Can't interrupt directory_files_internal run from timer-event-handler

Can't interrupt directory_files_internal run from timer-event-handler

[Klaus Zeitler]

The following problem occurs with CVS emacs + ECB under Solaris 5.8.

While ECB performs its stealth update activities, emacs sometimes seems to
hang for a long time.
gdb shows the following function and backtrace:

directory_files_internal (directory=16076323, full=3313665, match=3313665, 
    nosort=3313713, attrs=0, id_format=3313665)
    at /vol/freeware/cvs/emacs/head/emacs/src/dired.c:226
226           if (dp == NULL && errno == EAGAIN)
(gdb) p directory
$1 = 16076323
(gdb) pr
"/vol/cld_scratch"

(gdb) xbacktrace
 "directory-files"
 "apply"
 "ecb-directory-files"
 "ecb-check-emptyness-of-dir"
 "ecb-stealthy-empty-dir-check"
 "ecb-stealthy-updates"
 "apply"
 "byte-code"
 "timer-event-handler"

I think emacs waits for readdir in dired.c:223 to complete and unfortunately
C-g doesn't work during stealthy update.

The culprit BTW is a mount point to a VAX cluster that e.g. need 20 minutes
to return from 'M-x eval-expression (directory-files "/vol/cld_scratch")'.
While I can interrupt 'M-x eval-expression (directory-files "/vol/cld_scratch")'
with C-g just fine, C-g has no effect when this command is run from 
timer-event-handler.

-- 
 ------------------------------------------
|  Klaus Zeitler      Lucent Technologies  |
|  Email:             kzeitler@lucent.com  |
 ------------------------------------------
---
The only problem with being a man of leisure
is that you can never stop and take a rest.

Re: Can't interrupt directory_files_internal run from timer-event-handler

[Richard M. Stallman]

Timer code normally turns off C-g so that you won't interrupt it
by accident.  (After all, you can't predict when the timer code
will be running.)  If ECB is going to do something potentially
slow from a timer, it should probably bind inhibit-quit to nil
around that code, after displaying a message in the echo area
to inform the user what is happening.

Alternatively, it could use start-process to run ls asychronously
as a way to find out whether the directory is empty.  The process
sentinel could take action based on whether the output was empty.
Then you wouldn't have any such problem.
I suggest forwarding this to the ECB developers.

Re: Can't interrupt directory_files_internal run from timer-event-handler

[Eli Zaretskii]

> From: "Richard M. Stallman" <rms@gnu.org>
> Date: Fri, 05 Aug 2005 16:38:14 -0400
> Cc: emacs-devel@gnu.org
> 
> Alternatively, it could use start-process to run ls asychronously
> as a way to find out whether the directory is empty.

There's no need to run any program just to know if a directory is
empty.  Emacs has primitives which will tell that directly (e.g.,
file-attributes) and do that faster.

Re: Can't interrupt directory_files_internal run from timer-event-handler

[Richard M. Stallman]

    There's no need to run any program just to know if a directory is
    empty.  Emacs has primitives which will tell that directly (e.g.,
    file-attributes) and do that faster.

How do you determine that from file-attributes?  It is not obvious.
You might try to do it from the number of hard links to the directory,
which will normally be 2 for an empty directory, since users normally
can't create alternate names for a directory.  But isn't it possible
for root to create them?  Then an empty directory could have a link count
greater than two.

Re: Can't interrupt directory_files_internal run from timer-event-handler

[Eli Zaretskii]

> From: "Richard M. Stallman" <rms@gnu.org>
> CC: kzeitler@lucent.com, emacs-devel@gnu.org
> Date: Sun, 07 Aug 2005 13:15:29 -0400
> 
>     There's no need to run any program just to know if a directory is
>     empty.  Emacs has primitives which will tell that directly (e.g.,
>     file-attributes) and do that faster.
> 
> How do you determine that from file-attributes?  It is not obvious.

A combination of file-attributes (looking at the number of links) and
directory-files should do that, I think.

> You might try to do it from the number of hard links to the directory,
> which will normally be 2 for an empty directory, since users normally
> can't create alternate names for a directory.  But isn't it possible
> for root to create them?

I'm not an expert on this, but I think it's impossible to create such
alternate names, since GNU Find uses this fact to detect directories
with no subdirectories.

