- Inkscape Devel - Inkscape Mailman

Race condition in sp_gradient_invalidate_vector()?
by Ralf Stephan 04 Jul '05

04 Jul '05

Hello, both bugs 1038594 and 1196143 appear to be 'fixed' by not g_free'ing gr->color. This is a text gradient buffer mangled in both bugs (in 1196143 you can see some destruction of the picture when repainting after dialog release). In 1038594, valgrind tells me about attempted reads from the same address which was freed before, and gtk likely reuses the memory, and this fast. This would mean that there is a copy of the color buf somewhere that is not updated but I still am not well into the code. I would propose memory handling of this buffer to be moved into GC, before the release, or using malloc/free if that works. https://sourceforge.net/tracker/?func=detail&atid=604306&aid=1038594&group_… https://sourceforge.net/tracker/?func=detail&atid=604306&aid=1196143&group_… Regards, ralf

5 7

Removing C-Style casts
by Ben Fowler 04 Jul '05

04 Jul '05

In the file libnr/nr-point.h, there are some C-Style casts in the code: /** \brief A function to lower the precision of the point \param places The number of decimal places that should be in the final number. This function accomplishes its goal by doing a 10^places and multipling the X and Y coordinates by that. It then casts that to an integer (to get rid of all decimal points) and then divides by 10^places back again. */ inline void round (int places = 0) { _pt[X] = (Coord)(Inkscape::round((double)_pt[X] * pow(10, places)) / pow(10, places)); _pt[Y] = (Coord)(Inkscape::round((double)_pt[Y] * pow(10, places)) / pow(10, places)); return; } Now casting _pt[] to a double is no-op, and possibly the (Coord) cast is also unnecessary, or could be converted to a function-style one: Coord( ... ) . My problem centres on ensuring that the code matches the comment, which refers to casting to an int. I am fairly sure that the Inkscape::round( ) function actually is a wrapper around std::floor( ) which does everything in IEEE double precision. IMHO, the Coord type should have a round( ) function, which I have had to provide under a different name: namespace NR { /** - * A "real" type with sufficient precision for coordinates. + * A floating point type with sufficient precision for coordinates. * * You may safely assume that double (or even float) provides enough precision for storing * on-canvas points, and hence that double provides enough precision for dot products of * differences of on-canvas points. */ -typedef double Coord; + typedef double Coord; + + /** + \brief A function to lower the precision of a co-ordinate + \param places The number of decimal places that should remain in + this co-ordinate. + + This function accomplishes its goal by multipling the co-ordinate by + (10^places). It then uses the Inkscape round( ) function (currently + implemented with std::floor( )) to remove the surplus precision and + then divides by (10^places) to scale back again. + */ + inline Coord set_precision (Coord c, int places = 0) { +// g_assert( places > 0 ); // Assert on do-nothing calls and negative number of decimal places + double factor = pow(10, places); + return Coord((Inkscape::round(c * factor) / factor));; + } } /* namespace NR */ and then I can re-do the function as: /** \brief A function to lower the precision of the point \param places The number of decimal places that should remain in the co-ordinates of the point. */ inline void round (int places = 0) { // g_assert( places > 0 ); // Assert on do-nothing calls and negative number of decimal places _pt[X] = NR::set_precision(_pt[X], places); _pt[Y] = NR::set_precision(_pt[Y], places); return; } Is this the way to go? Would these changes be acceptable in any way? Should Coord be a class rather than a typedef? Should the round( ) function(s) which are really floor( ) functions be renamed? Should the round functions(s) which are really set_precision( ) (if any) be renamed? Should any of these functions be templates e.g. (untested) template <class T> inline void round( int places=0 ) { // round co-ordinates to places decimal places } How are modifications to working code meant to be tested? I would be happy to submit patches for review, or pursue this in any other way: I am trying to see whether it is viable to remove C-Style casts as per http://www.inkscape.org/cgi-bin/wiki.pl?InkscapeJanitors

1 3

Win32 testers needed!
by Bryce Harrington 03 Jul '05

03 Jul '05

Hi all, We are closing in on the 0.42 release, and getting the last few bugs fixed. There are a _number_ of bugs which appear to be Windows-only. Unfortunately, since the developers are linux-based, it's hard to determine how to fix them and to check that the fixes worked. Thus, if you are a Windows Inkscape user, we really need your help right now, in the run up to the release. Please review the Bug Tracker, and in particular look at the bugs that are ranked at '9'. Install both 0.41 and a version from CVS, and try to recreate each bug on each copy. Then add a comment to the bug report about whether you can or cannot replicate the error. Make sure you're logged into SourceForge, so the developers can contact you with more questions or to verify if a change fixed it. If you do not have a SourceForge account, please include your email address. If you want to help even more, it is extremely valuable to have backtraces. You can learn how to make backtraces on the Report A Bug page on the website. The more you can do to help narrow in on where the bug occurs, the more quickly it can be fixed. Please try to find a couple evenings over the next week to devote to doing some testing; even if you can only test a few bugs, you will help make Inkscape on Windows that much more robust. Thanks, Bryce

