On 2018-09-10, David Gibson wrote: > On Thu, Sep 06, 2018 at 03:34:35PM +1000, David Gibson wrote: >> On Tue, Aug 21, 2018 at 11:02:37AM +0200, Hector Oron wrote: >> > 2018-08-16 22:30 GMT+02:00 Rob Herring <robh@xxxxxxxxxx>: >> > > On Thu, Aug 16, 2018 at 1:43 PM Vagrant Cascadian <vagrant@xxxxxxxxxx> wrote: >> > >> On Jul 2018-07-23: >> > >> > Hi everyone, I've just tagged and released v1.4.7 of the device tree >> > >> > compiler. ... >> > When attempting to run tests on platforms that have no valgrind >> > available, few test cases are unable to run. >> > Valgrind has become a dependency of `make check` instead the old `make >> > checkm`. The offensive commits are: >> > >> > 4b8fcc3 libfdt: Add fdt_check_full() function >> > c14223f tests: Use valgrind client requests for better checking >> > 5b67d2b tests: Better handling of valgrind errors saving blobs >> > 85bce8b tests: Correction to vg_prepare_blob() >> > >> > Could you consider reverting those or move the affected testcases to >> > `make checkm` realm, that way devicetree-compiler testsuite would >> > still be functional on platforms (i.e. ARMv5tel) that have no valgrind >> > support. >> >> Sorry I've taken so long to reply. >> >> It's not really an issue of removing specific testcases. make checkm >> doesn't run different tests - it runs the same tests, all running >> under valgrind. We needed the valgrind interface things to prevent >> some incorrect errors, and catch some other errors which otherwise we >> wouldn't pick up. >> >> In theory, fixing the problem for systems without valgrind is easy: we >> can just define the valgrind calls to be no-ops in that case. >> >> The difficulty is that we need to compile the code differently for >> that case, which means more conditional compilation than we currently >> have, and is kind of awkward in the Makefiles. >> >> I'm sure it's possible to sort it out, I've just been pretty busy with >> my day job, so I haven't had a chance to look at it properly. > > Ok, I think I've fixed this now in master, and I've set up the Travis > build to check building without valgrind to stop it breaking again in > future. Thanks! Updated the Debian package with the patch: b94c056b137e59deefc62fbfe0cd3a23edfcc07c Make valgrind optional With that it builds on architectures with valgrind and without valgrind for all of Debian's release architectures: https://buildd.debian.org/device-tree-compiler A few unrelated build failures remain for some of the more obscure architectures, if you're curious. live well, vagrant
Attachment:
signature.asc
Description: PGP signature
