tramp no longer recognizes when remote files changed

Hi *,

when a buffer visits a file that gets changes outside of
Emacs the
user is warned the next time he tries to alter the buffer in
question:

  foo changed on disk; really edit the buffer? (y, n, r or
C-h)

This extremely useful feature used to work with tramp and
remote
files, too.  Now some time (days?) ago it stopped working,
if I edit a
remote file that changed in the meantime on disk I get no
warning, and
if I save the buffer all changes made outside of my Emacs
session are
silently lost.

Cheers
sascha
-- 
Sascha Wilde : "There are 10 types of people in the
world. 
             : Those who understand binary and those who
don't."


_______________________________________________
Emacs-devel mailing list
Emacs-develgnu.org
htt
p://lists.gnu.org/mailman/listinfo/emacs-devel

Sascha Wilde <wildesha-bang.de> writes:

> Hi *,

Hi,

> when a buffer visits a file that gets changes outside
of Emacs the
> user is warned the next time he tries to alter the
buffer in question:
>
>   foo changed on disk; really edit the buffer? (y, n, r
or C-h)
>
> This extremely useful feature used to work with tramp
and remote
> files, too.  Now some time (days?) ago it stopped
working, if I edit a
> remote file that changed in the meantime on disk I get
no warning, and
> if I save the buffer all changes made outside of my
Emacs session are
> silently lost.

That's the price ...

Tramp 2.1, brought to Emacs trunk some weeks ago, caches
file
information in order to speed up. It uses cached information
for
example when `file-attributes' is called, and cannot
recognize
external file changes therefore.

Maybe one could introduce an option whether such caches
shall be used,
or not. Tramp checks already `file-precious-flag' when it
copies files
(if set, Tramp applies "cksum" on both the source
and target files in
order to recognize errors). It could also stop using cached
information in such a case - slowing down performance ...

> Cheers
> sascha

Best regards, Michael.



_______________________________________________
Emacs-devel mailing list
Emacs-develgnu.org
htt
p://lists.gnu.org/mailman/listinfo/emacs-devel

Michael Albinus <michael.albinusgmx.de> wrote:
> Sascha Wilde <wildesha-bang.de> writes:
>>   foo changed on disk; really edit the buffer? (y,
n, r or C-h)
>>
>> This extremely useful feature used to work with
tramp and remote
>> files, too.  Now some time (days?) ago it stopped
working,
[...]

> That's the price ...
>
> Tramp 2.1, brought to Emacs trunk some weeks ago,
caches file
> information in order to speed up.

Yes, I recognized the, sometimes tremendous speed up, and
indeed it
made me happy -- but if _this_ is the price, it is much to
high (IMO).

I often remotely edit files which intentionally change
outside Emacs,
and the risk of loosing data with the new tramp version is
high and a
major step backwards compared to the situation before.

[...]
> Maybe one could introduce an option whether such caches
shall be used,
> or not.

Yes, I would highly appreciate that!

I don't know, but I would guess that some of the
optimization could be
even kept.  For example when opening an remote file for the
first time
the new version is much faster, and at this point of time
no
changes would be needed to satisfy my needs, right?

Cheers
sascha
-- 
Sascha Wilde
... mein Opa [...]  würde an dieser Stelle zu Dir sagen:
Junge, such Dir 
ne Frau, bau Dir ein Haus, mach ein Kind und laß' die Finger
von dem Zeug,
das Du gerade machst. -- Michael Winklhofer in
d.a.e.auktionshaeuser


_______________________________________________
Emacs-devel mailing list
Emacs-develgnu.org
htt
p://lists.gnu.org/mailman/listinfo/emacs-devel

Sascha Wilde <wildesha-bang.de> writes:

>> Maybe one could introduce an option whether such
caches shall be used,
>> or not.
>
> Yes, I would highly appreciate that!

Could you, please, try the appended patch? You must set
`file-precious-flag' to t.

