On 07/10/2015 09:25 PM, enh wrote: > (already submitted as https://android-review.googlesource.com/159035/ > --- i was under pressure because i didn't notice the problem until i'd > already broken all the 32-bit builds. as usual i tested on 64-bit. > i've removed the non-standard basename_r and dirname_r from 64-bit
Sigh. There's already a basename_r. Well of course there is. It sounds like a good idea, personally I'd submit it to posix rather than remove it. A better fix would be to move basename_r() to portability.c and rely on libc to provide it if they have it. (I assume that your basename_r() and the implementation I did aren't too far apart behavior-wise?) I need to dig up my AOSP build and try to chip the toolchain out to test build toybox. (Will static bionic builds run on ubuntu's kernel?) > bionic, but i can't fix the past. the build failure is because the > 32-bit bionic basename_r had the signature int basename_r(const char*, > char*, size_t). looking ahead maybe i should add a _BIONIC_NO_CRUFT so > non-historic 32-bit code can opt out of this kind of pollution.) Seriously, I'd just submit this one to the Austin group for the next posix/susvX spec. What the gnu guys did is a way bigger deviation than bionic here. > commit f8b41e81fca2b53410f80c95f111c754d5eff99a > Author: Elliott Hughes <[email protected]> > Date: Fri Jul 10 19:11:18 2015 -0700 I've applied this, but I'm open to using bionic's basename_r() instead of mine instead. FYI. Rob _______________________________________________ Toybox mailing list [email protected] http://lists.landley.net/listinfo.cgi/toybox-landley.net
