[imp-hackers] Another topic: ANSI standard test suite

[Sam Steingold]

On Thu, Jun 25, 2009 at 9:42 AM, Ville
Voutilainen<ville.voutilainen at gmail.com> wrote:
> The ansi-test repository is not a per-implementation fork, or at least
> is not meant to be
> one. We can just as well use that as a "common fork", if that's fine
> with other people.
> In addition to removal of incorrect tests, the ansi-test repo merely
> contains small changes
> that allow the suite to run on abcl, in addition to everything else it
> already runs on.

this is not good enough.
the fact that the tests reside in the abcl's repo means that people
would have to think "how is this related to abcl? is the gcl's repo
better?" &c.

As I said, I think there should be a separate project and gcl and abcl
should explicitly refer to it (instead of copying it).
E.g., clisp has a target in Makefile.devel which downloads the
ansi-test dir from gcl.

Moreover, since this is an ANSI test suite, no #+clisp/#+gcl/#+abcl
should ever be present there.
if several results are permitted by the standard, each should be
accepted from any implementation.

e.g., suppose ANSI said that (foo) could return 16, 42 or 17, but never 12.
then the test should look like this:

(member (foo) '(16 42 17)) ==> T

and not

(foo) ==>
#+good-implementation-1 16
#+good-implementation-2 42
#+good-implementation-3 17
#+bad-implementation-1 12


-- 
Sam Steingold <http://sds.podval.org>