Error in egcs FORTRAN I/O?

Error in egcs FORTRAN I/O?

[Paul Bunyk]

Hi! 

I'm trying to use egcs 1.0 to compile our simulation program (mix of
FORTRAN and C). Everything went really great (and performance jumped
20% up in comparison to Absoft F77) until I run into an I/O error
(reproducible). I'm trying to execute the statement
	READ(10,END=100,FMT='(I4)') NERR1
and it fails (with run-time error) in the middle of the file. 'errno' is
set to 29 (#define ESPIPE          29      /* Illegal seek */) and I
suspect that it happens right on the 4-K block boundary. 

Another interesting observation is that in the following code from
'do_fio' which triggers the error (at line 472), the value of n is
exactly 0 (or my gdb does not show it correctly?).

(gdb) frame 7
#7  0x8069d4a in do_fio (number=0x81600bc, ptr=0x81d4b6c "", len=4)
    at /tmp/egcs-1.0/gcc/f/runtime/libI77/fmt.c:472
472                             errfl(f__elist->cierr,errno,"fmt");
(gdb) list
467                     if(ptr==NULL)
468                             return((*f__doend)());
469                     f__cnt[f__cp]--;
470                     f__workdone=1;
471                     if((n=(*f__doed)(p,ptr,len))>0)
472                             errfl(f__elist->cierr,errno,"fmt");
473                     if(n<0)
474                             err(f__elist->ciend,(EOF),"fmt");
475                     continue;
476             case STACK:
(gdb) p n
$39 = 0
(gdb) p errno
$40 = 29

I'm using 
	egcs-2.90.21 971202 (egcs-1.00 release) 
RedHat 5.0 on an i386 
	Linux version 2.0.32 (root@porky.redhat.com) 
	(gcc version 2.7.2.3) #1 Wed Nov 19 00:46:45 EST 1997)

Any ideas/suggestions what I can try to do next? I have the session
running in background so I can dig deeper if I'd know where to look...

Paul

-- 
  ("`-''-/").___..--''"`-._   UNIX *is* user-friendly, he is just very 
   `6_ 6  )   `-.  (     ).`-.__.`) picky about who his friends are...
   (_Y_.)'  ._   )  `._ `. ``-..-'      Paul Bunyk, Research Scientist
 _..`--'_..-_/  /--'_.' ,'art by           (and part-time UN*X sysadm)
(il),-''  (li),'  ((!.-' F. Lee http://pbunyk.physics.sunysb.edu/~paul

Re: Error in egcs FORTRAN I/O?

[Toon Moene]

Paul,

>  I'm trying to use egcs 1.0 to compile our simulation
>  program (mix of FORTRAN and C). Everything went really
>  great (and performance jumped 20% up in comparison to
>  Absoft F77) until I run into an I/O error (reproducible).
>  I'm trying to execute the statement
>  	READ(10,END=100,FMT='(I4)') NERR1 and it fails
>  (with run-time error) in the middle of the file. 'errno'
>  is set to 29 (#define ESPIPE          29      /* Illegal
>  seek */) and I suspect that it happens right on the 4-K
>  block boundary.

[ On Red Hat 5.0-Intel ]

This suggests that it is a library problem (it could be libf2c.a  
that is at fault, or an underlying library - such as the libc  
library on your OS).  Therefore, the following questions are  
relevant:

    * The libc version on your system (find out by "echo /lib/libc.so.*")

    * The binutils version on your system (find out with "as -v" or  
"ld -v")

    * The version of ldd/ld.so on your system (find out with "ldd -v")

(Copied straight from the pre-release questionnaire ...)

A complete Fortran program showing the problem would be nice too.

Thanks,
Toon.

Re: Error in egcs FORTRAN I/O?

