[Toybox] [PATCH] Specify owner when testing tar with /dev/null

[enh]

On Thu, Dec 7, 2023 at 9:45 AM Colin Cross <ccross at google.com> wrote:
>
> On Thu, Dec 7, 2023 at 8:26 AM enh <enh at google.com> wrote:
> >
> > On Wed, Dec 6, 2023 at 6:44 PM Rob Landley <rob at landley.net> wrote:
> > >
> > > On 12/6/23 18:15, Colin Cross wrote:
> > > >> Who _does_ own /dev/null in your test environment?
> > > >
> > > > "nobody":
> > > >
> > > >     FAIL: tar pass /dev/null
> > > >     echo -ne '' | tar c --mtime @0 --group sys /dev/null 2>/dev/null | LST
> > > >     --- expected 2023-12-07 00:03:55.797797713 +0000
> > > >     +++ actual 2023-12-07 00:03:55.805797762 +0000
> > > >     @@ -1 +1 @@
> > > >     -crw-rw-rw- root/sys 1,3 1970-01-01 00:00 dev/null
> > > >     +crw-rw-rw- nobody/sys 1,3 1970-01-01 00:00 dev/null
> > > >     external/toybox/toybox-gtests.cpp:168: Failure
> > > >     Expected equality of these values:
> > > >       exit_status
> > > >         Which is: 1
> > > >       0
> > >
> > > Does bionic still have getuid() stubbed out? (I.E. does toybox's "stat -c %U/%G
> > > /dev/null" work?)
> >
> > you mean getpwuid(), i assume? but i don't think that's relevant,
> > because i think ccross is talking about host testing. that's in an
> > nsjail set up by build/soong/ui/build/sandbox_linux.go.
>
> This isn't the Soong nsjail sandbox, this is a sandbox created by some
> combination of a python script (atest), which calls a java tool
> (tradefed), which runs bazel, which runs the test.  My guess is that
> the sandboxing is all happening in bazel's test runner.

ah, is that sandbox relevant now Android's not moving to bazel then?
or does the soong sandbox have the same problem with this test?