Re: tests/yesno.sh, kinds of line (matching/non-matching, selected/rejected)

[Charles Levert]

* On Sunday 2005-11-13 at 17:36:03 +0000, Julian Foad wrote:
> Charles Levert wrote:
> >You can see by my design of the GREP_COLORS
> >framework that I just plain forgot about -v and
> >confused things:
> >
> >   ml:  terminfo-like string capability for matching lines
> >   cx:  terminfo-like string capability for context lines
> [...]
> >Should I have created something like this:
> >
> >   ml:  terminfo-like string capability for matching lines
> >   nl:  terminfo-like string capability for non-matching lines
> >
> >Or, like this:
> >
> >   sl:  terminfo-like string capability for selected lines
> >   rl:  terminfo-like string capability for rejected lines
> 
> This last pair, roughly; certainly the first of the pair should be 
> "selected" and not "matching".
> 
> The two sorts of lines that Grep prints (and which should be 
> distinguishable by colour) are the ones it was asked to find (selected 
> lines), and context lines (which are some of the rejected lines).  It only 
> prints rejected lines if and when they are context lines.  Therefore I 
> think it is clearer to call the printed non-selected lines "context lines" 
> rather than "rejected lines":

I'd like to define "rejected" somewhere, just
because it's a term that covers all non-selected
lines, not just their printed subset, but
"context" is indeed more accurate for this
very subset.  Maybe the full manual could have
a "Concepts and Terminology" @section in the
"Introduction" @chapter, or something.


>   sl:  terminfo-like string capability for selected lines
>   cx:  terminfo-like string capability for context lines
> 
> [...]
> >Is it too late to just incompatibly change this,
> >since there hasn't been an offical release
> >since the ill-designed version was introduced?
> 
> No, it's not too late.

One consequence to document is that now,
both sl and cx must be able to "support" mt.
By this I mean that the code intentionally does

   sgr_start(sl or cx)
   fwrite(non-matched text)
   /* no sgr_end() here */
   sgr_start(mt)
   fwrite(matched text)
   sgr_end()
   sgr_start(sl or cx)
   fwrite(non-matched text)
   sgr_end()
   fputs("\n")

for matching lines, so users must choose their
colors compatibly (e.g., if mt doesn't specify
a background color but sl and cx do).


> This is all experimental until we release it.  I'm 
> glad you discovered this problem before we have made a release.

I noticed that grep.el currently does
GREP_COLORS="...:ml=:cx=:...".  I designed
parse_grep_colors() to ignore unknown
capabilities, so now the legacy empty ml= would
just be ignored which corresponds to its default
being empty.  We can notify the grep.el people
of the change, as they seem to be the single
early adopters.