39 EDT ---
Thanks Behdad & Kevin,
My fix in #25 just disables the problematic optimization in GCC-4.4.x,
I know it does not fix the strict aliasing issue (I thought it was different
issue). I will try another fix following to comment #19.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
7:02 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #31 from mpsuzuki(a)hiroshima-u.ac.jp 2009-09-26 08:02:58 EDT ---
Created an attachment (id=362757)
--> (https://bugzilla.redhat.com/attachment.cgi?id=362757)
Patch to introduce the unions to the parts violating strict aliasing.
Behdad,
Following is what you suggested?
Attached is a patch to do similar modifications for all part
warned as violating strict aliasing.
diff --git a/src/cache/ftccmap.c b/src/cache/ftccmap.c
index 50a6189..8fb3387 100644
--- a/src/cache/ftccmap.c
+++ b/src/cache/ftccmap.c
@@ -283,11 +283,15 @@
{
FTC_Cache cache = FTC_CACHE( cmap_cache );
FTC_CMapQueryRec query;
- FTC_CMapNode node;
FT_Error error;
FT_UInt gindex = 0;
FT_UInt32 hash;
FT_Int no_cmap_change = 0;
+ union
+ {
+ FTC_Node node;
+ FTC_CMapNode cmap;
+ } ftc_unode;
if ( cmap_index < 0 )
@@ -371,20 +375,20 @@
#if 1
FTC_CACHE_LOOKUP_CMP( cache, ftc_cmap_node_compare, hash, &query,
- node, error );
+ ftc_unode.node, error );
#else
- error = FTC_Cache_Lookup( cache, hash, &query, (FTC_Node*) &node );
+ error = FTC_Cache_Lookup( cache, hash, &query, &(ftc_unode.node) );
#endif
if ( error )
goto Exit;
- FT_ASSERT( (FT_UInt)( char_code - node->first ) < FTC_CMAP_INDICES_MAX );
+ FT_ASSERT( (FT_UInt)( char_code - ftc_unode.cmap->first ) <
FTC_CMAP_INDICES_MAX );
/* something rotten can happen with rogue clients */
- if ( (FT_UInt)( char_code - node->first >= FTC_CMAP_INDICES_MAX ) )
+ if ( (FT_UInt)( char_code - ftc_unode.cmap->first >= FTC_CMAP_INDICES_MAX
) )
return 0;
- gindex = node->indices[char_code - node->first];
+ gindex = ftc_unode.cmap->indices[char_code - ftc_unode.cmap->first];
if ( gindex == FTC_CMAP_UNKNOWN )
{
FT_Face face;
@@ -392,7 +396,7 @@
gindex = 0;
- error = FTC_Manager_LookupFace( cache->manager, node->face_id, &face );
+ error = FTC_Manager_LookupFace( cache->manager, ftc_unode.cmap->face_id,
&face );
if ( error )
goto Exit;
@@ -413,7 +417,7 @@
FT_Set_Charmap( face, old );
}
- node->indices[char_code - node->first] = (FT_UShort)gindex;
+ ftc_unode.cmap->indices[char_code - ftc_unode.cmap->first] =
(FT_UShort)gindex;
}
Exit:
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
7:05 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #32 from mpsuzuki(a)hiroshima-u.ac.jp 2009-09-26 08:05:08 EDT ---
Created an attachment (id=362758)
--> (https://bugzilla.redhat.com/attachment.cgi?id=362758)
source rpm including the 2nd patch proposed by mpsuzuki
Here is new source rpm including my 2nd patch.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
7:10 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #33 from Kevin Kofler <kevin(a)tigcc.ticalc.org> 2009-09-26 08:10:24 EDT
---
This one should now work with GCC and other real-world compilers. It's still
not valid C because the C standard explicitly does not require a union to be
100% overlapping storage (if you assign one member, the value of the other is
undefined), in fact handling "union" as "struct" would still be a
compliant
implementation of "union". But GCC *does* guarantee that you can use
"union"
that way and other compilers in practical use do as well.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
7:24 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #34 from mpsuzuki(a)hiroshima-u.ac.jp 2009-09-26 08:24:05 EDT ---
Thank you very quick reply.
Hmm, I should rework. You suggested to pass FTC_Node variable
to FTC_CACHE_LOOKUP_CMP() and cast it to FTC_CMapNode
by caller. It is safer than using union?
