Re: Introducing 'unrecognized and 'ignored

[Eric S. Raymond]

Alexandru Harsanyi <address@hidden>:
> I had a look at the problem last night and wrote the code to load a 
> cvsignore file and produce a regexp to match the ignored files, however the 
> whole mechanism required by `vc-BACKEND-ignorable-file-p' is quite tricky 
> to implement corectly (in a way that both Emacs and the backend agree 100% 
> on what files are ignored).  Here are two of the problems I found when I 
> looked at an implementation for CVS:
>
>  * CVS will ignore a file only if it is unregistered.  For example .o files 
> are ignored from a CVS status listing, but if an .o file is registered than 
> that file will show up in the listing.

That is actually sensible behavior.  From CVS, surprisingly so :-)

>                                      This requires checking whether a 
> file is registered or not, a thing which we try to avoid.

Not necessarily.  See below.

>  * There's a 'per-repository' cvsignore file at $CVSROOT/CVSROOT/cvsignore 
> which Emacs cannot access directly for remote repositories.

That, on the other hand, is kind of a showstopper.

> With extended vc-state states I don't think there is a need for 
> `vc-ignorable-file-p' at all:  `vc-BACKEND-dir-state' should set the state 
> for each registered file and for files it can register.  `vc-dired-hook' 
> can than simply look for the vc-state property and if it is missing it can 
> assume that the file is ignored.

Thinking about it, you're right.  We should consider that what
dir-state does is update the state properties for as many files as it
can, which means we can call (vc-state) on every file in the dired
listing and only get expensive operations for the ones dir-state
didn't report a status on.

Then the problem of speeding up vc-dired-hook will reduce to a simpler
one -- how to make the individual directory-status commands in each
VCS return information for as many files as possible? Ideally, we want
them to return status on *all* files beneath the current directory.

Some VCSes will make this easy. Mercurial, for example, has hg status -A,
which is documented to return status flags for all files.  Now that we 
have 'unregistered and 'ignored, this means that the individual
vc-state calls will never have to do their own check.  Mercurial is
going to be really fast.

Some make it hard; CVS, for example.  I don't know any way to make its
dir-state set 'ignored files, and it will set 'unregistered only if the
stay-local-p test fails and dir-state goes down the path that parses
vc-status output (something we want to avoid because it's dog-slow).

Most are in between.  There is probably a way to beat Subversion into
reporting status on all files, but I don't know what it is yet.

I'll tackle this for Subversion -- would you look at making the CVS
dir-state better?  The goal is to get it to somehow set 'ignored 
and 'unregistered whenever appropriate.

There is also a minor can of worms near 'added.  It's not really a
fully-supported state yet.  It should be.  And 'removed, too.
-- 
                <a href="http://www.catb.org/~esr/";>Eric S. Raymond</a>