bug#29450: 26.0.90; No check for nil in some filenotify functions

bug#29450: 26.0.90; No check for nil in some filenotify functions

[John Wiegley]

The documentation for find-file-name-handler says:

    find-file-name-handler is a built-in function in ‘C source code’.
    
    (find-file-name-handler FILENAME OPERATION)
    
    Return FILENAME’s handler function for OPERATION, if it has one.
    Otherwise, return nil.

However, several of the functions in filenotify use the return value of this
function without checking if it's nil or not:

    (defun file-notify-rm-watch (descriptor)
      "Remove an existing watch specified by its DESCRIPTOR.
    DESCRIPTOR should be an object returned by `file-notify-add-watch'."
      (when-let* ((watch (gethash descriptor file-notify-descriptors)))
        (let ((handler (find-file-name-handler
                        (file-notify--watch-directory watch)
                        'file-notify-rm-watch)))
          (condition-case nil
              (if handler
                  ;; A file name handler could exist even if there is no
                  ;; local file notification support.
                  (funcall handler 'file-notify-rm-watch descriptor)

I've been getting several errors with a backtrace like nil(48). This is likely
because some package has done something wrong, but even still, filenotify
should be more defensive.

-- 
John Wiegley                  GPG fingerprint = 4710 CF98 AF9B 327B B80F
http://newartisans.com                          60E1 46C4 BD1A 7AC1 4BA2

bug#29450: 26.0.90; No check for nil in some filenotify functions

[Eli Zaretskii]

> From: "John Wiegley" <johnw@gnu.org>
> Date: Sat, 25 Nov 2017 22:18:07 -0800
> 
> The documentation for find-file-name-handler says:
> 
>     find-file-name-handler is a built-in function in ‘C source code’.
>     
>     (find-file-name-handler FILENAME OPERATION)
>     
>     Return FILENAME’s handler function for OPERATION, if it has one.
>     Otherwise, return nil.
> 
> However, several of the functions in filenotify use the return value of this
> function without checking if it's nil or not:

Maybe I'm blind, but I cannot find any place in filenotify where the
handler is used without checking, including in the case you show:

>     (defun file-notify-rm-watch (descriptor)
>       "Remove an existing watch specified by its DESCRIPTOR.
>     DESCRIPTOR should be an object returned by `file-notify-add-watch'."
>       (when-let* ((watch (gethash descriptor file-notify-descriptors)))
>         (let ((handler (find-file-name-handler
>                         (file-notify--watch-directory watch)
>                         'file-notify-rm-watch)))
>           (condition-case nil
>               (if handler <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
>                   ;; A file name handler could exist even if there is no
>                   ;; local file notification support.
>                   (funcall handler 'file-notify-rm-watch descriptor)

What am I missing?

> I've been getting several errors with a backtrace like nil(48). This is likely
> because some package has done something wrong, but even still, filenotify
> should be more defensive.

Can you show a full backtrace like that?

bug#29450: 26.0.90; No check for nil in some filenotify functions

[Glenn Morris]

Eli Zaretskii wrote:

>> From: "John Wiegley" <johnw@gnu.org>
>> Date: Sat, 25 Nov 2017 22:18:07 -0800
>> 
>> The documentation for find-file-name-handler says:
>> 
>>     find-file-name-handler is a built-in function in ‘C source code’.
>>     
>>     (find-file-name-handler FILENAME OPERATION)
>>     
>>     Return FILENAME’s handler function for OPERATION, if it has one.
>>     Otherwise, return nil.
>> 
>> However, several of the functions in filenotify use the return value of this
>> function without checking if it's nil or not:
>
> Maybe I'm blind, but I cannot find any place in filenotify where the
> handler is used without checking, including in the case you show:
>
>>     (defun file-notify-rm-watch (descriptor)
>>       "Remove an existing watch specified by its DESCRIPTOR.
>>     DESCRIPTOR should be an object returned by `file-notify-add-watch'."
>>       (when-let* ((watch (gethash descriptor file-notify-descriptors)))
>>         (let ((handler (find-file-name-handler
>>                         (file-notify--watch-directory watch)
>>                         'file-notify-rm-watch)))
>>           (condition-case nil
>>               (if handler <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
>>                   ;; A file name handler could exist even if there is no
>>                   ;; local file notification support.
>>                   (funcall handler 'file-notify-rm-watch descriptor)
>
> What am I missing?
>
>> I've been getting several errors with a backtrace like nil(48). This is likely
>> because some package has done something wrong, but even still, filenotify
>> should be more defensive.
>
> Can you show a full backtrace like that?


