g77.f-torture/execute/alpha1.f is strange

g77.f-torture/execute/alpha1.f is strange

[H.J. Lu]

Here is g77.f-torture/execute/alpha1.f. It may call exit (1) if I
doesn't happen to be 0. I don't remember what Fortran should be in
this case. I think I is default to INTEGER. But I don't know if
it should be initialized to 0. From the x86 asm code, gcc doesn't
initialize it at all. In any case, I got the exit value 1 on
linux/x86/glibc 2 and linux/alpha/glibc 2.0.5c.


-- 
H.J. Lu (hjl@gnu.ai.mit.edu)
---g77.f-torture/execute/alpha1.f---
      REAL*8 A,B,C
      REAL*4 RARRAY(19)/19*(-1)/
      INTEGER BOTTOM,RIGHT
      INTEGER IARRAY(19)/0,0,0,0,0,0,0,0,0,0,0,0,13,14,0,0,0,0,0/
      EQUIVALENCE (RARRAY(13),BOTTOM),(RARRAY(14),RIGHT)
C
      IF(I.NE.0) call exit(1)
C gcc: Internal compiler error: program f771 got fatal signal 11
C  at this point!
      END

Re: g77.f-torture/execute/alpha1.f is strange

[Toon Moene]

>  Here is g77.f-torture/execute/alpha1.f. It may call exit
>  (1) if I doesn't happen to be 0. I don't remember what
>  Fortran should be in this case. I think I is default to
>  INTEGER. But I don't know if it should be initialized to
>  0. From the x86 asm code, gcc doesn't initialize it at
>  all. In any case, I got the exit value 1 on
>  linux/x86/glibc 2 and linux/alpha/glibc 2.0.5c.
>
>
>  -- H.J. Lu (hjl@gnu.ai.mit.edu)
>  ---g77.f-torture/execute/alpha1.f---
>        REAL*8 A,B,C
>        REAL*4 RARRAY(19)/19*(-1)/
>        INTEGER BOTTOM,RIGHT
>        INTEGER IARRAY(19)/0,0,0,0,0,0,0,0,0,0,0,0,13,14,0,0,0,0,0/
>        EQUIVALENCE (RARRAY(13),BOTTOM),(RARRAY(14),RIGHT)
>  C
>        IF(I.NE.0) call exit(1)
>C gcc: Internal compiler error: program f771 got fatal signal 11
>C  at this point!
>        END

Yep, just using a variable I without initialising it is a deadly  
sin - you might be glad g77 didn't start WW III on it - as per the  
comp.lang.fortran FAQ.  Probably the right optional hardware wasn't  
installed ?

This code will only complete normally by chance, so I think it  
needs updating.

Toon.

Re: g77.f-torture/execute/alpha1.f is strange

[Dave Love]

>>>>> "Toon" == Toon Moene <toon@moene.indiv.nluug.nl> writes:

 Toon> This code will only complete normally by chance, so I think it  
 Toon> needs updating.

Yes.  I didn't understand what it was about when I looked at this
stuff; I think it appeared to be testing for a compiler abort and
looked as though it might be in the wrong directory, though I don't
understand the testing stuff.  I also didn't recognize some of the
other things there, but anyway, could regression tests be commented
with an indication of the problem and fix made?

Is Kate C-in-C this stuff if I extract more test cases from the bug
archives now I can get at them again?

Re: g77.f-torture/execute/alpha1.f is strange

[Jeffrey A Law]

  In message < m0xRqJB-0004ecC@ocean.lucon.org >you write:
  > Here is g77.f-torture/execute/alpha1.f. It may call exit (1) if I
  > doesn't happen to be 0. I don't remember what Fortran should be in
  > this case. I think I is default to INTEGER. But I don't know if
  > it should be initialized to 0. From the x86 asm code, gcc doesn't
  > initialize it at all. In any case, I got the exit value 1 on
  > linux/x86/glibc 2 and linux/alpha/glibc 2.0.5c.
Thanks.  Yes, the initialization needs to be explicit.  I've updated
the test.
jeff

Re: g77.f-torture/execute/alpha1.f is strange

[Jeffrey A Law]

  In message < 199711031846.SAA03155@phal.cygnus.co.uk >you write:
  > > Thanks.  Yes, the initialization needs to be explicit.  I've
  > > updatedthe test.
  > 
  > Hmmm, since the compiler aborted, I think the orifinal test should
  > be copied
It could be, but I'm not planning on it.

jeff

Re: g77.f-torture/execute/alpha1.f is strange

[Toon Moene]

Jeff wrote:

>  Thanks.  Yes, the initialization needs to be explicit.
>  I've updated the test.

