> On 2020-Apr-02, at 11:23, Louis Dionne <ldio...@apple.com> wrote:
> 
> 
> 
>> On Apr 2, 2020, at 14:05, Duncan Exon Smith <dexonsm...@apple.com 
>> <mailto:dexonsm...@apple.com>> wrote:
>> 
>>> It looks like the following bots assigned to you started failing recently:
>> 
>> I agree the bots should be fixed, but...
>> 
>>> I strongly suspect this is due to the fact that your builder is running as 
>>> root. Libc++'s filesystem tests are known to fail when run as root, because 
>>> some of them need to check for failure to access some files for which 
>>> there's no permission. This doesn't seem to work when run as root.
>> 
>> It seems like these tests could have a line like:
>> ```
>> REQUIRES: -user=root
>> ```
>> and libcxx's lit target could export a `user=root` feature when tests were 
>> being run as root.  WDYT?
>> 
>> WDYT?
>> 
> 
> I think you mean `// UNSUPPORTED: -user=root`? Yes, I think it would make 
> sense. Either that or actually spend the time to make them work as root.

I meant `REQUIRES: !user=root`, I used the wrong negation.

> 
> FWIW, I think the root cause of the issue is that "traverse" permissions in 
> directories are not honoured when run as root:
> 
>     root $ d=$(mktemp -d)
>     root $ touch $d/foo
>     root $ chmod 000 $d
>     root $ stat $d/foo
>     16777221 234259215 -rw-r--r-- 1 root wheel 0 0 <...>
> 
> When I would expect:
> 
>     ldionne $ d=$(mktemp -d)
>     ldionne $ touch $d/foo
>     ldionne $ chmod 000 $d
>     ldionne $ stat $d/foo
>     stat: <...>/tmp.JoY6Ixmo/foo: stat: Permission denied
> 
> Does that make sense? I didn't know 'root' had a free pass like that.
> 
> Louis

_______________________________________________
linaro-toolchain mailing list
linaro-toolchain@lists.linaro.org
https://lists.linaro.org/mailman/listinfo/linaro-toolchain

Reply via email to