Hi Steve, On Wednesday, 30 August 2017 3:34 PM, Steve Langasek wrote: > I'm taking a look at these patches now, prompted by the fact that Ubuntu's > incomplete multiarch transition for libfreetype6-dev has broken at least one > reverse-dependency in contrib (astromenace).
Thanks for looking in on this. We definitely don't want broken packages. > It sounds like you've only looked at runtime dependencies. The primary > concern is with packages that *build*-depend on libfreetype6-dev; there are > far more than 41 of those, I count 243 in main alone. What command did you use to determine this? I'm obvously doing something wrong, as "apt-cache rdepends libfreetype6-dev" only gives me 41 packages. As to their use of freetype-config, the script I was using was grepping built packages, not sources. So that was a big mistake. :-( Anway, taking another look at this, I still found 17 packages using the script. If you also include rdepends on libfreetype6, around 160 packages seem to use freetype-config (out of 953 rdepends packages). > Are you willing to do a rebuild test of these reverse-dependencies against > libfreetype6-dev with freetype-config removed, to verify that none of them > regress? Yes. But it will take a while! We might also take an intermediate step by keeping freetype-config, but enforcing the use of pkg-config by removing the 'else' block after the check for pkg-config (see line 37). Aside from flagging that the script will be removed in a future release, this would also solve the multi-arch conflict in freetype-config. >>On Mon, Aug 28, 2017 at 02:49:22AM +0000, Hugh McMaster wrote: > >> --- debian/rules 2017-08-27 17:59:43.522265000 +1000 >> +++ debian/rules 2017-08-27 19:46:36.000000000 +1000 >> @@ -83,6 +83,7 @@ >> dh_auto_build -D $(ft2demos_u) -- TOP_DIR=../$(freetype_u) \ >> OBJ_DIR=../$(freetype_u)/objs >> endif >> + sed -i 's/\(\#define SIZEOF_LONG\).*/\1 __SIZEOF_LONG__/' >> $(freetype_u)/builds/unix/ftconfig.h >> >> override_dh_auto_clean: >> rm -f unpack-stamp patch-stamp > > I would prefer to see this as a patch to the source, not as post-processing > from debian/rules. However, that means a bit of fiddling in > builds/unix/configure.ac, so I'm ok to proceed with this approach for now. TBH, I'm not sure if it's possible to make this change in builds/unix/configure.ac. #define SIZEOF_LONG __SIZEOF_LONG__ became #define SIZEOF_LONG 8 when I ran initial tests on amd64. So I was forced to make the change after dh_build had finished, in order to prevent the substitution. That said, I'm not too familiar with autoconf, so this could be possible. -- Hugh McMaster