[Bug libfortran/22570] Null Characters instead of blanks in t extoutput

[Steven Bosscher]

On Thursday 21 July 2005 15:36, THOMAS Paul Richard 169137 wrote:
> Steven,
>
> > On Thursday 21 July 2005 12:11, Steven Bosscher wrote:
> > > Thanks for your efforts.  Sadly these patches still do not
> >
> > fix mgrid :-(
>
> The thanks are mine for bearing with me!  During an out-of-body experience
> during a meeting, I understood the 4th(!) problem.
>
> All the test cases consisted, one way or another, of a single record.
> Whilst, the position counters were being reset at the start of a transfer,
> they were not being rest after each record.
>
> The enclosed patch runs each of your testcases, for which thanks, and has
> been regtested on Cygwin_NT.
>
> To recapitulate:
>
> The four buglets that I have fixed are:
>
> (1) The receipt of a slash should clear the pending skips and spaces(last
> night's patch);
> (2) The skips and spaces need to accumulate to cope with sequences of tabs
> and X-edits,
> as mentioned and tested this morning; and
> (3) The function read.c(read_x)must ignore current_unit->bytes_left, except
> where the
> record length is known.
> (4) Make sure the position counters are reset for a new record.
>
> I will incorporate these four fixes into one patch and provide one or,
> possibly, two test cases.

Your patch with these fixes (and Jerry's fix!) revives four of the many
failing SPEC benchmarks:

                     UNPATCHED CVS TODAY:           PATCHED CVS TODAY:
   168.wupwise       1600  184             872*     1600  183             876*
   171.swim          3100  445             697*     3100  444             699*
   172.mgrid                                        1800  247             729*
   173.applu                                        2100  330             636*
   178.galgel                                X                             X
   187.facerec                               X                             X
   189.lucas                                 X                             X
   191.fma3d                                 X                             X
   200.sixtrack                                     1100  259             424*
   301.apsi                                         2600  313             832*

That still leaves four broken.  I know for sure that galgel is broken
due to rth's tree-complex rewrite.  I will investigate the other ones.

Paul, can you commit that patch when you have it?  Jerry, your patch
is OK too (it seems I didn't already say that...?).

Gr.
Steven