@Kai771: Yep, I was relying on the third-party stuff obeying my env more than forcing everything in the Makefile (because I always have my env set, and also to keep the Makefile tweaks to the minimum), but you got the idea, and discovered how fun it can be to track down... ^^. (It also stems from the fact that I'm a Gentoo user, and following the 'Gentoo way', I'm more inclined to fix buildsystems to honor the env rather than to 'fix' a buildsystem by forcing more stuff down its throat
).
There's also the added fun that mupdf builds the two *dump binaries for your host, not your target, so you can't have ARM stuff passed to your host's x86/x86_64 compiler, hence the whole double defines in my diff, once with -march & co, once without
.
This may not concern the TC you used, but GCC's stack-smashing protection handling & glibc's fortify support may also be things to keep in mind
. Everything depends on the TC you use (for exemple, while I do need to kill ssp & fortify, I don't have to define GNU_SOURCE on my setup).
That said, between the two of us, we've covered most of what one might encounter when trying this kind of stuff
.
(The glibc_2.15 stuff is interesting, though... my best guess would be the CS/MG TC handling fortify/ssp in a weird, or at least different way than I'm used to, since it seems to be disabled by default, and actually passing those flags/defines made things worse...). Or, possibly more likely, since the 2.15 stuff was pulled by libm, that's it's defaulting to the 'old' soft (not softfp) float ABI, and asking for the vfp fpu with these settings pulled some extra soft fp stuff from libm...