Question: suppress ask-user-about-lock?

Question: suppress ask-user-about-lock?

[Qiantan Hong]

Is there a standard way to cause lock-buffer immediately fail?

(let ((noninteractive t)) (lock-buffer)) is an obvious one,
but not sure if it’s good practice.


Best,
Qiantan

Re: Question: suppress ask-user-about-lock?

[Michael Albinus]

Qiantan Hong <qhong@mit.edu> writes:

Hi,

> Is there a standard way to cause lock-buffer immediately fail?

Do you mean with "fail" that lock-buffer shall do nothing?

> (let ((noninteractive t)) (lock-buffer)) is an obvious one,
> but not sure if it’s good practice.

Set or bind create-lockfiles to nil.

> Best,
> Qiantan

Best regards, Michael.

Re: Question: suppress ask-user-about-lock?

[Qiantan Hong]

>> Is there a standard way to cause lock-buffer immediately fail?
> 
> Do you mean with "fail" that lock-buffer shall do nothing?
No, I mean it should signal ‘file-locked without asking the user
(through ask-user-about-lock).
I intended to use it as a mutex in Elisp program.

Re: Question: suppress ask-user-about-lock?

[Michael Albinus]

Qiantan Hong <qhong@mit.edu> writes:

Hi,

>>> Is there a standard way to cause lock-buffer immediately fail?
>> 
>> Do you mean with "fail" that lock-buffer shall do nothing?
> No, I mean it should signal ‘file-locked without asking the user
> (through ask-user-about-lock).
> I intended to use it as a mutex in Elisp program.

What about (untested)

(cl-letf (((symbol-function #'ask-user-about-lock)
           (lambda (file proponent)
             (signal 'file-locked (list file proponent)))))

...)

Best regards, Michael.

Re: Question: suppress ask-user-about-lock?

[Qiantan Hong]

> What about (untested)
> 
> (cl-letf (((symbol-function #'ask-user-about-lock)
>           (lambda (file proponent)
>             (signal 'file-locked (list file proponent)))))
> 
> ...)
I guess it will break under multi-thread environment?
Because FLET, or function cells in general, are not thread local.

I also feel like the above is as hacky as my
(let ((noninteractive t)) (lock-buffer))
Does anyone have more recommendations?


Best,
Qiantan

Re: Question: suppress ask-user-about-lock?

[Eli Zaretskii]

> From: Qiantan Hong <qhong@mit.edu>
> Date: Wed, 8 Dec 2021 19:22:44 +0000
> Cc: "emacs-devel@gnu.org" <emacs-devel@gnu.org>
> 
> No, I mean it should signal ‘file-locked without asking the user
> (through ask-user-about-lock).
> I intended to use it as a mutex in Elisp program.

But we already have mutexes in Emacs, so why not use the real thing?
What am I missing?

Re: Question: suppress ask-user-about-lock?

[Michael Albinus]

Qiantan Hong <qhong@mit.edu> writes:

> I also feel like the above is as hacky as my
> (let ((noninteractive t)) (lock-buffer))

The docstring of ask-user-about-lock encourages to redefine the
function. There should be less side effects than binding noninteractive.

> Best,
> Qiantan

Best regards, Michael.

Re: Question: suppress ask-user-about-lock?

[Qiantan Hong]

> On Dec 8, 2021, at 12:04 PM, Eli Zaretskii <eliz@gnu.org> wrote:
> 
>> From: Qiantan Hong <qhong@mit.edu>
>> Date: Wed, 8 Dec 2021 19:22:44 +0000
>> Cc: "emacs-devel@gnu.org" <emacs-devel@gnu.org>
>> 
>> No, I mean it should signal ‘file-locked without asking the user
>> (through ask-user-about-lock).
>> I intended to use it as a mutex in Elisp program.
> 
> But we already have mutexes in Emacs, so why not use the real thing?
> What am I missing?
I’m using it to lock a key value store file when writing to it,
so Emacsen have mutual exclusive write access to it.

Seems that “mutex" usually refers to a particular kind of
intraprocess mutual exclusive lock, I apologize for the confusion.
I’m referring to something similar to flock(2).

Re: Question: suppress ask-user-about-lock?

[Matt Armstrong]

Qiantan Hong <qhong@mit.edu> writes:

>> What about (untested)
>> 
>> (cl-letf (((symbol-function #'ask-user-about-lock)
>>           (lambda (file proponent)
>>             (signal 'file-locked (list file proponent)))))
>> 
>> ...)
>
> I guess it will break under multi-thread environment?  Because FLET,
> or function cells in general, are not thread local.

(info "(elisp)Threads") talks about `let' bindings being thread local.
I always took that to apply to the other let-like bindings as well.  It
looks like `cl-letf' uses `let*' under the hood.


> I also feel like the above is as hacky as my (let ((noninteractive t))
> (lock-buffer)) Does anyone have more recommendations?

It think it feels hacky because `ask-user-about-lock' is a decades old
API and was probably introduced before the abnormal hook convention took
hold (info "(elisp)Standard Hooks").  I suspect that back then
redefining functions as a customization point was more common.

If the API were added today I think you'd see an
`ask-user-about-lock-function' (or functions) var that is bound to
#'ask-user-about-lock by default.  The Emacs C layer would then funcall
it.

Re: Question: suppress ask-user-about-lock?

[Qiantan Hong]

> (info "(elisp)Threads") talks about `let' bindings being thread local.
> I always took that to apply to the other let-like bindings as well.  It
> looks like `cl-letf' uses `let*' under the hood.
For LETFing a symbol-function, it FSETs under the hood, which is not
safe. In fact it is impossible to do it using thread-safe let only,
because code outside of the CL-LETF calls the function from
the function cell, and to have your binding affect them the only
way is to set the thread-unsafe function cell.