Re: versioning test

[Ralf Wildenhues]

Hi Bob,

* Bob Friesenhahn wrote on Sat, Jan 24, 2009 at 05:18:32PM CET:
> On Sat, 24 Jan 2009, Ralf Wildenhues wrote:
>
>> Here is a testsuite addition to get some exposure to versioning.
>> OK to push (and add Mike to THANKS)?  It'd be good if somebody
>> proof-read it so there are no silly typos or thinkos.
>
> I read it and did not see any silly typos or thinkos, but then again it 
> is not easy reading.  It should be interesting to see where these tests 
> report a failure.

Well, I did test that setting $version_type to "qnx" on my GNU/Linux
system would cause the failure that Mike reported, so one can expect
they are not completely useless, e.g., for porting to a new platform,
or for ensuring that future changes to ltmain will not break documented
semantics.

OTOH, the tests need a gloss over.  Mike's testing already exposed one
bug in them, in that one cannot assume the major version of a library is
$current - $age, it might just be $current (plus or minus one or so).

Another bit, my test for this hypothesis:

|    If two libraries have identical CURRENT and AGE numbers, then the
| dynamic linker chooses the library with the greater REVISION number.

is too weak for what I thought of.  On GNU/Linux, you can install two
libraries differing only in revision, and `ldconfig -n $libdir' will
take care to let the respective versioned symlink point to the newer
revision.  Cool, huh?  I didn't even know that, but I think that's what
the sentence above tried to imply.  Problem of course is, this surely
doesn't happen for systems that don't encode a revision into the name,
and may not either hold for some that do.  I'd have to test, and will
try to come up with a better patch (and documentation fix) for this
hypothesis then.

Thanks for the review.

Cheers,
Ralf