[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
bug#19873: Ill-formed regular expression is constructed in forward-parag
From: |
Alan Mackenzie |
Subject: |
bug#19873: Ill-formed regular expression is constructed in forward-paragraph. |
Date: |
Thu, 2 Dec 2021 20:45:17 +0000 |
Hello, Lars.
On Thu, Dec 02, 2021 at 11:39:51 +0100, Lars Ingebrigtsen wrote:
> Alan Mackenzie <acm@muc.de> writes:
> > I think this idea is workable, but you'll have to check for one or both
> > of paragraph-s{tart,eparate} starting with "[ \t]+". A good strategy
> > here might be to begin the target regexp with "^[ \t]*", then begin one
> > or both components with "[ \t]" (without the "*").
> > There may be other gotchas which I haven't thought about yet.
> > One needs a twisted mind to do this sort of thing properly, so I offer my
> > services to review your upcoming patch. ;-)
> The problem seems rather intractable to me. Is there really any way to
> examine a regexp to determine "does this in practice match [ \t]*"?
Back when the bug was new, I started writing a library to analyse a
regular expression and convert it into an equivalent well formed regular
expression. It's actually working, but is incomplete. It's currently
2757 lines long, including pretty complete unit testing. I actually
looked at it again at the start of November.
> I wonder whether instead of trying to construct a better overall regexp
> could rewrite the loop. That is, instead of searching for sp-parstart,
> search for parstart "\\|" parsep, and then check whether
> (match-beginning 0) of that comes after "^[ \t]*". Or something along
> those lines.
> But I don't know whether that'd be any faster in practice.
It strikes me as one of these things which needs to be done
systematically, which, as I said, I've already tried (and not yet given
up). The question presents itself, would the effort be better spent
improving Emacs's regexp engine?
> Do you have a test case that demonstrates the slowness? In that case I
> could try to see whether there's any alternate approach here that's
> faster.
Martin Rudalics had the original testcase. The slowness was exponential
with the number of spaces typed, I think.
> --
> (domestic pets only, the antidote for overdose, milk.)
> bloggy blog: http://lars.ingebrigtsen.no
--
Alan Mackenzie (Nuremberg, Germany).