Re: Can't interrupt directory_files_internal run from timer-event-handler

[Johan Vromans]

Eli Zaretskii <eliz@gnu.org> writes:

> I'm not an expert on this, but I think it's impossible to create such
> alternate names, since GNU Find uses this fact to detect directories
> with no subdirectories.

GNU Find also has a command line option to defeat this heuristic,
probably with reason.

-- Johan

Re: Can't interrupt directory_files_internal run from timer-event-handler

[Johan Vromans]

Johan Vromans <jvromans@squirrel.nl> writes:

> GNU Find also has a command line option to defeat this heuristic,
> probably with reason.

Sorry, I have to correct this. I got the information from the manual
page for 'find' under the assumption that a GNU/Linux system would
have GNU find. However, it turned out to be the case that my system
(Fedora Core 2) has a different find utility (from the Secure Locate
kit).

-- Johan

Re: Can't interrupt directory_files_internal run from timer-event-handler

[Richard M. Stallman]

    A combination of file-attributes (looking at the number of links) and
    directory-files should do that, I think.

But...it is using directory-files now, and that is what causes the
problem.  The question is how to avoid using using directory-files.

    I'm not an expert on this, but I think it's impossible to create such
    alternate names, since GNU Find uses this fact to detect directories
    with no subdirectories.

Linux seems to refuse to do this even if you are root.
But I believe that root could do it on Unix at some point.
Maybe that was only in old versions in the early 80s.

It might be that that case is so peculiar that it can safely be
ignored here.  If so, looking at the link count would be ok.

Re: Can't interrupt directory_files_internal run from timer-event-handler

[Stefan Monnier]

> Timer code normally turns off C-g so that you won't interrupt it
> by accident.  (After all, you can't predict when the timer code
> will be running.)  If ECB is going to do something potentially
> slow from a timer, it should probably bind inhibit-quit to nil
> around that code, after displaying a message in the echo area

Indeed.  Except that just binding inhibit-quit to nil may not quite do
what we want: it should use with-local-quit or something equivalent.


        Stefan

Re: Can't interrupt directory_files_internal run from timer-event-handler

[Stefan Monnier]

> There's no need to run any program just to know if a directory is
> empty.  Emacs has primitives which will tell that directly (e.g.,
> file-attributes) and do that faster.

If directory-files blocks (on a failed NFS server), file-attributes is
pretty likely to block just as well, in which case `ls' may be a good way to
do the equivalent of running directory-files asynchronously.


        Stefan

Re: Can't interrupt directory_files_internal run from timer-event-handler

[Sascha Wilde]

[-- Attachment #1.1: Type: text/plain, Size: 1222 bytes --]

On Mon, Aug 08, 2005 at 08:27:40PM -0400, Richard M. Stallman wrote:
>     I'm not an expert on this, but I think it's impossible to create such
>     alternate names, since GNU Find uses this fact to detect directories
>     with no subdirectories.
> 
> Linux seems to refuse to do this even if you are root.
> But I believe that root could do it on Unix at some point.
> Maybe that was only in old versions in the early 80s.

Yes, indeed Unix v5 (fifth edition from 1974, that is _not_ SysV!),
does allow root to make hard links on directories.  But the seventh
edition from 1978 already prohibits it.  (sorry, at the moment I don't
have a v6 at hand to test it there...  ;-))

Every more modern Unix I tried, whether free or proprietary, whether
SysV or BSD based, refuses to do this even for root.

So I'd think that this case can be safely ignored.  I don't know about
non unixoid systems, though.

cheers
sascha
-- 
Parents strongly cautioned  --  this  posting  is  intended for mature
audiences  over  18.  It  may  contain some material that many parents
would not find suitable for children and may include intense violence,
sexual situations, coarse language and suggestive dialogue.

[-- Attachment #1.2: Type: application/pgp-signature, Size: 189 bytes --]

[-- Attachment #2: Type: text/plain, Size: 142 bytes --]

_______________________________________________
Emacs-devel mailing list
Emacs-devel@gnu.org
http://lists.gnu.org/mailman/listinfo/emacs-devel

Re: Can't interrupt directory_files_internal run from timer-event-handler

[Richard M. Stallman]

    Yes, indeed Unix v5 (fifth edition from 1974, that is _not_ SysV!),
    does allow root to make hard links on directories.  But the seventh
    edition from 1978 already prohibits it.  (sorry, at the moment I don't
    have a v6 at hand to test it there...  ;-))

Thanks for looking this up.