[Paul Bunyk]

Toon Moene writes:
 > Paul,
 > 
 > >  I'm trying to use egcs 1.0 to compile our simulation
 > >  program (mix of FORTRAN and C). Everything went really
 > >  great (and performance jumped 20% up in comparison to
 > >  Absoft F77) until I run into an I/O error (reproducible).
 > >  I'm trying to execute the statement
 > >  	READ(10,END=100,FMT='(I4)') NERR1 and it fails
 > >  (with run-time error) in the middle of the file. 'errno'
 > >  is set to 29 (#define ESPIPE          29      /* Illegal
 > >  seek */) and I suspect that it happens right on the 4-K
 > >  block boundary.
 > 
 > [ On Red Hat 5.0-Intel ]
 > 
 > This suggests that it is a library problem (it could be libf2c.a  
 > that is at fault, or an underlying library - such as the libc  
 > library on your OS).  Therefore, the following questions are  
 > relevant:
 > 
 >     * The libc version on your system (find out by "echo /lib/libc.so.*")

RedHat now packages Linux with new GNU libc, that might be the source
of this problem:
	lib/libc.so.6 -> libc-2.0.5.so

 > 
 >     * The binutils version on your system (find out with "as -v" or  
 > "ld -v")

pbunyk:~> as -v
GNU assembler version 2.8.1 (i686-pc-linux-gnu), using BFD version linux-2.8.1.0.1
pbunyk:~> ld -v
GNU ld version 2.8.1 (with BFD linux-2.8.1.0.1)

 > 
 >     * The version of ldd/ld.so on your system (find out with "ldd -v")

pbunyk:~> ldd --version 
ldd (GNU libc) 2.0.5

 > 
 > (Copied straight from the pre-release questionnaire ...)
 > 
 > A complete Fortran program showing the problem would be nice too.

I would send it, but it is rather large and not that easy to build;
moreover, it takes probably an hour of runtime to get to this
problem. This is why I'm suggesting to look into the details
myself... Or maybe someone have encouuntered a similar problem and
patch already exists...

 > 
 > Thanks,
 > Toon.

Thank you,

Paul

-- 
  ("`-''-/").___..--''"`-._   UNIX *is* user-friendly, he is just very 
   `6_ 6  )   `-.  (     ).`-.__.`) picky about who his friends are...
   (_Y_.)'  ._   )  `._ `. ``-..-'      Paul Bunyk, Research Scientist
 _..`--'_..-_/  /--'_.' ,'art by           (and part-time UN*X sysadm)
(il),-''  (li),'  ((!.-' F. Lee http://pbunyk.physics.sunysb.edu/~paul

Re: Error in egcs FORTRAN I/O?

[Toon Moene]

I asked:

 > A complete Fortran program showing the problem would be nice too.

And you replied:

>  I would send it, but it is rather large and not that easy
>  to build; moreover, it takes probably an hour of runtime
>  to get to this problem. This is why I'm suggesting to
>  look into the details myself... Or maybe someone have
>  encouuntered a similar problem and patch already exists...

Hmm, yeah, I see - that makes my previous questions also rather  
unhelpfull, as we can't rerun your program that way.

Perhaps some other remarks and questions can help us out:

1. Is the point at which the error is provoked the first
   time through this read statement ?

2. Or is it the first time I/O is done on unit 10 ?

As far as I can see from the libI77 part of the g77 (and f2c)  
runtime library, the function that f__doed points to is probably  
rd_ed (file rdfmt.c), because you're reading a formatted file using  
an `edit string' (I4); it indeed does an fseek (sometimes), and it  
returns errno (why n is still 0 is also not clear to me).

However, I don't see why fseek should fail on unit 10.  Normally,  
unit 10 is either opened by an explicit OPEN statement, or by the  
first I/O on that unit (in which case a file with name `fort.10' is  
opened).  The only way you can get a non-seekable file that way is  
having a named pipe `fort.10' before starting your program.   
However, at open time it is determined whether the file can be  
seeked [sic] (member `useek' of the file unit structure that libI77  
uses internally).  This member is checked before the seek in rd_ed  
is done.

Therefore, I highly suspect one of the underlying components of  
your system (libc or the kernel itself) being at fault here.

I am sorry I cannot help you further.

Toon.

Re: Error in egcs FORTRAN I/O?

[Dave Love]

Since C was mentioned, I'd suspect any attempt to mix C and Fortran
i/o to the same stream, problems with which I've not resolved, but
otherwise this isn't a familiar type of problem.  I'm willing to try
the experiment on a libc5 system if it mainly takes machine time
rather than mine.

Re: Error in egcs FORTRAN I/O?

