On Mon, Oct 28, 2019 at 07:53:24AM +0100, Martin Pitt wrote: > Olly Betts [2019-10-27 11:52 +1300]: > > However, then the build fails for me running tests using ccache - the > > problem is that $HOME is /sbuild-nonexistent under sbuild, which > > doesn't exist, and cache tries to create its cache under $HOME/.ccache > > by default. > > I don't have that, but I also have failing tests.
I think this is #935817 - if so it's not tests using ccache that are the problem, but that meson automatically tries to use ccache if it's installed, plus mk-sbuild seems to install ccache in chroots. (That seems a mis-feature to me - the usual ways to use ccache are to either something like `CC="ccache gcc"`, or to add /usr/lib/ccache to your PATH ahead of /usr/bin, which both require a concious extra action, but I think that's a good thing - it's surprising for building some software to create a persistent hidden directory under your home directory potentially using a lot of diskspace just because the package happens to use meson for its build system and you're on a multi-user system which happens to have ccache installed). It seems brittle for a build to fail because an extra package is installed, especially one like ccache which might reasonably have been added to a chroot to speed up builds. > > [2/3] /usr/bin/arachne-pnr -q -d 1k -p '../test cases/fpga/1 > > simple/spin.pcf' spin.blif -o spin.asc > > FAILED: spin.asc > > /usr/bin/arachne-pnr -q -d 1k -p '../test cases/fpga/1 simple/spin.pcf' > > spin.blif -o spin.asc > > spin.blif:50: fatal error: invalid formal-actual > > ninja: build stopped: subcommand failed. > > For me, fpga also fails, but the failure looks different: I also see that eatmydata seems to cause problems (#917501), which again is something that might reasonably be in use. I've been running my package builds under it for the best part of a decade (including many NMUs) and this is the first issue I've hit. I tried temporarily disabling that, but meson still fails to build. Cheers, Olly