Re: [PATCH 0/6] KUnit integration for Rust doctests

[David Gow]

On Thu, 15 Jun 2023 at 16:20, Miguel Ojeda
<miguel.ojeda.sandonis@xxxxxxxxx> wrote:
>
> On Thu, Jun 15, 2023 at 3:44 AM Boqun Feng <boqun.feng@xxxxxxxxx> wrote:
> >
> > Great work! I've played this for a while, and it's really useful ;-)
>
> Thanks!
>
> > The assertion warning only says line 35 but which file? Yes, the
> > ".._sync_lock_spinlock_rs" name does provide the lead, however since we
> > generate the test code, so we actually know the line # for each real
> > test body, so I come up a way to give us the following:
> >
> >         [..] # rust_doctest_kernel_sync_lock_spinlock_rs_0: ASSERTION FAILED at rust/kernel/sync/lock/spinlock.rs:61
> >         [..] Expected e.c == 11 to be true, but is false
> >         [..] [FAILED] rust_doctest_kernel_sync_lock_spinlock_rs_0
> >
> > Thoughts?

I like it.

A part of me would like to keep the
kernel::kunit::info(format_args!("    # Doctest from line {line}\n"));

If only so we can preserve the line information when the tests
actually pass. This is something that'd probably ultimately fit as a
"test attribute":
https://lore.kernel.org/linux-kselftest/20230610005149.1145665-1-rmoar@xxxxxxxxxx/

For C tests we're not bothering outputting line numbers now (though
again, are considering if we can do it with an attribute), but those
tests have much more searchable names, so I think it still makes sense
for the Rust ones.

How about printing something like:
    # source_line: {}

kunit.py will hide this when the test passes unless the user
explicitly passes --raw_output.

>
> Sounds good to me. However, David/Philip, is this OK or do you really
> need/use the actual/compiled source file there? If you don't need it,
> does it need to exist / be a real file at least? If the latter answer
> is a "yes", which I guess it may be likely, then:

I don't think there's anything automated using the file:line in
assertion messages, it's just for human consumption.
This may change in the future (and there are probably some text
editors around which will turn a path like that into, e.g., a
clickable link now), but we're not currently doing anything which
would actually open the file.

>
> > +        let src_file = format!("rust/kernel/{}", file.replace("_rs", ".rs").replace("_", "/"));
>
> This would not work for files with a `_` in their name, like
> `locked_by.rs`. I guess we could still find the real filename based on
> that information walking the dir, which is another hack I recall
> considering at some point.
>
> Otherwise, if "fake" filenames in the line above are OK for
> David/Philip (I suspect they may want to open them for reporting?),
> then I guess the `file` one may be good enough and eventually we
> should get `rustdoc` to give us the proper metadata anyway.

Personally, I'd think a "fake filename" is okay (especially if it's
temporary until we can get the right one), though I'd prefer there to
be some indication that it's "fake": maybe leaving the _ separator in,
or wrapping it in brackets, or something? Unless the whole
walk-the-filesystem technique ends up being worth doing, so we don't
have a significant chance of the filename being dud.

-- David

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature