"gfc_code2string(): Bad code" error for gfortran-4.3.2 on Darwin PPC

[Chris Talley]

Hi Tobias,

I followed your instruction but ran into a problem. The "gfortran-432  
-ffree-form -S -v dummy_mod.F" command returned:

  /usr/local/gcc432/libexec/gcc/powerpc-apple-darwin8.11.0/4.3.2/ 
f951 /var/tmp//ccxfGHJm.f -fPIC -quiet -dumpbase dummy_mod.F -mmacosx- 
version-min=10.4 -auxbase dummy_mod -version -ffree-form -o  
dummy_mod.s -fpreprocessed -fintrinsic-modules-path /usr/local/gcc432/ 
lib/gcc/powerpc-apple-darwin8.11.0/4.3.2/finclude

I then ran "gdb --args /usr/local/gcc432/libexec/gcc/powerpc-apple- 
darwin8.11.0/4.3.2/f951 /var/tmp//ccxfGHJm.f -fPIC -quiet -dumpbase  
dummy_mod.F -mmacosx-version-min=10.4 -auxbase dummy_mod -version - 
ffree-form -o dummy_mod.s -fpreprocessed -fintrinsic-modules-path / 
usr/local/gcc432/lib/gcc/powerpc-apple-darwin8.11.0/4.3.2/finclude"

After setting the breakpoint and typing 'run', I get the following  
output:

 >Starting program: /usr/local/gcc432/libexec/gcc/powerpc-apple- 
darwin8.11.0/4.3.2/f951 >/var/tmp//ccTmieJp.f -fPIC -quiet -dumpbase  
dummy_mod.F -mmacosx-version-min=10.4 -auxbase dummy_mod ->version - 
ffree-form -o dummy_mod.s -fpreprocessed -fintrinsic-modules-path >/ 
usr/local/gcc432/lib/gcc/powerpc-apple-darwin8.11.0/4.3.2/finclude
 >Reading symbols for shared libraries .++ done
 >GNU F95 (GCC) version 4.3.2 (powerpc-apple-darwin8.11.0)
 >        compiled by GNU C version 4.3.2, GMP version 4.2.2, MPFR  
version 2.3.1.
 >GGC heuristics: --param ggc-min-expand=100 --param ggc-min- 
heapsize=131072
 >Error: Can't open file '/var/tmp//ccTmieJp.f'
 ><built-in>:0: fatal error: can't open input file: /var/tmp//ccTmieJp.f
 >compilation terminated.
 >
 >Program exited with code 01.

After that I played around with the logical line in dummy_mod.F, and  
found that kind type of the logical must be the same as the kind type  
of the constant being assigned as follows:

    logical (kind=1) :: MEMCHEKDONE = .false._1

or

    logical :: MEMCHEKDONE = .false.

Both of these lines compile with 4.3.2, whereas the original

    logical (kind=1) :: MEMCHEKDONE = .false.

compiled with 4.2.4. Also, remember that the error only shows up with  
the mismatched logical kinds in a module being used by a module  
containing the 'CNT = count(A == 1)' line.

Could this be a case of a 1-byte logical being initialized by a 4- 
byte logical, resulting in memory corruption that only shows up when  
the 'CNT = count(A == 1)' line is present? Shouldn't gfortran auto  
convert the 4-byte logical in the 'dummy_mod.F' as it apparently did  
in 4.2.4?


Thanks,

Chris Talley
Senior Aerospace Engineer
CFD Research Corporation
215 Wynn Drive
Huntsville, AL 35805
256-726-4835
cst at cfdrc dot com

> Hi Chris,
>
> Chris Talley wrote:
> > By changing the line in dummy_mod.F from
> > > logical (kind=1) :: MEMCHEKDONE = .false.
> > to
> > > logical :: MEMCHEKDONE = .false.
> > the code compiles without an error.
> >
> > What is the difference between a 'logical (kind=1)' and a 'logical'?
>
> "logical" has the same size as an "integer", which is [usually] 4  
> bytes.
> "logical(kind=1)" has the same storage size as "integer(kind=1)",  
> namely
> 1 byte.
>
> If I recall correctly, "logical(kind=4)" is a tiny bit faster while
> "logical(kind=1)" uses less memory (a quarter). In most programs,
> neither the storage size for logical variables is negligible.
>
> (Unless your variable "MEMCHEKDONE" is in a COMMON block, changing the
> size should cause no algorithmic problems.)
>
>  * * *
>
> Coming back to your problem: Similar to Daniel, I cannot reproduce it
> on x86-64-linux with either gfortran 4.3 nor with today's 4.4.0.
>
> The "gfc_code2string()" function is used to read/write data from/to
> module files. At a glance I cannot see what could go wrong there.
>
> Can you run gfortran in a debugger and provide thus more information.
> Steps:
> 1. Run: "gfortran-432 -ffree-form -S -v dummy_mod.F"
>    and find in the output the line containing
>    .../4.3.2/f951 ...
> 2. Run: gdb --args .../4.3.2/f951 ...
> 3. Set a break point in "gdb" at gfc_code2string
> 4. Type "run"
>
> When you are in "gfc_code2string", can you print the value of "m- 
> >string"
> and of "code"? It would be interesting to know for which values it
> fails.
>
> Tobias