Re: [Sbcl-devel] Test Report, 1.0.37.59, Linux/ppc and Linux/x86-64

[Tobias C. R. <tc...@fr...>]

Alastair Bridgewater <ala...@gm...> writes:

> Hello all,
>
> I recently procured a PPC G5 system (now running linux), and thought
> "why not try SBCL on this thing?" This led to a few things, including
> trouble running the test suite both on PPC and on x86-64.
>
> Three tests hung on Linux/ppc, one on Linux/x86-64. A patch
> conditioning all three tests out (for locating the failing tests, not
> for application to CVS) is below.
>
> A few things come to mind:
>
>   1. The dynamic-extent tests are partially conditioned on
> :STACK-ALLOCATE-WHATEVER and related features, but this does not seem
> to  be consistent.
>
>   2. The test suite should probably be audited for the presence of
> bare assert forms (outside of a with-test form), and such asserts be
> wrapped.
>
>   3. There are sufficiently few x86-64 failures that it may be
> possible to have a completely clean test result with a bit of work (it
> should at least be possible to reduce it to two expected failures).
>
>   4. I should build and run the test suite more often.
>
> That's all I have at this point, though I may act on some of this over
> the next week.

A major gripe of the test suite is that it's not easily loadable into a
running sbcl image. That means no M-., no arglist display in Slime,
can't interactively test new test definitions, and so on.

I wonder if we couldn't extend sb-rt to include test-related utility
functions, and use SB-RT for the tests/, making sure that each .lisp
file starts with a proper in-package etc.

  -T.