[Toybox] [PATCH] tar.test: don't test non-`-p` behavior as root.

[enh]

On Thu, Aug 1, 2024 at 10:08 PM Rob Landley <rob at landley.net> wrote:
>
> On 8/1/24 10:24, enh wrote:
> >> > i'm not sure what you're talking about? this isn't a hash-related test?
> ...
> >> The tar "ownership" test.
> >>
> >> FAIL: tar ownership
> ...
> > yes, on Android (as root, the only way we test) only two tests are skipped:
> >
> > SKIP: tar ownership
> > SKIP: tar honor umask
> >
> > the reason for the former (as hinted at by the `skipnot chown
> > nobody:nogroup dir/file 2>/dev/null`) is that there is no "nogroup" on
> > Android.
>
> $ grep ^no /etc/group
> nogroup:x:65534:
> nobody:x:65534:
>
> On debian it appears to be an alias for nobody. Which probably shows up in "ls"
> because it's first in the /etc/groups file. I have no idea why debian did that.
>
> > an alternative here would be to use nobody:nobody, which works on all
> > of Android/Debian/macOS. i don't know if we considered that at the
> > time? it sounds familiar now i'm typing it out...
>
> Ok, here's the problem:
>
> $ git annotate tests/tar.test | grep '"ownership"' -A1
> 43d398ad5       (Elliott Hughes 2019-07-18 15:52:55 -0700       192)testing "ownership"
> "$TAR dir/file | SUM 3" \
> 43d398ad5       (Elliott Hughes 2019-07-18 15:52:55 -0700       193)
> "2d7b96c7025987215f5a41f10eaa84311160afdb\n" "" ""
> $ git show 43d398ad5
> commit 43d398ad5d7b14fb344fc2e5338177761b9a199a
> Author: Elliott Hughes <enh at google.com>
> Date:   Thu Jul 18 15:52:55 2019 -0700
>
>     test.tar: fix tar tests on Android.
> ...
>     Fix the block/char special tests (the expectations were missing, but
>     they're never normally run on the host because the mknod fails).
>
>     Likewise add the missing expectation to the "ownership" test, which
>     isn't normally run on the host because the chown fails.
> ...
>     With this patch, all the tar tests now pass for me on both the device
>     and host.
>
> The test came from you, and the hash came from you, and I was wondering what had
> changed to produce a hatch that differed from the one you provided in the commit.
>
> Especially since:
>
> TAR='tar c --owner root --group sys --mtime @1234567890'
>
> And thus:
>
> skipnot chown nobody:nogroup dir/file 2>/dev/null
> testing "ownership" "$TAR dir/file | SUM 3" \
>   "2d7b96c7025987215f5a41f10eaa84311160afdb\n" ""
>
> In theory shouldn't care what user:group is on the file?
>
> It looks like I checked in an unfinished test (hence the blank output), but the
> REASON it was unfinished was a design issue, namely the $TAR macro disguises
> variable ownership so you can't test with it so I need to override the fields to
> provide a sstable tarball manually in this test, and I wasn't sure what they were.
>
> Expecially since the UID/GID variance on mac is badly defined. It says:
>
> $ grep nobody /etc/group
> nobody:*:-2:
>
> And in _theory_ that's getting typecast to (unsigned short) and thus trimmed to
> 65534 by the system somewhere, except on mac:
>
> $ echo '#include <sys/stat.h>' | cc -E - | grep gid_t
> typedef __uint32_t __darwin_gid_t;
> typedef __darwin_gid_t gid_t;
>
> If __uint32_t is 16 bits on MacOS it's... about what I expect there, really. And
> if not, then it's getting truncated in the kernel, or by the filesystem, or...
>
> Anyway, if this test doesn't work for you, and you submitted it, then me redoing
> it shouldn't break you. Modulo "does the result test the same on android" is
> something I can't answer until you try it and tell me. :)

as i said in a recent checkin comment, my number one mistake when
working on toybox is that i'm likely to _only_ test on the host, just
because it's more convenient, and then assume that the device is the
same (despite all the fairly obviously large differences). i wouldn't
be at all surprised if that's what happened here too.

let me know when you check something in that you want me to try... (i
say "me" but tbh, unless i'm actively debugging a device test failure,
it's most likely the CI bots doing all the work :-) )

> (I still want an AOSP "lunch shellprompt" target that builds _just_ a little
> qemu test system, but I'd still have to download and regularishly update a
> half-terabyte of AOSP source to build it...)
>
> Rob