bug#36564: 27.0.50; Wrong number of errors in compilation mode-line

[Juri Linkov]

>> I tried compilation--ensure-parse, and it updates the number of errors,
>> so I replaced font-lock-ensure with compilation--ensure-parse.
>
> Thanks, that's much better.
> I'm still worried about the performance cost of running
> compilation--ensure-parse every time we get a few chars from the
> process filter.
>
> Not sure how/where to delay it, tho.  Maybe some idle timer?

IIUC, after adding compilation--ensure-parse there is no
performance degradation in case when the compilation buffer
is displayed during compilation, because compilation--ensure-parse
is called on the same-sized chunks as when the buffer is fontified
by font-lock.

I noticed this when tried to debug the problem of fontifying
diff hunks, but I failed to find a solution.  The problem is this:
sometimes diff-mode doesn't refine some hunks during font-lock
when the first part of the hunk emitted by diff-process-filter
(from the diff command comparing files) is fontified partly,
then after emitting the remaining part of the hunk it remains
unfontified.