Re: [openvrml-develop] compiling 0.12.2 on Slackware 8.0

[Greg R. <ne...@po...>]

Braden,

>> (2) The configure test for -lglut apparently assumes libglut.so is "hard
>>     linked" against X11 libraries (very bad practice, but that's another
>>     topic); this is not the case on my box.  This patch to configure itself
>>     suffices, but I suppose configure.ac or aclocal.m4 or something is
>>     what really should be fixed:

> I believe this has already been addressed in CVS (on the 0.12 branch).

OK.  However it gets fixed, it needs to propagate to the lookat makefile,
too; I eventually got the library and app compiled, but lookat failed on
the link step.  (Same fix as to the configure script.)

> I'm not familiar with -pipe; but if it's just a flag set for the
> compiler, users who need it can add it to CFLAGS when running configure.

True, but it's not well known, and this is the only package I've ever
compiled that requires so much memory and/or /tmp space.

>> (4) After all that, a continued "make" promptly blew away the results of
>>     previous half hour and tried the same doomed command yet again.  I'll
>>     have to change the ownership of the stupid file to root or hard-link
>>     it somewhere or something.  I have no clue why this happened...

> Me either... But then I'm not 100% clear on what happened.

libtool stupidity.  I just copied the actual g++ step, which puts the
object file in .libs, but the makefile dependency was on the version 
in the main directory (which apparently is where libtool puts it after
it's successfully built).  I don't know why libtool uses hidden directories
and a two-step compilation/copy, but it was a painful approach this time.

> That would have the side effect of not squawking very loudly when the
> package to support a feature desired by the user can't be found. I'm not
> sure that's desirable. Inspection of "configure --help" tells users what
> OpenVRML will look for, which should give them an idea of what they
> should be configuring --without.

But the README's description can be misinterpreted to mean that the
extra feature will be used if found, but not otherwise.  (At least,
that's how I read it.)  And other configure scripts succeed in similar
cases (e.g., OpenSSH's --with-md5-passwords -- either way compiles,
though only one may work).  Perhaps it's simply a documentation issue.

> Thanks for the feedback!

You bet!  Very nice to have a stable background node. :-)

Greg