Andreas Enge <andr...@enge.fr> skribis: > On Mon, Nov 03, 2014 at 10:00:38AM +0100, Ludovic Courtès wrote: >> Same problem as with Valgrind, so I would suggest disabling builds on >> MIPS, as done in commit 67a86d3. > > Speaking of valgrind, how is transitivity of disabled builds handled, > on hydra and by the users?
There’s nothing preventing users to build Valgrind on mips64el-linux, currently. On Hydra, build-aux/hydra/gnu-system.scm uses ‘package-transitive-supported-systems’, which returns the intersection of a package’s ‘supported-systems’ and that of its inputs (with the assumption that implicit inputs are all supported everywhere.) > I noticed that some packages do not build on mipss due to a dependency > on valgrind, for instance petsc-openmpi and petsc-complex-openmpi: > http://hydra.gnu.org/build/132921 > http://hydra.gnu.org/build/132781 > It was a bit confusing to me at first, since valgrind was not in the > "still failing jobs" tab, until I realised it was also not in the "still > succeeding jobs" tab and in fact disabled. > > The two packages in question depend indirectly on valgrind via openmpi. > Now openmpi seems to be disabled for mips on hydra also, although it is > not explicitly disabled in the package recipe. These two other packages > are not disabled, however. So does this mean that the dependency graph > for disabled inputs is followed only to depth 1? Oh right, that’s a bug. Fixed in commit c37a74b. Thanks for the report! Ludo’.
