On Mon, 31 Oct 2016 07:39:48 +0100 Sebastiaan Couwenberg <sebas...@xs4all.nl> wrote: > On 10/31/2016 05:13 AM, Anders Kaseorg wrote: > > On Sun, 30 Oct 2016, Bas Couwenberg wrote: > >> The recent git upload FTBFS on mips64el due to OOM, the missing build on > >> mips64el is preventing the qgis rebuild as part of the ongoing gdal > >> transition (#842288). > >> > >> The build was performed on mipsel-manda-02 which is known to have issues > >> like these (although git built successfully on this buildd before). > >> Perhaps a rebuild on another buildd will be sufficient. > > > > Eep. > > > > Looking at > > https://buildd.debian.org/status/logs.php?pkg=git&arch=mips64el, the build > > has been attempted three times now, including once on mipsel-aql-02, > > failing all three times with out of memory errors. None of the failing > > tests (t0023-crlf-am.sh, t0064-sha1-array.sh, t0302-credential-store.sh) > > have changed between 2.10.1 and 2.10.2, and 2.10.1 built with no trouble. > > These tests should be using basically no memory (about 4 MB each). > > > > Maybe another test running in parallel is using memory and causing these > > tests to fail? Could these build machines be so underpowered that they > > shouldnât be setting DEB_BUILD_OPTIONS=parallel=4? > > > > Unless someone familiar with the mips64el buildd situation has a better > > plan, Iâm unsure how to proceed except by disabling individual failing > > tests, or the entire test suite, for mips64el. > > > > Any ideas? > > Disabling the tests on mips64el is reasonable. > > You can also do a build on the mips64el porterbox if that succeeds > without changes you can just upload that.
It is better to understand what is going on first. git is an important piece of software and the tests are here for a reason. As Bas said, we have porterboxes where it is very easy to install packages and try to reproduce this, so I'd suggest that as a first step. The machines should be fine, they have 8GB of memory, which should be sufficient for parallel=4. Cheers, Emilio