VMD-L Mailing List

From: John Duff (jfduff_at_mtu.edu)
Date: Sun Oct 08 2000 - 14:52:46 CDT

Justin,

I tried your suggestion without any luck. I played around with it for
a while, and couldn't figure out a workaround with #defining
or #undefining things (such as __cplusplus or __THROW). I checked out the
docs for lex/flex, and there's an option to compile for C++ (either you
use "flex++" or "lex/flex -+"). I tried that too, and got a bunch of
new errors and still saw the isatty error. So perhaps the bug ultimately
resides with lex/flex, where it needs to be patched to handle the new
prototyping in /usr/include/unistd.h. In any case, I thought of a different
workaround for vmd. In the Makefile, change the rule for compiling AtomLexer.l:

<from>
.l.o:
       $(ECHO) "Converting " $< " --> "$@ " ..."; \
       $(LEX) $<
       $(MOVE) lex.yy.c lex.yy.C
       $(CCPP) $(CPPFLAGS) -I../src -c lex.yy.C \
               -o ../LINUX/$@
       $(DELETE) lex.yy.C

<to>
.l.o:
        $(ECHO) "Converting " $< " --> "$@ " ..."; \
        $(LEX) $<
        if [[ `grep isatty /usr/include/unistd.h | grep __THROW` ]] ; then \
          echo "Modifying lex.yy.c"; \
          sed 's/extern int isatty YY_PROTO(( int ));/extern int isatty YY_PROTO(( int )) __THROW;/' lex.yy.c > lex.yy.C; \
          $(DELETE) lex.yy.c; \
        else \
          $(MOVE) lex.yy.c lex.yy.C; \
        fi;
        $(CCPP) $(CPPFLAGS) -I../src -c lex.yy.C \
                -o ../LINUX/$@
        $(DELETE) lex.yy.C

This should work for linux installations that have the new gcc/g++, without
breaking those with the old gcc/g++. If you see a problem with this, or
think of a better workaround, I'd be curious to know.

-John

>
> Hi,
>
> I ran into this problem with the new gcc/g++ on RH7, but I had so many
> other problems with the rest of the distribution that I didn't pursue
> it ;->
>
> I would guess that the reason VMD dies right away is not due to any
> Xfree/DRI problems, but incompatibilities with libstd++. Could be wrong...
> In any case, the way we do lex.yy.cc is kinda hokey, but since not just
> Inform but three or four other C++ classes are 'exposed' to AtomLexer,
> it would be a bit of work to make AtomLexer pure C.
>
> Here's an idea, I have no idea if this would work or not, but what if
> you put
>
> #ifndef __cplusplus
> #define __cplusplus
> #endif
>
> at the top of AtomLexer.l? If you could try it and let us know how it
> goes, that would be cool.
>
> Thanks for the well-researched note.
>
> Justin
>
>
>
>
> On Sat, Oct 07, 2000 at 01:57:48AM -0400, John Duff wrote:
> > Hello,
> >
> > I've run into a bug when trying to compile vmd using redhat 7.0 (i386).
> >
> > System
> > ------
> > 800MHz Athlon TBird (Abit KT7 mobo)
> > 256M Atlas Precision RAM
> > Creative Labs 3D Blaster Annihilator 2 vid card (NVIDIA GeForce 2 GTS 32M DDR)
> > RedHat 7.0 (2.2.16-22 kernel)
> > XFree86 4.0.1
> > NVIDIA linux 0.9.5 drivers
> >
> >
> > Synopsis
> > --------
> > After downloading the precompiled vmd1.6a3 Linux DRI-OpenGL package, I
> > configured and installed it, but when I ran it, it immediately died after
> > very briefly popping up the console window. This seems to be a known problem
> > (http://www.ks.uiuc.edu/Research/vmd/mailing_list/vmd-l/0093.html). I tried
> > fixing up the links to libGL.so like the mailing list email suggested, but
> > still it didn't work. So I downloaded the source and when I tried compiling
> > I got the following error message:
> >
> > Converting AtomLexer.l --> AtomLexer.o ...
> > lex.yy.c:1659: declaration of `int isatty (int)' throws different
> > exceptions
> > /usr/include/unistd.h:670: than previous declaration `int isatty (int)
> > throw ()'
> > make: *** [AtomLexer.o] Error 1
> >
> >
> > After investigating this, I found the following (quick and dirty) workaround.
> > Modify line 670 in /usr/include/unistd.h:
> >
> > <from>
> > extern int isatty (int __fd); __THROW;
> > <to>
> > extern int isatty (int __fd);
> >
> >
> > After I did this, vmd compiled smoothly and started up without any problems
> > after doing a "make install". Note that after compiling vmd, you should
> > undo the change to line 670 in unistd.h, as this is not a bug in unistd.h.
> >
> >
> > More Details
> > ------------
> > Obviously the fix is a quick hack just to enable vmd to compile.
> >
> > /usr/include/unistd.h is in the glibc-devel-2.1.92-14 package. It looks
> > like all the function prototypes in unistd.h now have "__THROW" appended
> > to them, which is something new in this file, different from the version in
> > the package found in RedHat 6.2 (glibc-devel-2.1.3-15). Furthermore, __THROW
> > is defined in /usr/include/malloc.h, and it resolves to "throw" only if you're
> > compiling C++ programs, which makes sense. VMD's Makefile first uses lex
> > to compile the AtomLexer.l file to lex.yy.c, renames lex.yy.c to lex.yy.C,
> > and then uses g++ to compile lex.yy.C. That looks like the source of the
> > problem, since the use of g++ causes __THROW to be "throw" instead of "" in
> > the prototype of isatty (line 670 /usr/include/unistd.h), which is different
> > than the C prototype of isatty in lex.yy.c (line 1659). I tried fixing
> > the Makefile to use gcc instead of g++ to compile lex.yy.c, but it looks like
> > AtomLexer.l includes Inform.h, which is a C++ header file (I guess that's
> > why you used g++ to compile it). So, if all this is accurate, it looks like
> > you need to fix AtomLexer.l to be able to be compiled using a C compiler,
> > not C++.
> >
> >
> > -John
> > jfduff_at_mtu.edu
>
> --
>
> Justin Gullingsrud 3111 Beckman Institute
> H: (217) 384-4220 I got a million ideas that I ain't even rocked yet...
> W: (217) 244-8946 -- Mike D
>