On Tue, 14 Jan 2020 at 11:23, Richard Purdie <
richard.purdie@linuxfoundation.org> wrote:

> For extra data I reran this on a builder where it failed. This time it
> worked out ok. I can think of two different things we could try:
>
> a) Add some debug to the code so that when a unknown user error occurs,
> it triggers a mention of the filename with the issue
>
> b) Potentially try turning off any parallel install code for python.
> I'm starting to wonder if that is racing somehow over file ownership.
> Making it single thread may make the issue appear (or it may
> disappear).
>

I tried setting PARALLEL_MAKEINST to "", and it didn't make any difference
(python is using standard autotools). I guess we'll have to catch the build
directory when it happens again somehow.

Alex