[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH] tests: make address@hidden run sanitizers
From: |
Fam Zheng |
Subject: |
Re: [Qemu-devel] [PATCH] tests: make address@hidden run sanitizers |
Date: |
Tue, 13 Mar 2018 01:41:02 +0800 |
User-agent: |
Mutt/1.9.2 (2017-12-15) |
On Mon, 03/12 18:17, Marc-André Lureau wrote:
> Hi
>
> On Mon, Mar 12, 2018 at 6:11 PM, Fam Zheng <address@hidden> wrote:
> > On Mon, 03/12 15:11, Marc-André Lureau wrote:
> >> Hi Fam,
> >>
> >> It would be great if patchew could run the debug/sanitizer build.
> >> Unfortunately, except x86 target, the run is full of ASAN leaks or
> >> warnings.
> >>
> >> Do you think patchew could learn to do regression testing? By that I
> >> mean that patchew could compare a patchset output with the current
> >> master (that output should be cached for multiple tests). The output
> >> to compare here would be the errors reported by ASAN during the build.
> >> (eventually, this could be used for other metrics some day). If so,
> >> would you be willing to help me implementing it?
> >
> > The question is can we simply rely on the exit code of cmp or diff for that
> > kind
> > of comparison? If there is any out-of-order texts, that simple diff won't
> > work.
>
> If the tests are run sequentially, i think it should produce the same
> output when run multiple times (to be verified).
In that case the test will last for a long time. Currently patchew isn't very
good as multi-tasking. But that is one option we can explore (like we can devote
one VM for asan test).
>
> However, if there are leaks already, and the address of allocation or
> symbols change, this may produce diff on existing lines.
Maybe we can come up with a 'make address@hidden' or a
relatively simple 'make address@hidden | sed -e "..."' command that
can be fed to diff.
>
> What I would like to see is the diff of added (or removed) lines.
>
> I suppose that should be possible, I'll do some experiments.
>
> > Anything more intelligent/sensible should go as a policy, and patchew should
> > continue focus on mechanisms.
>
> hmm ok :)
Like said above, I think it's possible to postprocess the output to make it easy
to diff/cmp. By mechanism/policy I mean the place to draw a line to code
Patchew. However patchew has a database where we can write test script snippets
(as configurations) that is "the policy".
Fam