[patch,fix, PING] PR21875 Internal Array IO and NIST final fix

[Jerry DeLisle]

Steven Bosscher wrote:
> On Tuesday 13 September 2005 21:55, Jerry DeLisle wrote:
> 
>>Attached is revised patch for internal unit array I/O
>>
>>This revision affects only the frontend patches which are attached with new
>>test cases.
>>
>>(
>>See http://gcc.gnu.org/ml/fortran/2005-09/msg00176.html for frontend patch
>>to trans-io.c
>>)
>>
>>Significant changes relate to checking for contiguous array sections and
>>adding error messages for bad arrays and end of file condition.
>>
>>I would like to request we try to get this into 4.0.2 since it resolves
>>significant deficiencies in NIST, and other test suites.
>>
>>Regression tested OK, NIST tested OK
>>
>>OK for 4.1?
> 
> 
> There are some formatting errors (space after "(int)" in next_record_w,
> wrongly indented "for" in get_array_unit_len, and some in mem_alloc_w_at)
> but it looks OK to me otherwise.  And you're the library gurus by now
> anyway... ;-)

OK I will fix.

> 
> The only surprising bit is this:
> +   if (*len < 0)  /* This should never happen */
> +     *len = 0;
> + 
> So if it never happens, why have it?

I put that in there to flag out an issue in review (which it has).  There was a 
point where len was actually going negative and causing a segfault.  I put that 
there to catch that condition long enough for me to see what was happening to 
fix it.

Its sort of a policy question.  Leaving it may prevent a future segfault if 
something starts going wrong again.  It would allow us to see bad output rather 
than a segfault.  Other than that, there is no need.  I will take it out.

With these changes I assume that it is OK to commit to 4.0 after retesting.

> 
> The front-end bits from the patch are OK.
> 
> 
>>4.0.2 if approved by Mark? 
> 

I will await comment from Mark on the 4.0.2 side

> 
> I agree it would be good to have this in 4.0 also, but it is probably
> too late for 4.0.2 by now.  I have added Mark to the CC:.
> 
> Gr.
> Steven
> 
>