On 11/17/20 10:47 AM, Mark Gray wrote: > On 17/11/2020 00:27, Ilya Maximets wrote: >> On 11/16/20 6:11 PM, Mark Gray wrote: >>> On 16/11/2020 16:33, Ilya Maximets wrote: >>>> On 11/16/20 5:28 PM, Stokes, Ian wrote: > >> One more important thing. >> >> This commit broke windows build: >> https://ci.appveyor.com/project/blp/ovs/builds/36338443/job/bjsq27m27kworj4e >> >> Please, check this out. > > Thanks for letting me know. It looks like it cannot find the generated > file when building. However, the command used to generate it is being > executed (and presumably not failing as there is no error) as per the > dependency in the 'automake.mk' file.
OVSIDL_BUILT targets depends on ovsdb/ovsdb-idlc.in, but ovsdb/ovsdb-idlc.in depends on nothing. The build is parallel, so we might just see output of the commands printed in some order while actually dirs.py generation happened at the same time with execution of ovsdb/ovsdb-idlc.in. I'm guessing, but that is very likely since we're running 'make -j4' and these are two first commands executed. In general, I think you need to make 'ovsdb/ovsdb-idlc.in' target depend on dirs.py target to make things work in correct order. We might also need to check other targets that might use dirs.py, but doesn't have build dependencies on it. It looks like a pure luck that linux build is not broken. > > As a quick test, I generated a 'dirs.py' file and pushed it to a private > branch. The build will then not fail at that point. I also added an "ls" > command into the build and the file is there with the correct > permissions (644). > > As I don't have a Windows machine to test this on, hopefully Alin can help? You could configure and use appveyor for your own github repo. This is not a great way to debug things, but at least something. > > Mark >> >> Best regards, Ilya Maximets. >> > _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev