[opam-devel] [MirageOS-devel] opam 1.1.1 and aspcud 1.9

[David Allsopp]

Louis Gesbert wrote:
> For this specific bug,running "make tests" would have been enough
> (I just tested on my VM). For being defensive and having ./configure
> refuse newer version of dependencies, the m4 macros don't seem to
> provide version filtering (only API testing. Here the API was
> compatible, anyway that would have failed at build-time). Does
> ocamlfind even provide an easy way to get the version of a package?

Version filtering isn't what I meant, though - I meant semantic filtering, i.e. a relevant test case which fails for this "broken" version but works in the older versions (or corrected newer versions). IMO, version filtering is something packagers do with package meta-data; developers should use version filtering only because of API changes which isn't applicable here.

> But my feeling is that what happened really shouldn't happen -- I
> mean, a small, but API compatible change, that would cause the older
> version to break _but not the newer version, just by chance_, so 
> that it could get unnoticed ? What are the odds of this happening 
> _again_ ? ;)

Again, in my opinion, "once bitten, twice shy". My own personal experience is that if something like this has happened it is *very* likely it will happen again (and that's in no way a negative comment on any developers concerned!)

There are plenty of other instances in configure scripts where semantics are tested rather than a version compatibility matrix (usually when probing the C compiler) so I don't think it's too unusual a thing to want to do?


David