1 0

New gtk bundle
by Bob Jamison 03 Jul '05

03 Jul '05

Hi, all, I uploaded another copy of the Win32 libs for the build. This one has xmlparse.dll and libtiff3.dll added, and some unnecessary files removed. It is here: http://inkscape.org/win32/ It is 7-zipped to reduce the size. 24mb is much nicer than 52. I added some lines to the 'dist' target of Makefile.mingw to use these files. Bob

2 2

win32 crash on creating about memory dialog
by Adib Taraben 02 Jul '05

02 Jul '05

Hi, on win32 the current cvs crashes after pressing "about memory" function. I use the new gtk26 libs from Bob. I tried this on linux and there it is ok. I have no glue whats going on as I don't know gtk. I think that has to be resolved before release, thats why I post here. Attached there is the bt: HTH, Adib. --- gdb: kernel event for pid=3976 tid=2572 code=EXCEPTION_DEBUG_EVENT) gdb: Target exception EXCEPTION_ACCESS_VIOLATION at 0x05013750 Program received signal SIGSEGV, Segmentation fault. 0x05013750 in Glib::Object::Object (this=0x8326d98, __vtt_parm=0x8e4830, construct_params=@0x22ec80) at object.cc:214 in object.cc (gdb) 214 object.cc: No such file or directory. bt #0 0x05013750 in Glib::Object::Object (this=0x8326d98, __vtt_parm=0x8e4830, construct_params=@0x22ec80) at object.cc:214 #1 0x05fb5557 in Gtk::Object::Object (this=0x8326d98, __vtt_parm=0x8e482c, construct_params=@0x22ec80) at object.cc:35 #2 0x0073d895 in _fu56___ZTTN3Gtk14TreeViewColumnE () at c:/gtk26/include/gtkmm-2.4/gtkmm/treeviewcolumn.h:848 #3 0x0073df3c in Gtk::TreeView::append_column<Glib::ustring> (this=0x8348fc4, title=@0x22edb0, model_column=@0x8348fa0) at c:/gtk26/include/gtkmm-2.4/gtkmm/treeview.h:1365 #4 0x007730f3 in Inkscape::UI::Dialog::Memory::Private::Private ( this=0x8348f90) at ui/dialog/memory.cpp:93 #5 0x0066ce55 in Inkscape::UI::Dialog::Memory::Memory (this=0x8337b98) at ui/dialog/memory.cpp:206 #6 0x00772d16 in Inkscape::UI::Dialog::Memory::create () at ./ui/dialog/memory.h:23 #7 0x007727a3 in Inkscape::UI::Dialog::(anonymous namespace)::create<Inkscape::UI::Dialog::Memory> () at ui/dialog/dialog-manager.cpp:46 #8 0x00511f91 in Inkscape::UI::Dialog::DialogManager::getDialog ( this=0x72b9dd8, name=1162) at ui/dialog/dialog-manager.cpp:137 #9 0x00511ff0 in Inkscape::UI::Dialog::DialogManager::showDialog ( this=0x72b9dd8, name=1162) at ui/dialog/dialog-manager.cpp:156 #10 0x00511fd6 in Inkscape::UI::Dialog::DialogManager::showDialog ( this=0x72b9dd8, name=0x85a9d2 "Memory") at ui/dialog/dialog-manager.cpp:149 #11 0x004d4076 in Inkscape::HelpVerb::perform (action=0x74cf990, data=0x95, pdata=0x0) at verbs.cpp:1472 #12 0x004e062a in sp_action_perform (action=0x74cf990, data=0x0) at helper/action.cpp:135 #13 0x0043767f in sp_ui_menu_activate (object=0x7526b28, action=0x74cf990) at interface.cpp:284

2 1

Re: [ inkscape-Patches-1000235 ] Save as Flash (.swf)
by Ted Gould 02 Jul '05

02 Jul '05

Raymond Penners wrote: > > definitly lacking on some ways that your plugin is good, but > we'd like to start using the extension approach in the > future, and spalah can do some things your code can't. > Also, we really don't want to add a dependency on ming. I > realize it is a configure option, but the reality is that no > one compiles from source. If we don't make it the default, > no one will ever see the code. It's the sad reality of open > source today. > > Thanks for your contribution, I hope that you will consider > contributing to Spalah, or to Inkscape in other ways. Sorry > again to take so long to get back to you. > > I am wondering, does this mean that Spalah will be distributed > out-of-the box with Inkscape? If not, then IMHO not having a simple > "Save as .swf" feature in Inkscape would be a real shame. Well, we're still trying to figure all that out. On one hand we really want to reduce the size of Inkscape, make it easier to install and faster to load. Which means that we want to make as much external as possible. On the other hand, most people aren't of the mindset to install lots of different packages. On Linux this is pretty easy, most distributions have reasonable package management. On Windows, it is much less so. But it still leaves the issue: how do users know where to go to get these features? We need some way to advertise them also. I realize that wasn't much of an answer to your question, but that is more because we don't quite have one yet. It's quite a thorny issue. --Ted

