On 2015-07-21 19:02, Johannes Schauer wrote: > Hi, > > Quoting gustavo panizzo (gfa) (2015-07-21 08:05:47) >> On 2015-07-20 16:34, Johannes Schauer wrote: >>> crossbuilding is not equal to native building. I think Daniel was looking >>> for a >>> way to test if their packages build natively on hurd-i386 or kfreebsd-amd64. >>> But crossbuilding from linux (which I'll just assume Daniel runs on their >>> host) >>> to hurd or kfreebsd will lead to different results than natively building on >>> these kernels in (probably) most cases. >> would that make the test useless? I build test using qemu-static-XXXX >> before upload to debian to avoid FTBFS when the package reach the buildd >> network > > are you using qemu-static-XXXX to build on linux for hurd or kfreebsd? no, in the quick test i did it didn't work (their binfmt appears to be the same)
> > The *only* 100% sure way to know if your package builds on the buildds is... > to > upload and build your packages on the buildds. that's the thing i want to avoid, useless uploads > > The second best way is to build either on real hardware (your own or a > porterbox) or in a virtual machine which emulates a whole machine, including a > kernel. > > The third best way is to only emulate the architecture but not the kernel. > This > is what qemu-user does. This will probably let you run into kernel specific > problems. [ snip ] well what I do worked in the test I did, for example trinity 1.5-1 FTBFS on mips, mipsel and sparc due a missing include header, that i could catch using qemu-static-XXX. I agree that more complex interactions between the userspace and the kernel may not work. I tough that qemu-static-XXX would be better than you show, but hey! I didn't load a kernel so it is kinda expected I didn't want to but I think i will create a set of vm, one for each debian arch, and automate the build on it, I looked how to create a buildd machine but is think is too much. I will stick to libvirt+scripts to launch the build on different arch. :( i didn't want to use more vm, pbuilder is so cool and easy to maintain > > And sure, if your package does noting fancy but just calls cp a couple of > times > to put everything in the right location, then nothing will go wrong with the > qemu-user method. But if your package is that simple, then you probably also > don't need to test it. again, dependencies and includes sometimes are different across arch, that could be spotted > > cheers, josch > cheers -- 1AE0 322E B8F7 4717 BDEA BF1D 44BB 1BA7 9F6C 6333 keybase: http://keybase.io/gfa -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected] Archive: https://lists.debian.org/[email protected]