> Cheers
> sascha

Best regards, Michael.

*** /home/albinus/src/tramp/lisp/tramp-cache.el	2007-07-27
11:14:10.000000000 +0200
---
/home/albinus/src/emacs/lisp/net/tramp-cache.el	2007-07-10
20:45:30.000000000 +0200
***************
*** 108,124 ****
  
  (defun tramp-set-file-property (vec file property value)
    "Set the PROPERTY of FILE to VALUE, in the cache
context of VEC.
! It is not set when `file-precious-flag' is non-nil. 
Returns VALUE."
!   (unless file-precious-flag
!     ;; Unify localname.
!     (setq vec (copy-sequence vec))
!     (aset vec 3 (directory-file-name file))
!     (let ((hash (or (gethash vec tramp-cache-data)
! 		    (puthash vec (make-hash-table :test 'equal)
! 			     tramp-cache-data))))
!       (puthash property value hash)
!       (tramp-message vec 8 "%s %s %s" file
property value)))
!   value)
  
  (defun tramp-flush-file-property (vec file)
    "Remove all properties of FILE in the cache context
of VEC."
--- 108,123 ----
  
  (defun tramp-set-file-property (vec file property value)
    "Set the PROPERTY of FILE to VALUE, in the cache
context of VEC.
! Returns VALUE."
!   ;; Unify localname.
!   (setq vec (copy-sequence vec))
!   (aset vec 3 (directory-file-name file))
!   (let ((hash (or (gethash vec tramp-cache-data)
! 		  (puthash vec (make-hash-table :test 'equal)
! 			   tramp-cache-data))))
!     (puthash property value hash)
!     (tramp-message vec 8 "%s %s %s" file
property value)
!     value))
  
  (defun tramp-flush-file-property (vec file)
    "Remove all properties of FILE in the cache context
of VEC."



_______________________________________________
Emacs-devel mailing list
Emacs-develgnu.org
htt
p://lists.gnu.org/mailman/listinfo/emacs-devel

Michael Albinus <michael.albinusgmx.de> wrote:
> Sascha Wilde <wildesha-bang.de> writes:
>
>>> Maybe one could introduce an option whether
such caches shall be used,
>>> or not.
>>
>> Yes, I would highly appreciate that!
>
> Could you, please, try the appended patch? You must
set
> `file-precious-flag' to t.

This seems to fix it!  Thanks a lot!

sascha
-- 
Sascha Wilde
To become a Jedi, use Emacs you have to.


_______________________________________________
Emacs-devel mailing list
Emacs-develgnu.org
htt
p://lists.gnu.org/mailman/listinfo/emacs-devel

>>> Maybe one could introduce an option whether
such caches shall be used,
>>> or not.
>> Yes, I would highly appreciate that!
> Could you, please, try the appended patch? You must
set
> `file-precious-flag' to t.

How 'bout an idle hook that flushes the cache after 0.5s of
idleness?


        Stefan


_______________________________________________
Emacs-devel mailing list
Emacs-develgnu.org
htt
p://lists.gnu.org/mailman/listinfo/emacs-devel

    Tramp 2.1, brought to Emacs trunk some weeks ago, caches
file
    information in order to speed up. It uses cached
information for
    example when `file-attributes' is called, and cannot
recognize
    external file changes therefore.

Does it make sense to block use of the cached data
specifically in the
case of checking whether a file has changed for C-x C-f?

What are the main occasions on which the cache is useful?

    > Could you, please, try the appended patch? You must
set
    > `file-precious-flag' to t.

    This seems to fix it!  Thanks a lot!

`file-precious-flag' does other things.  It should not be
reused to
control whether Tramp checks for a changed file.

Having a parameter for users to set is not a very good
solution.
First let's look for a way to make Tramp do the right
combination
of things by default.  Michael, can you try?