3 2

LJ Editor's Choice 2005
by unknown＠example.com 02 Jul '05

02 Jul '05

I just flipped to page 82 of the August 2005 issue of Linux Journal. Congratulations to all you devs for the Editor's Choice Award for Graphics Software. Aaron Spike

5 4

object.cpp build failure on Solaris
by Christopher Wiedel 02 Jul '05

02 Jul '05

I've been trying to build Inkscape 0.41 on sparc-sun-solaris2.8, and have run into a bit of a problem. It configures fine, and when I start 'make' it initially goes well, occasionally popping out warnings like: node-context.cpp: In function `void nodepath_event_attr_changed(SPRepr*, const gchar*, const gchar*, const gchar*, bool, void*)': node-context.cpp:313: warning: cast from `GTypeInstance*' to `SPItem*' increases required alignment of target type node-context.cpp:322: warning: cast from `GTypeInstance*' to `SPItem*' increases required alignment of target type However, it blows up when it hits src/object-edit.cpp. Or more accurately, something inside it. It throws out a great many of the same type of warnings as the ones above. This doesn't appear to be the problem, since other files had the same warnings, but were able to compile. It winds up with this: object-edit.cpp: In function `void sp_offset_offset_set(SPItem*, const NR::Point&, const NR::Point&, unsigned int)': object-edit.cpp:897: warning: cast from `GTypeInstance*' to `SPOffset*' increases required alignment of target type object-edit.cpp: In function `NR::Point sp_offset_offset_get(SPItem*)': object-edit.cpp:909: warning: cast from `GTypeInstance*' to `SPOffset*' increases required alignment of target type make[2]: *** [object-edit.o] Error 1 make[2]: Leaving directory `/usr/foo/inkscape/0.41/inkscape-0.41/src' make[1]: *** [all-recursive] Error 1 make[1]: Leaving directory `/usr/foo/inkscape/0.41/inkscape-0.41' make: *** [all] Error 2 This is using gcc 3.3.2. It doesn't appear to have any actual error message that I can find. I would appreciate any pointers. Thanks. -- Christopher Wiedel

2 1

Fwd: Re: [Inkscape-devel] Inkboard CVS updates
by David Yip 01 Jul '05

01 Jul '05

On Friday 01 July 2005 01:04 pm, mental@...3... wrote: > I suspect the element and its sodipodi:type attribute might be > getting transmitted separately. > > i.e. the remote client ends up getting a bare <svg:path>, and > immediately creates an SPPath object. The sodipodi:type attribute > which would have instructed it to create an SPEllipse or whatever > might not arrive until it's too late. Actually, it seems like that's exactly what's going on, judging from the way Inkboard processes messages: the new object message arrives, so Inkboard creates a new object. However, only _then_ does it actually parse the type attribute and set it. I'm reworking received_new_object_helper to extract and set the type attribute before it creates the new object. Hopefully, this'll fix the problem. -- David

1 0

Inkboard CVS updates
by David Yip 01 Jul '05

01 Jul '05

I have committed a number of fixes to the Inkboard CVS <http://sourceforge.net/projects/inkboard> that address crashes related to buffer overflows. Instructions for building and running Inkboard can be found in my request for Inkboard testers that was posted to this list. Inkboard should now be much more stable than it previously was, sans conditions like invalid Jabber usernames or passwords. (I haven't gotten around to debugging those yet; I've been chasing down another bug, which is described below.) In a session between two peers Alice and Bob, if Alice sends a star, spiral, or ellipse, it will show up correctly on Bob's canvas, and (almost) any manipulations that Alice makes will be correctly replicated on Bob's canvas. However, if Bob manipulates the star, spiral, or ellipse in any way, the changes will not be correctly replicated on Alice's canvas. Additionally, Bob will not see the object as a star, spiral, or ellipse -- he will only see it as a path. I think these two problems are related, and since I'm not sure where to start with addressing the first problem, I've been focusing on the second, in the hope that it will cure the first :) I don't think this is a case of missing information: if Bob saves the document, exits, and then reopens it, any star, spiral, or ellipse will be correctly recognized as such, and not just a path. I think there's just something that Inkboard is just not triggering correctly. If anyone has any ideas on where to look or what to look for, I'd greatly appreciate them. Another, smaller bug: gradient editing is not replicated properly. Pressing "add" will replicate the changes, but the gradient objects themselves fall out of sync. Finally, just some thoughts about the near-future integration with Inkscape: Ideally, I'd like to begin merging Inkscape with Inkboard only when all of the bugs have been ironed out of the Inkboard code. That's rather unrealistic, though, especially since the act of merging may expose new bugs anyway! :) So it may be a good idea to create a new branch in the Inkscape codebase specifically for integration with Inkboard, and then merge branches when an acceptable level of robustness is achieved. -- David

3 3