Colorizing error messages

[David Malcolm]

On Thu, 2016-10-27 at 11:31 -0700, Jerry DeLisle wrote:
> Hi all,
> 
> I notice that our front end diagnostics are colorized but the runtime
> messages 
> are not.  Does anyone know how we could add the color at runtime?
> 
> Jerry

(sorry for the belated response; I just got back from vacation)

I maintain the diagnostics subsystem, but I'm not very familiar with
Fortran.

The frontend diagnostic colorization happens on the host via diagnostic
-color.c, which by default looks to see if stderr is directly connected
to a tty and colorizes if it is.

I'm guessing that the fortran runtime messages you refer to are the
ones emitted by libgfortran/runtime/error.c.

It's not clear to me exactly how you'd want the colorized errors to
look.  Presumably diagnostic-color.c could be also built for the
*target* and integrated into libgfortran/runtime/error.c somehow
(perhaps calling colorize_init from main.c's init, or the first time an
error is emitted?).   If so, there might need to be a licensing tweak,
to ensure that diagnostic-color.c has the runtime library exception (I
don't think it does at the moment).

You'd also want to have automated test coverage for this.  I found
testing of colorization to be a little awkward to do in DejaGnu.  For
the frontend diagnostics, we have tests like
 gcc/testsuite/gcc.dg/plugin/diagnostic-test-show-locus-color.c
though IIRC this merely tests the colorization of the source-code and
carets/underlining, not of the messages themselves.

Is it worth tracking this idea as a PR?

Hope this is helpful
Dave