Re: name.lib vs. -lname

[Ralf Wildenhues]

Hi Peter,

* Peter Ekberg wrote on Tue, Feb 14, 2006 at 11:38:28AM CET:
> On Sun, Feb 12, 2006 at 11:17:04AM +0100, Ralf Wildenhues wrote:
> > | @@ -4082,6 +4127,17 @@
> > |         *) tmp_libs="$tmp_libs $deplib" ;;
> > |         esac
> > |         ;;
> > | +     -l*)
> > | +       if test "X$with_gcc" != Xyes; then
> > | +         case $host_os/$linkmode in
> > | +         mingw*/prog | cygwin*/prog)
> > | +           # assume MSVC
> > | +           deplib=`$ECHO "X${deplib}.lib" | $Xsed -e 's/^-l//'`
> > 
> > See above (func_stripname).  I still don't like this `lib' special
> > casing at all.  It should equally apply to GCC.  Also with_gcc != yes
> > should not imply MSVC.
> > 
> > I also believe this is not the right place to fix this at all.  It
> > should be done at the start of the deplib search algorithm.  I am
> > very reluctant to change that part before 2.0.
> 
> The rationale is to keep the -lname form as long as possible so
> that libtool sees normal stuff as long as possible. If you do
> the conversion to name.lib earlier, you have to add handling
> for that form in other places where the -lname form is
> currently expected. It quickly became a mess when I tried it.
> That said, I might have started at the wrong spot and that was
> the cause of the mess, as my libtool-fu is weakish.

Yeah, that part of the code is the biggest mess.
It's possibly a good spot, or even needs to be done later.
The testsuite will be much better soon to be able to expose
it when it's wrong.  I'm reluctant to change the code before that.

> Heck, I'm not even sure where "the start of the deplib search"
> is. Yes, I can take a guess, but I'm really mostly fumbling.

Exactly.  My point again: libtool behavior is well defined only
when we have mostly exhaustive test suite.  Then you can go and
change and see when it does the right thing.

Yes, actually restructuring so that it's more readable would be
good, too.  :)  Actually, that restructuring is much helped by
testsuite exposure, so I think the order of doing the work is ok.

> But the comment about gcc is wrong, I think. There are no
> libraries named name.lib when you are using gcc, and it expects
> -lname so there is no need to mangle the deplib. Or what did you
> mean?

Can you create a library with MSVC/link and use it with GCC/ld?
How about the other way round?  If it has C interfaces only?

Cheers,
Ralf