Re: [libtool 2.1a] testsuite: 5 9 15 failed

[Eric Blake]

> Hehe, failure of the java one is new.  I believe it's due to my
> virtually nonexistant knowledge of the language.  :)
> 
> Can anybody take a look at the source I put in libtool HEAD
> tests/convenience.at whether it constitutes valid Java?
> Alternatively, a pointer to some language docs might help as well.

It is valid Java (having been a former contributor to the jikes compiler
for Java), although naming a method A when the class is named A is
not normal practice (if you intended for those to be constructors, get
rid of the void keyword in [ABC].java).  Unfortunately, I don't have
much gcj experience (partly because gcj on cygwin is always lagging
in support), so I don't know how well gcj works at compiling Java files
to .o or pulling in libraries.  I do note that all of your java classes are
defined to be in the anonymous package, and that the anonymous
package tends to include only the current directory, so if libtool is
moving files around to temporary directories before compilation it may
be breaking the ability of the compiler to see all other classes in the
anonymous package.

> 
> The other two failures are specific to the win32 environment.
> 5 is a bug in the test, I believe, I've seen it before, but was unsure
> whether to just skip the test on win32.
> 15 is a genuine bug in ltmain which lingered there for the last 5 years.
> On a cold and boring day I might try to fix it, but noone has
> encountered in practice yet.  :)
> (actually I hope the bug reports provide an incentive to fix it.)

OK, that's nice to know.

> 
> Thanks a lot for reporting these!

The reason I ask is that I would like to get a successful compile of m4
from CVS head on cygwin (and that currently depends on CVS autoconf,
automake, and libtool), and I have still never succeeded in that.  It always
appears to be a libtool bug breaking the compile, so I first want to make
sure libtool is doing the right things.

--
Eric Blake