[craig]

Just rediscovered this old email.  Is it still pertinent, to egcs 1.1.2,
or the 1.2 (gcc 2.95) version being prepared for release?  If so, do
you have more information, per a more complete bug report?  (If not
still pertinent, feel free to just ignore this email!)

        tq vm, (burley)


>I'm trying to use egcs 1.0 to compile our simulation program (mix of
>FORTRAN and C). Everything went really great (and performance jumped
>20% up in comparison to Absoft F77) until I run into an I/O error
>(reproducible). I'm trying to execute the statement
>	READ(10,END=100,FMT='(I4)') NERR1
>and it fails (with run-time error) in the middle of the file. 'errno' is
>set to 29 (#define ESPIPE          29      /* Illegal seek */) and I
>suspect that it happens right on the 4-K block boundary. 
>
>Another interesting observation is that in the following code from
>'do_fio' which triggers the error (at line 472), the value of n is
>exactly 0 (or my gdb does not show it correctly?).
>
>(gdb) frame 7
>#7  0x8069d4a in do_fio (number=0x81600bc, ptr=0x81d4b6c "", len=4)
>    at /tmp/egcs-1.0/gcc/f/runtime/libI77/fmt.c:472
>472                             errfl(f__elist->cierr,errno,"fmt");
>(gdb) list
>467                     if(ptr==NULL)
>468                             return((*f__doend)());
>469                     f__cnt[f__cp]--;
>470                     f__workdone=1;
>471                     if((n=(*f__doed)(p,ptr,len))>0)
>472                             errfl(f__elist->cierr,errno,"fmt");
>473                     if(n<0)
>474                             err(f__elist->ciend,(EOF),"fmt");
>475                     continue;
>476             case STACK:
>(gdb) p n
>$39 = 0
>(gdb) p errno
>$40 = 29
>
>I'm using 
>	egcs-2.90.21 971202 (egcs-1.00 release) 
>RedHat 5.0 on an i386 
>	Linux version 2.0.32 (root@porky.redhat.com) 
>	(gcc version 2.7.2.3) #1 Wed Nov 19 00:46:45 EST 1997)
>
>Any ideas/suggestions what I can try to do next? I have the session
>running in background so I can dig deeper if I'd know where to look...

Re: Error in egcs FORTRAN I/O?

[craig]

Just rediscovered this old email.  Is it still pertinent, to egcs 1.1.2,
or the 1.2 (gcc 2.95) version being prepared for release?  If so, do
you have more information, per a more complete bug report?  (If not
still pertinent, feel free to just ignore this email!)

        tq vm, (burley)


>I'm trying to use egcs 1.0 to compile our simulation program (mix of
>FORTRAN and C). Everything went really great (and performance jumped
>20% up in comparison to Absoft F77) until I run into an I/O error
>(reproducible). I'm trying to execute the statement
>	READ(10,END=100,FMT='(I4)') NERR1
>and it fails (with run-time error) in the middle of the file. 'errno' is
>set to 29 (#define ESPIPE          29      /* Illegal seek */) and I
>suspect that it happens right on the 4-K block boundary. 
>
>Another interesting observation is that in the following code from
>'do_fio' which triggers the error (at line 472), the value of n is
>exactly 0 (or my gdb does not show it correctly?).
>
>(gdb) frame 7
>#7  0x8069d4a in do_fio (number=0x81600bc, ptr=0x81d4b6c "", len=4)
>    at /tmp/egcs-1.0/gcc/f/runtime/libI77/fmt.c:472
>472                             errfl(f__elist->cierr,errno,"fmt");
>(gdb) list
>467                     if(ptr==NULL)
>468                             return((*f__doend)());
>469                     f__cnt[f__cp]--;
>470                     f__workdone=1;
>471                     if((n=(*f__doed)(p,ptr,len))>0)
>472                             errfl(f__elist->cierr,errno,"fmt");
>473                     if(n<0)
>474                             err(f__elist->ciend,(EOF),"fmt");
>475                     continue;
>476             case STACK:
>(gdb) p n
>$39 = 0
>(gdb) p errno
>$40 = 29
>
>I'm using 
>	egcs-2.90.21 971202 (egcs-1.00 release) 
>RedHat 5.0 on an i386 
>	Linux version 2.0.32 (root@porky.redhat.com) 
>	(gcc version 2.7.2.3) #1 Wed Nov 19 00:46:45 EST 1997)
>
>Any ideas/suggestions what I can try to do next? I have the session
>running in background so I can dig deeper if I'd know where to look...