OTOH, it might simply be that Dave Love was right (which seems to  
be supported by the comment in the code) that this was a *compiler*  
crash, and that it therefore should reside in the `compile'  
directory and not the `execute' directory.

HTH,
Toon.

Re: g77.f-torture/execute/alpha1.f is strange

[Joern Rennecke]

> Thanks.  Yes, the initialization needs to be explicit.  I've updated
> the test.

Hmmm, since the compiler aborted, I think the orifinal test should be copied
into the compile or noncompile directory (depending on whatever that fortran
language says should happen when you try to compile this code)

Re: g77.f-torture/execute/alpha1.f is strange

[Dave Love]

>>>>> "Jeffrey" == Jeffrey A Law <law@cygnus.com> writes:

 >> Hmmm, since the compiler aborted, I think the orifinal test should
 >> be copied
 Jeffrey> It could be, but I'm not planning on it.

It sounds as though the test is then useless, though I can't prove my
suspicion about what the bug was.  FWIW, I got the original alpha-test
report from 1996-12-01 (thanks, nemo).

--------------

Well, I have a new one and I'll repeat the last one...
both these code segments compile fine under g77 linux, but
cause Internal compiler errors on the Alpha.

[...]
=================================================
      REAL*8 A,B,C
      REAL*4 RARRAY(19)/19*(-1)/
      INTEGER BOTTOM,RIGHT
      INTEGER IARRAY(19)/0,0,0,0,0,0,0,0,0,0,0,0,13,14,0,0,0,0,0/
      EQUIVALENCE (RARRAY(13),BOTTOM),(RARRAY(14),RIGHT)
C
      IF(I.NE.0) STOP
C gcc: Internal compiler error: program f771 got fatal signal 11
C  at this point!
      END

Re: g77.f-torture/execute/alpha1.f is strange

[Jeffrey A Law]

  In message < 9711032343.AA10247@moene.indiv.nluug.nl >you write:
  > Jeff wrote:
  > 
  > >  Thanks.  Yes, the initialization needs to be explicit.
  > >  I've updated the test.
  > 
  > OTOH, it might simply be that Dave Love was right (which seems to  
  > be supported by the comment in the code) that this was a *compiler*  
  > crash, and that it therefore should reside in the `compile'  
  > directory and not the `execute' directory.
ok ok ok  :-)  I'll take out the initialization and move the test..


jeff

Re: g77.f-torture/execute/alpha1.f is strange

[Craig Burley]

>ok ok ok  :-)  I'll take out the initialization and move the test..

That restores the test to its original status, but its status
as "executable" also is preserved by compiling it with
"-finit-local-zero", which might be useful as well, though
in this case not involving a past bug-fix I'm aware of
(though, if the test had any local variables of COMPLEX type,
it would ;-).

(This is an example of why I'm not thrilled with categorizing
the *source* code of a test based on its *uses*, since the
use can straddle categories...I think of uses as attributes,
perhaps, I guess.  But that's not worth discussing right now.)

        tq vm, (burley)

Re: g77.f-torture/execute/alpha1.f is strange

[Dave Love]

>>>>> "Kate" == Kate Hedstrom <kate@imcs.rutgers.edu> writes:

 Kate> You can take over if you think that makes more sense.

I don't know dejagnu.  (Doubtless a poor reflexion...)

[I can't currently read the g77 mail archives as I thought I could.]

Re: g77.f-torture/execute/alpha1.f is strange

[Kate Hedstrom]

>  Toon> This code will only complete normally by chance, so I think it  
>  Toon> needs updating.
> 
> Yes.  I didn't understand what it was about when I looked at this
> stuff; I think it appeared to be testing for a compiler abort and
> looked as though it might be in the wrong directory, though I don't
> understand the testing stuff.  I also didn't recognize some of the
> other things there, but anyway, could regression tests be commented
> with an indication of the problem and fix made?

Sure they could be commented, although all I have is some code snippets
which once failed and no indication of what was wrong or what was
fixed.

> Is Kate C-in-C this stuff if I extract more test cases from the bug
> archives now I can get at them again?

As far as I know I still am, although I have to admit that I've been
working on other things lately.  You can take over if you think that
makes more sense.

Kate