GNOME Bugzilla – Bug 730859
MinGW build fails on missing atk.def
Last modified: 2014-06-03 16:00:54 UTC
Created attachment 277351 [details] [review] Fix for atk/Makefile.am Since 41442d82 we no longer generate atk.def, but we still read it when linking libatk on MinGW. This causes a build failure: CC atkwindow.lo CC atk-enum-types.lo /usr/bin/i686-w64-mingw32-windres atk.rc atk-win32-res.o CCLD libatk-1.0.la libtool: link: symbol file `atk.def' does not exist make[3]: *** [libatk-1.0.la] Error 1
Review of attachment 277351 [details] [review]: Just testing, and the patch fails to apply with the last master. Could you update it?
CCing Chun-wei Fan, that is the one that have more experience with Win32 related compilation.
The patch was developed against 22d344af, which is still the latest master, so I'm not sure why it's not applying for you. I just double-checked and it applies for me.
(In reply to comment #3) > The patch was developed against 22d344af, which is still the latest master, so > I'm not sure why it's not applying for you. My bad, I was applying on the wrong branch. > I just double-checked and it applies for me. Yes, it applies correctly, and I was able to compile and install atk without problems on a Linux environment. But I don't have the proper environment to test this on Win32, so if you don't mind, we will wait a little to see if Chun-wei Fan have something to say about the patch. Again, sorry for my mistake.
Hi, Sorry, I was away for a few days, so I wasn't able to see this bug until now. I would not be the one that would know the best on MinGW builds of ATK (as I don't usually use it at all), so I can't really say much about it, so I think I might CC LRN (who does the MinGW builds) to see whether this would be okay for him as well. With blessings, thank you!
May i suggest using > -export-symbols-regex '^atk_' instead?
Since ATK is now using visibility-based symbol exporting (as of 41442d82, the commit that introduced the bug), isn't it true that we no longer need -export-symbols* at all?
If ATK uses visibility the same way GTK does - then yes, no need for -export-symbols* of any kind.
Hi, The original intent of the mentioned patch is to use visibility, just as what GLib and GTK+ does... the -export-symbols* (for MinGW builds) should not be needed, as far as I can tell. With blessings!
So, patch is OK. What do we do now? Its status is "reviewed" currently. Should someone change it to "ready to push" or something?
Review of attachment 277351 [details] [review]: (In reply to comment #10) > So, patch is OK. What do we do now? Its status is "reviewed" currently. Should > someone change it to "ready to push" or something? Just doing. Thanks everybody.
Hi, I have pushed the patch as cbf8e493f, so closing this bug. Thanks, with blessings!