[Toybox] [PATCH] Try to debug modinfo test flake.

[enh]

On Mon, Jun 12, 2023 at 12:41 PM Rob Landley <rob at landley.net> wrote:

> On 6/12/23 09:27, enh wrote:
> > On Sat, Jun 10, 2023 at 12:58 PM Rob Landley <rob at landley.net
> > <mailto:rob at landley.net>> wrote:
> >
> >     On 6/9/23 18:42, enh wrote:
> >     >     >     Try commit a5b5749cc7f0
> >     >     >
> >     >     > that actually seems to have _broken_ things more reliably?
> looking
> >     at your
> >     >     > change, it seems like you forgot to actually touch the -F
> test?
> >     >
> >     >     Sigh, I spent so much effort convincing myself I didn't need
> to quote
> >     stuff I
> >     >     missed the case where I did. :P
> >     >
> >     >     Try commit 87abbcad8f8c
> >     >
> >     > (yes, that works [on Android and Debian both]. thanks!)
> >
> >     Which I believe means the test is finding no modules, which raises
> questions
> >     about the test environment...
> >
> > yeah, let me know if you want to try to debug that further (or just
> check in
> > test changes that will blow up with useful diagnostics),
>
> It's basically "find /dir -name '*.ko'" produces results, or it does not.
> Not a
> huge amount of subtlety there. Since it's no longer trying to discard
> stderr
> then if nothing is written to stderr, it most likely means there's nothing
> in
> that directory. (Even the directory not being there or not being readable
> should
> produce an error to stderr, which would show up in the log...)
>
> Which means it's a VM setup issue, which is not really my area. :)
>

yeah, me neither. i suppose i should +Alistair Delva <adelva at google.com> in
case he's interested (internal bug for this was
https://b.corp.google.com/issues/279863718). but the 0.08% failure rate
makes me think this clearly isn't a normal state for those VMs to be in!


> (Mac not having any linux kernel modules is... unsurprising.)
>
> > but personally i'm
> > prepared to stick my head in the sand where virtualized servers are
> concerned.
> > (though obviously now we have the get out clause, we wouldn't notice if
> we
> > started skipping all these tests in configurations we care about too.)
>
> I need to circle back to getting the tests to run in mkroot, then at least
> I
> should be testing them. In which context "skipped" is similar to an error,
> modulo the "this is only tested on glibc" and so on ones. Finite
> whitelist...
>
> Rob
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.landley.net/pipermail/toybox-landley.net/attachments/20230612/70552669/attachment.htm>