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
11:36 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #35 from mpsuzuki(a)hiroshima-u.ac.jp 2009-09-26 12:36:37 EDT ---
Created an attachment (id=362766)
--> (https://bugzilla.redhat.com/attachment.cgi?id=362766)
Patch to cast the pointers allocated by FTC_XXX_LOOKUP_CMP() in the callers.
Kevin,
This is what you suggested in comment #19?
diff --git a/src/cache/ftccmap.c b/src/cache/ftccmap.c
index 50a6189..428cebb 100644
--- a/src/cache/ftccmap.c
+++ b/src/cache/ftccmap.c
@@ -283,7 +283,7 @@
{
FTC_Cache cache = FTC_CACHE( cmap_cache );
FTC_CMapQueryRec query;
- FTC_CMapNode node;
+ FTC_Node node;
FT_Error error;
FT_UInt gindex = 0;
FT_UInt32 hash;
@@ -373,18 +373,18 @@
FTC_CACHE_LOOKUP_CMP( cache, ftc_cmap_node_compare, hash, &query,
node, error );
#else
- error = FTC_Cache_Lookup( cache, hash, &query, (FTC_Node*) &node );
+ error = FTC_Cache_Lookup( cache, hash, &query, &node );
#endif
if ( error )
goto Exit;
- FT_ASSERT( (FT_UInt)( char_code - node->first ) < FTC_CMAP_INDICES_MAX );
+ FT_ASSERT( (FT_UInt)( char_code - FTC_CMAP_NODE( node )->first ) <
FTC_CMAP_INDICES_MAX );
/* something rotten can happen with rogue clients */
- if ( (FT_UInt)( char_code - node->first >= FTC_CMAP_INDICES_MAX ) )
+ if ( (FT_UInt)( char_code - FTC_CMAP_NODE( node )->first >=
FTC_CMAP_INDICES_MAX ) )
return 0;
- gindex = node->indices[char_code - node->first];
+ gindex = FTC_CMAP_NODE( node )->indices[char_code - FTC_CMAP_NODE( node
)->first];
if ( gindex == FTC_CMAP_UNKNOWN )
{
FT_Face face;
@@ -392,7 +392,7 @@
gindex = 0;
- error = FTC_Manager_LookupFace( cache->manager, node->face_id, &face );
+ error = FTC_Manager_LookupFace( cache->manager, FTC_CMAP_NODE( node
)->face_id, &face );
if ( error )
goto Exit;
@@ -413,7 +413,7 @@
FT_Set_Charmap( face, old );
}
- node->indices[char_code - node->first] = (FT_UShort)gindex;
+ FTC_CMAP_NODE( node )->indices[char_code - FTC_CMAP_NODE( node )->first]
= (FT_UShort)gindex;
}
Exit:
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
11:44 a.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #36 from mpsuzuki(a)hiroshima-u.ac.jp 2009-09-26 12:44:22 EDT ---
Created an attachment (id=362768)
--> (https://bugzilla.redhat.com/attachment.cgi?id=362768)
source rpm including the 3rd patch proposed by mpsuzuki
The source rpm including patch in comment #35.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
1:35 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #37 from Kevin Kofler <kevin(a)tigcc.ticalc.org> 2009-09-26 14:35:17 EDT
---
Yeah, that was my idea and I think that's the safest solution.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
6:05 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #38 from mpsuzuki(a)hiroshima-u.ac.jp 2009-09-26 19:05:31 EDT ---
Adam,
Could you test the new source rpm in comment #36?
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
12:37 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #39 from Adam Williamson <awilliam(a)redhat.com> 2009-09-28 13:37:17 EDT
---
yup, seems OK, navit runs without errors, nothing else seems to have any
problems.
--
Fedora Bugzappers volunteer triage team
https://fedoraproject.org/wiki/BugZappers
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
12:50 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #40 from mpsuzuki(a)hiroshima-u.ac.jp 2009-09-28 13:50:27 EDT ---
Thank you!
I will "add" the patch proposed in comment #35 to next release of freetype,
2.3.0.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
12:48 p.m.
New subject: [Bug 513582] segfault in FTC_CMapCache_Lookup()
Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug.
https://bugzilla.redhat.com/show_bug.cgi?id=513582
--- Comment #41 from mpsuzuki(a)hiroshima-u.ac.jp 2009-09-29 13:48:21 EDT ---
The patch in comment #35 is applied to git head. freetype-2.3.10 will include
it.
Thanks to everybody helped me.
--
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.
5331
days inactive
5398
days old
fonts-bugs@lists.fedoraproject.org
42 comments
1 participants
participants (1)
-
Red Hat Bugzilla