bug#51622: 29.0.50; [PATCH v2] Abbreviate remote home directories in `abbreviate-file-name'

[Michael Albinus]

Jim Porter <jporterbugs@gmail.com> writes:

Hi Jim,

> If I profile `tramp-handle-file-name-case-insensitive-p' on Emacs 29
> master, I see that 52% of the time is spent in `file-remote-p' (and a
> further 30% in `expand-file-name'). This is where the difference in
> performance comes from; I don't think my patch helps with that, but
> when testing, I eliminated the `file-remote-p' call (and the remote
> check) and saw performance improve. I just got mixed up about what the
> issue was.
>
> If the calls to `file-remote-p' and `expand-file-name' could be
> optimized or replaced, that would make this function a lot
> faster. I'll take a look at this and see what I can do to improve
> things. If you have any suggestions though, I'm happy to hear about
> them.

The check (file-remote-p filename nil 'connected) is applied to prevent
unintended work. It returns non-nil when there is already a connection
to the remote host. The intention is to avoid, that
`tramp-handle-file-name-case-insensitive-p' opens a new connection just
in order to see, whether the remote file system is case
insensitive. This is not needed when there's no connection yet.

Hmm, we could replace this check by something cheaper: is there already
a connection process?

--8<---------------cut here---------------start------------->8---
     (and (let ((non-essential t)) (tramp-connectable-p filename))
          ...
--8<---------------cut here---------------end--------------->8---

The `non-essential' variable must be bound to non-nil; otherwise
`tramp-connectable-p' would always return non-nil.

Does this help? In case of, you might offer a patch, with a comment why
we don't use (file-remote-p filename nil 'connected) .

For the `expand-file-name' case I have no offer yet. But let's see how
the timing changes with that change above.

Btw, when you run benchmark checks, you shall set (or let-bind)
`tramp-verbose' to 0. Writing debug data has a cost.

> (The main reason I'm digging into this right now is that you mentioned
> it would be nice if we didn't have to copy-and-paste so much code from
> `abbreviate-file-name'. I'm looking into making a new function called
> something like `directory-abbrev-apply', which *only* does the
> `directory-abbrev-alist' replacements so that Tramp can call
> this.

Good idea, I was also thinking about. For backward compatibility we will
keep a mirror of that function in tramp-compat.el, until we support only
Emacs 29+. Isn't this a glory future? :-)

Best regards, Michael.