Re: [Inkscape-devel] autopackage now requires explicit --enable-binreloc (fwd)
Bryce Harrington wrote:
- The code for enable-binreloc=auto (which was previously the default) checked the build environment rather than the execution environment.
??? What does this mean?
autopackage code prevents running inkscape from the build directory. (This problem is reduced if the source directory and the build directory are the same.)
It should fall back to using the configure-supplied PREFIX value.
I don't think this is possible except if you explicitly code fallback functions in the application.
- br_locate uses sscanf("%x-%x") and unsigned ints whereaslinux/fs/proc/array.c uses printf("%08lx-%08lx") and unsigned longs. This presumably leads to bugs on 64-bit platforms.
That's because Mike's patch uses an old version of BinReloc. The CVS BinReloc uses unsigned longs and "%lx" (maybe I should change that to "%08lx" instead? not sure).
- br_prepend_prefix needlessly requires that its path argument bewritable.
It doesn't. br_prepend_prefix doesn't modify the content of the path argument in any way. I just didn't use the 'const' keyword, that's all. ;)
- br_strcat uses calloc, only to write over the space. It ineffect calculates strlen(str1) 3 times and strlen(str2) twice.
Should be fixed in the autopackage CVS now.
- br_strndup has buffer overflow if size excedes the length of strby more than one.
Thanks for the tip. Fixed in CVS now.
- br_extract_prefix doesn't behave as its documentation indicates.
This is a typo in the documentation. Fixed.
Also it does an unnecessary strndup when it could use memrchr.
I'm not sure which line exactly you're referring to.
On Sat, 2004-07-10 at 11:45 +0200, Hongli Lai wrote:
I'm not sure which line exactly you're referring to.
I think we should ignore this one, memrchr isn't portable anyway. Thanks for fixing the other problems, I'll see if I can send an updated version of the code at some point (super busy right now with other stuff ...)
Hi Mike,
After testing out the new autopackage binaries, that include this new patch, there's still an issue:
After doing: rm -rf /usr/local/lib/inkscape rm -rf /usr/local/share/inkscape rm -rf /usr/lib/inkscape rm -rf /usr/share/inkscape
and then installing via autopackage, these dirs exist:
/usr/lib/inkscape /usr/share/inkscape
however when Inkscape is started up, it cannot locate its icons. Doing this resolves the issue:
ln -s /usr/lib/inkscape /usr/local/lib/inkscape ln -s /usr/share/inkscape /usr/local/share/inkscape
Is there a way we can get this fixed quickly, or a way we could kludge around it (like have the .autopackage create the symlinks as above)?
Bryce
On Sat, 2004-07-10 at 11:20 -0700, Bryce Harrington wrote:
however when Inkscape is started up, it cannot locate its icons. Doing this resolves the issue:
ln -s /usr/lib/inkscape /usr/local/lib/inkscape ln -s /usr/share/inkscape /usr/local/share/inkscapeIs there a way we can get this fixed quickly, or a way we could kludge around it (like have the .autopackage create the symlinks as above)?
Looks like binary relocatability wasn't enabled in that build. I'll take a look at it now.
thanks -mike
On Sat, Jul 10, 2004 at 07:32:37PM +0100, Mike Hearn wrote:
On Sat, 2004-07-10 at 11:20 -0700, Bryce Harrington wrote:
however when Inkscape is started up, it cannot locate its icons. Doing this resolves the issue:
ln -s /usr/lib/inkscape /usr/local/lib/inkscape ln -s /usr/share/inkscape /usr/local/share/inkscapeIs there a way we can get this fixed quickly, or a way we could kludge around it (like have the .autopackage create the symlinks as above)?
Looks like binary relocatability wasn't enabled in that build. I'll take a look at it now.
We've disabled binreloc by default so that some of the other developers aren't disturbed by it. In the meantime, if your nightly builds can be changed to add "--enable-binreloc" to the "configure" run, Inkscape will correctly compile binreloc.
Also: I'd like to suggest some changes to make autopackage more friendly for developers. Currently you use CFLAGS to set the -DENABLE_BINRELOC, but that doesn't work sanely if "configure" is re-run and the flag changes. For example, if I compile inkscape without binreloc, and then re-run "configure" with "--enable-binreloc", prefix.cpp is not recompiled (since none of its deps changed). I would recommend two changes (which I have made to inkscape already).
In "prefix.c", add:
#ifdef HAVE_CONFIG_H # include "config.h" #endif
which will give it a dep-check if people are using autoconf. And to actually put the elements into config.h, I included calles to AC_DEFINE to make it happen. This way people can continue to use BINRELOC_CFLAGS if they want to or they can sense "ENABLE_BINRELOC" in their config.h files.
I've attached patches against binreloc CVS ...
participants (4)
-
Bryce Harrington -
Hongli Lai -
Kees Cook -
Mike Hearn