_______________________________________________
Emacs-devel mailing list
Emacs-develgnu.org
htt
p://lists.gnu.org/mailman/listinfo/emacs-devel

Richard Stallman <rmsgnu.org> writes:

> What are the main occasions on which the cache is
useful?

I've made a test: Run 1000x (file-attributes
"/ssh:ford:/etc/hosts").
"ford" is another host in my LAN.

With enabled cache, it takes 1.821823 sec. Without cache, it
takes
60.103737 sec. And it is a nearby host.

A simple (find-file "/ssh:ford:/etc/hosts") shows
in the profiler the
following file name handler based functions:

expand-file-name                          127
file-exists-p                             42
file-attributes                           23
file-name-directory                       23
file-directory-p                          11
directory-file-name                       8
substitute-in-file-name                   7
file-name-nondirectory                    5
file-modes                                3
file-readable-p                           2
file-truename                             2
make-auto-save-file-name                  2
file-local-copy                           1
file-symlink-p                            1
file-writable-p                           1
insert-file-contents                      1
set-visited-file-modtime                  1

See f.e. the large number of `file-exists-p' calls. This is
mainly due
to vc, there are repeated tests for existence of

/
/etc
/etc/hosts
/etc/RCS/hosts,v
/etc/hosts,v
/etc/RCS/hosts
/etc/SCCS/s.hosts
/etc/s.hosts
/etc/.bzr
/.bzr
/etc/.hg
/.hg
/etc//=tagging-method
//=tagging-method
/etc/MCVS/CVS
/MCVS/CVS

> Having a parameter for users to set is not a very good
solution.
> First let's look for a way to make Tramp do the right
combination
> of things by default.  Michael, can you try?

Tramp flushes the cached data in
`verify-visited-file-modtime' now. That
seems to be sufficient.

Sascha, could you, please, test it?

Best regards, Michael.


_______________________________________________
Emacs-devel mailing list
Emacs-develgnu.org
htt
p://lists.gnu.org/mailman/listinfo/emacs-devel

    Tramp flushes the cached data in
`verify-visited-file-modtime' now. That
    seems to be sufficient.

That sounds plausible.  It could be that users will report a
couple of
other problems, and we might want to make a couple of other
operations
discard the cached data.  I assume you'll keep an eye out to
see
if this needs to be done.


_______________________________________________
Emacs-devel mailing list
Emacs-develgnu.org
htt
p://lists.gnu.org/mailman/listinfo/emacs-devel

Richard Stallman <rmsgnu.org> writes:

>     Tramp flushes the cached data in
`verify-visited-file-modtime' now. That
>     seems to be sufficient.
>
> That sounds plausible.  It could be that users will
report a couple of
> other problems, and we might want to make a couple of
other operations
> discard the cached data.  I assume you'll keep an eye
out to see
> if this needs to be done.

For sure. Tramp flushes already the cache in in a lot of
other file
name primitives; `verify-visited-file-modtime' has been
simply
overlooked by me. I thought I do it there also.

Best regards, Michael.



_______________________________________________
Emacs-devel mailing list
Emacs-develgnu.org
htt
p://lists.gnu.org/mailman/listinfo/emacs-devel

Michael Albinus <michael.albinusgmx.de> wrote:

> Tramp flushes the cached data in
`verify-visited-file-modtime' now. That
> seems to be sufficient.
>
> Sascha, could you, please, test it?

Just gave it a short try:
- used Emacs to edit remotehost:foo
- sshed to remotehost and issued echo bar>>foo
- back in Emacs trying to edit the foo buffer warns that the
file
  changed on disk
That's the desired behavior.  Thanks!

sascha
-- 
Sascha Wilde : xedit? Das sieht zwar wie vi aus als könne es
nix, aber im
             : Gegensatz zu vi kann es wirklich nix und nix
ist noch geschönt!
             : (Michael Core in dafc)


_______________________________________________
Emacs-devel mailing list
Emacs-develgnu.org
htt
p://lists.gnu.org/mailman/listinfo/emacs-devel