Re: Unbearably slow editing in .h files (was: patch: add-log.el: changelog find file under point)

[Alan Mackenzie <acm@muc.de>]

Hello, Stefan and Martin.

On Mon, Feb 18, 2008 at 03:09:42PM -0500, Stefan Monnier wrote:
> > The simple approach is based on searching and comparing the result with
> > that of `add-log-current-defun'.  It works pretty well most of the time
> > but may exhibit the nasty behavior described in

> > http://lists.gnu.org/archive/html/emacs-devel/2007-10/msg00443.html

, whose content is thus:

:Try the following:
:
:  emacs -Q src/lisp.h
:  C-s Fplist_put RET
:  C-x 4 a
:
:you should see Emacs get stuck in an infinite loop (not a bad, one: C-g
:stops it just fine).
:
:
:Stefan
:
:PS: BTW, am I the only one who experiences odd slowdowns when editing
:src/lisp.h, apparently due to font-lock occasionally taking a lot of time
:(like a second or two) to do its job.  Not easy to reproduce, sadly.

It does a lot of cacheing of things; that second or two is filling up
caches, which prevents it from being "reproduced" immediately.  ;-)

> > On my 1 GHz machine this means that I have to wait some 40 secs until
> > I'm told that there are no matches.  I can think of two ways to handle
> > this: (1) Convince Alan that the behavior of `c-beginning-of-defun' is
> > intolerable - AFAICT this has been tried before.

;-)  I think that lisp.h is an ill-conditioned file, in some way.  Or
are you telling me that the slowness happens in a lot of files.h?  The
intolerability could be due to searching for an unnested brace instead
of one at column zero.

(2) Fix this within
> > `add-log-current-defun' to not use `c-beginning-of-defun' - I'm unable
> > to do this myself because I completely fail to understand that part of
> > the function.

What is add-log-current-defun using c-beginning-of-defun for?  How many
times does it call c-b-o-d?  If it's calling it lots of times, why.  OK,
cancel that, I've just had a look at the function (in Emacs-22).  It's
using several sledgehammers to crack some tiny nuts, presumably dating
from the era in which the sledgehammer was the only available tool.  The
bit in add-log-current-defun dealing with C files could usefully use
`c-beginning-of-macro' and, in particular,
`c-where-wrt-to-brace-construct'.

The latter function painstakingly analyses where, in a defun, point is,
returning a value such as 'in-header or 'outwith-function.  It's a
somewhat slow function, is called at the start of every
c-{beginning,end}-of-defun call, and its result is not currently cached.
Using it directly, instead of the five explicit calls to
c-\(beginning\|end\)-of-defun would give a speed up of 5 or 2.5 in C-x 4
a.

> Nowadays, I find myself regularly using M-x fundamental-mode in .h files
> in order to be able to edit them sanely.  We have to do something to
> fix it.  Alan?

Which files.h are you talking about, Stefan?  CC Mode seems to work well
enough for me in my day job (maintaining typical old
"proprietary-quality" C code).  Maybe .../src/lisp.h and one or two
others are somewhat extreme in their use of macros, or something.  A
solution might be to set open-paren-in-column-0-is-defun-start in the
file local variables list, but I can't remember offhand if CC Mode
handles it properly (I suspect it doesn't).
 
>         Stefan

-- 
Alan.