[darcs-users] Crazy idea: get depends from tests

Trent W. Buck twb at cybersource.com.au
Fri Sep 11 01:59:23 UTC 2009


Florent Becker <florent.becker at ens-lyon.org> writes:

> The biggest limitation of darcs' patch theory is that the commutation
> between patches is not defined semantically, but only
> syntactically. This is because any semantical treatment of patches is
> undecidable. We partially deal with that undecidability by using
> tests.
>
> That's why I propose to add a --add-deps-by-tests flag to record (and
> amend- record). When recording a patch x with that flag, darcs would
> try to commute each patch between x and the last tag (or some other
> chosen patch) past x, then unpull it and run tests. If the tests fail,
> then x depends on that patch. While it would require a not-too slow
> testsuite, that system would be a real robustness argument for darcs'
> cherry-picking: whenever you do cherry-picking, you end up with a
> merge of to branches which pass the tests (even though the merge
> itself could well not pass).
>
> What do you think?

Would this complete in a reasonable time, on reasonable hardware, given
a reasonably comprehensive set of tests?  For Darcs, for example, you're
talking about recompiling Darcs and running the tests/ scripts for every
commutation.

While I like the idea in principle, I can't see how I could have both
this feature AND comprehensive testing, so I'd pick the latter.



More information about the darcs-users mailing list