gfortran 4.1.0-cvs cygwin build failure: ICE on environ.c (!)

[Tobias Schlüter]

Brooks Moses wrote:
> Ok, so.  Step 1: I fire up cvs, and check out the source as per the
> instructions on the Wiki.  All goes well.  The system that I'm using is
> my venerable home work machine, a 200MHz Pentium Pro running Win2k and
> Cygwin, with all the Cygwin stuff updated about two or three months
> ago.  Since the website (I don't remember where) mentioned depending on
> bison, I installed that from Cygwin before starting.

There should be no dependency on bison, nor on flex.  Probably, you should run
contrib/gcc_update from the root directory of your tree to make sure that cvs
didn't mess up the file dates.

> Step 2: Set up environment variables and run configure, as per a recent
> post in comp.lang.fortran (with the options taken from the post):
> 
>   setenv CFLAGS '-O2 -g'
>   setenv BOOT_CFLAGS '-O2'
>   cd gcc/gcc
> 
>   ../configure  --enable-threads=posix 
>       --enable-sjlj-exceptions --disable-libmudflap
>       --enable-languages=f95  
> 
> This, again, works fine.  (Well, I think it gave me an error about a
> missing library the first time, so I installed that via Cygwin setup,
> and then it ran fine.)

That looks like a weird way to build, even though it seems to work in
principle according to oyur description below.  The common way would be to do
'cd gcc; mkdir build; cd build; ../configure [insert options]'

> Step 3: run "make bootstrap-lean".
> 
> This failed, an hour or so in, due to my not having flex installed. 
> Since the error message mentioned that I shouldn't need flex installed
> unless I modified a .l file, I found this a bit odd, but I went and
> installed flex.
> 
> (Should I report that as a bug?  I saved the build output from around
> when it was doing that.)

I assume that somehow file dates were messed up in your checkout, I can't
think of another way how this could be necessary.

> So, I installed flex, and ran "make bootstrap-lean" again, and let
> things churn.
> 
> Twenty-six hours later, after getting through the compilations in the
> "stage 1" and "stage 2" directories, it crashed with the following
> output:
> 
> 
>>/cygdrive/e/gfortran/gcc/gcc/./gcc/xgcc -B/cygdrive/e/gfortran/gcc/gcc/./gcc/ -B/usr/local/i686-pc-c
>>ygwin/bin/ -B/usr/local/i686-pc-cygwin/lib/ -isystem /usr/local/i686-pc-cygwin/include -isystem /usr
>>/local/i686-pc-cygwin/sys-include -DHAVE_CONFIG_H -I. -I../../../libgfortran -I. -iquote../../../lib
>>gfortran/io -std=gnu99 -Wall -Wstrict-prototypes -Wmissing-prototypes -Wold-style-definition -Wextra
>> -Wwrite-strings -O2 -g -O2 -c ../../../libgfortran/runtime/environ.c -o environ.o
>>../../../libgfortran/runtime/environ.c: In function 'init_integer':
>>../../../libgfortran/runtime/environ.c:104: error: invariant not recomputed when ADDR_EXPR changed
>>&_ctype_D.1970[1];
>>
>>../../../libgfortran/runtime/environ.c:104: internal compiler error: verify_stmts failed.
>>Please submit a full bug report,
>>with preprocessed source if appropriate.
>>See <URL:http://gcc.gnu.org/bugs.html> for instructions.
>>make[3]: *** [environ.lo] Error 1
>>make[3]: Leaving directory `/cygdrive/e/gfortran/gcc/gcc/i686-pc-cygwin/libgfortran'
>>make[2]: *** [all] Error 2
>>make[2]: Leaving directory `/cygdrive/e/gfortran/gcc/gcc/i686-pc-cygwin/libgfortran'
>>make[1]: *** [all-target-libgfortran] Error 2
>>make[1]: Leaving directory `/cygdrive/e/gfortran/gcc/gcc'
>>make: *** [bootstrap-lean] Error 2

I'm sorry to tell you that this is a known bug which has been open since May
28th, see <http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21766>.  I have no idea
why it hasn't been resolved yet even though a patch for it has been posted
here: <http://gcc.gnu.org/ml/gcc-patches/2005-05/msg02543.html>.

> Unfortunately, it appears that the "make" process is now rather deeply
> wedged.  The first run-through had apparently just finished the "stage
> 2" build and cleared everything out of those directories before it
> crashed.  However, when I restart the process, it tries to run tail on
> all of the object files in the stage2 directories, which are now empty,
> and so it fails.
> 
> Thus, my questions: ought I to file the bug report about the build
> failure?  If so, what do I need to include in it to properly illustrate
> the problem?  And what can I do to unwedge my build process (preferably
> without having to redo the whole 26 hours of compilation!) once that's
> dealt with?

I'm not sure, I never use bootstrap-lean.  Maybe running 'make bubblestrap' or
'make restage2' will do what you need, but I don't know.  Maybe a bug report
on the failure when trying to run 'make bootstrap-lean' a second time would be
in place.  (I understand that this is a little late to say, but if you have
enough hard disk space, bootstrap-lean means no gain, it only makes rebuilding
harder.  OTOH, using 'make' instead of 'make bootstrap' will save you a lot of
time, and since gfortran is compiled with the gcc compiled during the build
process the whole bootstrap serves no particular purpose.  If you're modifying
the compiler it's a way of stress testing it that's basically why we do it)

HTH
- Tobi

- Tobi