filecache.el questions

filecache.el questions

[Ted Zlatanov]

Hi,

in using filecache.el I found two issues I have for now addressed on my
own:

1) the file-cache-alist is never saved.  This is a problem with large
file sets (incidentally, it would be nice if Emacs had a native way of
storing large data sets in an external database, as I mentioned).  I
wrote a serialization function that essentially dumps with prin1 and
princ and reads back with eval.  I also added a variable with the list
of locations that should be scanned at startup.

I wanted to ask if I should submit that as a patch.  filecache.el seems
pretty lean so I wasn't sure if I should add that kind of weight to it.

2) multiple alternatives are proposed sequentially (with repeated C-u
C-TAB).  For example, if hello.c is in /data and /data2:

/tmp/hello.c[C-TAB] -> /data/hello.c[C-u C-TAB] -> /data2/hello.c[C-u C-TAB]

The message displayed is just "1 of 2" and "2 of 2".  I modified the
message to say:

1 of 2
*1:/data/hello.c*
 2:/data2/hello.c

2 of 2
 1:/data/hello.c
*2:/data2/hello.c*

but that's just a hack, not as good as completing-read on the
alternatives (this is why I'm not submitting a patch).  Has
completing-read been considered for the multiple matches case?

Thanks
Ted

Re: filecache.el questions

[Ted Zlatanov]

(I heard from Paul, who I didn't CC correctly on my last e-mail so he
didn't see that this went to emacs-devel as well.  Fixing this.)

Paul Breton wrote:
> If the code is fairly small, I think it wouldn't hurt to add an
> optional way to save file-cache-alist. A list of locations sounds like
> overkill to me, however; I would have one variable to say whether or
> not to save it (defaulting to not), and another one to save where it
> should be saved.

> (Personally, I never save the cache because eventually it would get
> out of sync with the actual filesystem; initializing once per emacs
> session (which are fairly long-lived for me) works for me. However,
> people who use it differently might benefit from the new code).

I have to restart it often, and I need to scan a *lot* of directories,
so it makes sense for me.  

I added a file-cache-rescan function that will rescan and save, and
file-cache-initialize which will do a rescan if the save file doesn't
exist (intended to be called when you load the library).  It doesn't
have to have a list of file locations, you're right.  Does the
rescan/initialize mechanism seem OK?  The actual code is pretty
trivial, though I keep wishing it was even simpler through some Emacs
persistence mechanism.

> Similarly, it has never bothered me very much to simply cycle thru the
> various completions. I wouldn't use the display hack simply because it
> seems like it could get ugly when there are a lot of
> completions. Completing-read is a good idea; however, I haven't looked
> at this code in a long time and am still scratching my head over how
> the completion magic works :)

Here's a patch to do a completing-read when more than one directory is
available, using enable-recursive-minibuffers which is a bit hacky.  It
does work better than the current method, which just inserts the next
value in the list.  This replaces the current method; the two can
coexist as well.

Ted

Index: filecache.el
===================================================================
RCS file: /sources/emacs/emacs/lisp/filecache.el,v
retrieving revision 1.37
diff -c -r1.37 filecache.el
*** filecache.el	6 May 2008 07:57:35 -0000	1.37
--- filecache.el	1 Oct 2008 20:23:24 -0000
***************
*** 511,525 ****
       (t
        (let* ((minibuffer-dir (file-name-directory (minibuffer-contents)))
  	     (dir-list       (member minibuffer-dir directory-list))
  	     )
! 	(setq directory
! 	      ;; If the directory is in the list, return the next element
! 	      ;; Otherwise, return the first element
! 	      (if dir-list
! 		  (or (elt directory-list
! 			   (setq num (1+ (- len (length dir-list)))))
! 		      (elt directory-list (setq num 0)))
! 		(elt directory-list (setq num 0))))
  	)
        )
       )
--- 511,521 ----
       (t
        (let* ((minibuffer-dir (file-name-directory (minibuffer-contents)))
  	     (dir-list       (member minibuffer-dir directory-list))
+ 	     (enable-recursive-minibuffers t)
  	     )
! 	(setq directory (completing-read 
! 			 (format "Select a directory for %s: " file) 
! 			 directory-list nil t))
  	)
        )
       )

Re: filecache.el questions

[Mathias Dahl]

> in using filecache.el I found two issues I have for now addressed on my
> own:
>
> 1) the file-cache-alist is never saved.  This is a problem with large
> file sets (incidentally, it would be nice if Emacs had a native way of
> storing large data sets in an external database, as I mentioned).  I
> wrote a serialization function that essentially dumps with prin1 and
> princ and reads back with eval.  I also added a variable with the list
> of locations that should be scanned at startup.

I had the same problem when I learnt about this and I created a page
over at EmacsWiki about it:

http://www.emacswiki.org/cgi-bin/wiki/FileNameCache

Nowadays I use `anything' and the `anything-c-source-file-cache'
source instead of my old iswitchb hack. I even have a special command
when I want to use `anything' just to get files from the cache (the
cache is so large that I don't want it to be used for normal
invocations of `anything'):

(defun anything-file-cache ()
  (interactive)
  (let ((anything-sources
         (list anything-c-source-file-cache)))
    (anything)))

Enjoy!

/Mathias

Re: filecache.el questions

[Ted Zlatanov]

On Thu, 2 Oct 2008 13:36:36 +0200 "Mathias Dahl" <mathias.dahl@gmail.com> wrote: 

>> in using filecache.el I found two issues I have for now addressed on my
>> own:
>> 
>> 1) the file-cache-alist is never saved.  This is a problem with large
>> file sets (incidentally, it would be nice if Emacs had a native way of
>> storing large data sets in an external database, as I mentioned).  I
>> wrote a serialization function that essentially dumps with prin1 and
>> princ and reads back with eval.  I also added a variable with the list
>> of locations that should be scanned at startup.

MD> I had the same problem when I learnt about this and I created a page
MD> over at EmacsWiki about it:

MD> http://www.emacswiki.org/cgi-bin/wiki/FileNameCache

My save method is a bit more complicated, since it uses a method similar
to how Gnus saves its newsrc file, and do more protection of the
contents (print-quoted, print-readably, etc.).  The file has one big
setq statement:

(setq file-cache-alist '(...))

MD> Nowadays I use `anything' and the `anything-c-source-file-cache'
MD> source instead of my old iswitchb hack. I even have a special command
MD> when I want to use `anything' just to get files from the cache (the
MD> cache is so large that I don't want it to be used for normal
MD> invocations of `anything'):

MD> (defun anything-file-cache ()
MD>   (interactive)
MD>   (let ((anything-sources
MD>          (list anything-c-source-file-cache)))
MD>     (anything)))

That's a much better way than what I proposed.  I hope anything.el makes
it into the Emacs core, it's a very useful package.  I'll be using it
rather than play around with completing-read, etc.

Ted