[gfortran] patch for pr libfortran/21303

[Jerry DeLisle]

FX Coudert wrote:
>> I am using i686-pc-linux-gnu (aka Fedora Core 4)
> 
> 
> I am also using a i686-pc-linux-gnu. To be sure, I updated my tree, 
> removed other (pending) patches and bootstrapped from scratch. After 
> that, running the fmt_l.f90 testcase gives:
> 
> $ make check-fortran RUNTESTFLAGS="dg.exp=fmt_l.f90"
> [... snip ...]
>                 === gfortran Summary ===
> 
> # of expected passes            152
> /home/fxcoudert/svn/debug/ibin/gcc/testsuite/gfortran/../../gfortran  
> version 4.2.0 20060219 (experimental)
> 
> 
> To be sure, I attached the final patch one again (but it's really the 
> same as the previous diff with the updated fmt_l.f90).
> 
>> At compile time with -std=f95 -pedantic:
>>
>> The warning is:
>>
>> $ gfc -std=f95  fmt_l.f90
>>  In file fmt_l.f90:14
>>
>> or" }
>>     1
>> Warning: Positive width required in format string at (1).  Also notice 
>> the location is off at the end of the comment.
>>
>> So the message does not match the dg-warning.  I gather this is not 
>> what you intended.
> 
> 
> No, the warning I'm expecting happens with -std=gnu -pedantic. I could 
> also test the warning issued with -std=f95, which is as you noted 
> different (because it's the general warning for such problems), but that 
> would mean having either a run-time error (which is harder to test) or 
> no runtime output at all (and I'd like to test the runtime warning).
> 
> So, in short: I'm sure I want to use -std=gnu -pedantic as dg-options, 
> though we still have to figure out why it's working for me and not for 
> you. If someone else could spare a few cycles on this, I'd be glad.
> 
> FX
> 
> 
OK! After a clean bootstrap it passes regression and the test case as expected.

OK to commit to Trunk and 4.1 after the freeze.

Best regards,

Jerry