Any update?

bug#29450: 26.0.90; No check for nil in some filenotify functions

[John Wiegley]

>>>>> "GM" == Glenn Morris <rgm@gnu.org> writes:

>>> I've been getting several errors with a backtrace like nil(48). This is
>>> likely because some package has done something wrong, but even still,
>>> filenotify should be more defensive.
>> 
>> Can you show a full backtrace like that?

GM> Any update?

I worked around this with (setq auto-revert-use-notify nil). If it doesn't
happen for anyone else, you can close it. I'll look into it again if I ever
decide to re-enable this.

-- 
John Wiegley                  GPG fingerprint = 4710 CF98 AF9B 327B B80F
http://newartisans.com                          60E1 46C4 BD1A 7AC1 4BA2

bug#29450: 26.0.90; No check for nil in some filenotify functions

[Michael Albinus]

John Wiegley <johnw@gnu.org> writes:

Hi,

Sorry to jump in such late; I've overlooked this thread before.

>>>> I've been getting several errors with a backtrace like nil(48). This is
>>>> likely because some package has done something wrong, but even still,
>>>> filenotify should be more defensive.
>>> 
>>> Can you show a full backtrace like that?
>
> I worked around this with (setq auto-revert-use-notify nil). If it doesn't
> happen for anyone else, you can close it. I'll look into it again if I ever
> decide to re-enable this.

I still would like to see a full backtrace. You have quoted

--8<---------------cut here---------------start------------->8---
      (when-let* ((watch (gethash descriptor file-notify-descriptors)))
        (let ((handler (find-file-name-handler
                        (file-notify--watch-directory watch)
                        'file-notify-rm-watch)))
 --8<---------------cut here---------------end--------------->8---

I have the feeling that it isn't the return value of
`find-file-name-handler' being nil, but rather the
`file-notify--watch-directory' call.

Best regards, Michael.

bug#29450: 26.0.90; No check for nil in some filenotify functions

[John Wiegley]

>>>>> "MA" == Michael Albinus <michael.albinus@gmx.de> writes:

MA> I have the feeling that it isn't the return value of
MA> `find-file-name-handler' being nil, but rather the
MA> `file-notify--watch-directory' call.

I've had file-notify back on for a few days now, and this has not happened
again, so I'm fine with closing this issue.

However -- and this is a separate bug -- I had to disable it again because it
was stalling out Emacs regularly. I'd ran an operation that changed the state
of the disk, and then Emacs would be locked up for minutes at a time until I
hit C-g repeatedly. Finally, I ran Instruments to see what was going on, and
it was stuck in a busy loop reading directories, spending 92% of its time
calling delq over and over again.

But that's a separate issue, and disabling file-notify cured the performance
problem.

-- 
John Wiegley                  GPG fingerprint = 4710 CF98 AF9B 327B B80F
http://newartisans.com                          60E1 46C4 BD1A 7AC1 4BA2

bug#29450: 26.0.90; No check for nil in some filenotify functions

[Michael Albinus]

John Wiegley <johnw@gnu.org> writes:

Hi John,

> I've had file-notify back on for a few days now, and this has not happened
> again, so I'm fine with closing this issue.

I do it herewith.

> However -- and this is a separate bug -- I had to disable it again because it
> was stalling out Emacs regularly. I'd ran an operation that changed the state
> of the disk, and then Emacs would be locked up for minutes at a time until I
> hit C-g repeatedly. Finally, I ran Instruments to see what was going on, and
> it was stuck in a busy loop reading directories, spending 92% of its time
> calling delq over and over again.

In function `file-notify-handle-event', there's a commented `message'
call. Could you enable this, and send the output when this problem
happens, again?

> But that's a separate issue, and disabling file-notify cured the performance
> problem.

Do you write another bug report for this?

Best regards, Michael.