[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
bug#67900: 30.0.50; Emacs Crahes When Executing Command `consult-buffer'
From: |
Andrea Corallo |
Subject: |
bug#67900: 30.0.50; Emacs Crahes When Executing Command `consult-buffer' |
Date: |
Fri, 05 Jan 2024 16:46:46 -0500 |
User-agent: |
Gnus/5.13 (Gnus v5.13) |
Chang Xiaoduan <drcxd@sina.com> writes:
> Hello Andrea,
>
> I still doubt the method you suggest to locate the mis-compiled code
> here.
>
> Assume we have n functions all compiled with optimization level
> 2. Executing one of them, which may call the others, triggers a
> crash. Now if we make one of the function compiled with optimization
> level 1, and the crash can still be reproduced, then can we conclude
> that the function is not involved in the crash? I do not think so. Maybe
> that function and some other function both trigger the crash.
>
> Now assume we have n functions all compiled with optimization level 1
> and no crash. Making one function compiled with optimization level 2 and
> the program crashes. I think now it is safe to conclude that this
> function is involved in the crash.
>
> The following is what I have done:
>
> 1. I mark all `defun' in consult.el with `(declare (speed 1))'. I can
> still reproduce the crash.
>
> 2. In addition to step 1, I mark all `cl-defun' in consult.el with
> `(declare (speed 1))'. I can still reproduce the crash.
>
> 3. In addition to step 2, I mark all `defmacro' in consult.el with
> `(declare (speed 1))'. I do not know if this works with `defmacro' or
> not, I do it anyway. I can still reproduce the crash. However, I notice
> one of the `defmacro' is somehow special:
>
> #+begin_src eamcs-lisp
> (defmacro consult--define-state (type)
> "Define state function for TYPE."
> (declare (speed 1))
> `(defun ,(intern (format "consult--%s-state" type)) ()
> ,(format "State function for %ss with preview.
> The result can be passed as :state argument to `consult--read'." type)
> (consult--state-with-return (,(intern (format "consult--%s-preview"
> type)))
> #',(intern (format "consult--%s-action"
> type)))))
> #+end_src
>
>
>>From what I know about "macro" in C, this will expand to a function
> definition. I assume this is also true in Emacs Lisp, so:
>
> 4. In addition to step 3, I add the `declare' form in the macro
> definition, and now the code is:
>
> #+begin_src emacs-lisp
> (defmacro consult--define-state (type)
> "Define state function for TYPE."
> (declare (speed 1))
> `(defun ,(intern (format "consult--%s-state" type)) ()
> ,(format "State function for %ss with preview.
> The result can be passed as :state argument to `consult--read'." type)
> (declare (speed 1))
> (consult--state-with-return (,(intern (format "consult--%s-preview"
> type)))
> #',(intern (format "consult--%s-action"
> type)))))
> #+end_src
>
> Only after step 4, I can not reproduce the crash. If I regress to step
> 3, then the crash is reproducible. Thus, I think *at least* the function
> generated using this macro is involved in the crash. What do you think
> about it?
I think you did what I suggested, adding declares to narrow down the
issue to defuns. The fact that some defun is macro generated and that
one of these could have been the misscompiled one was as well suggested
by me in this thread.
I think the best now is to macro expand all those function and keep on
using the suggested declare method to narrow down exactly which of the
generated functions is the problematic one.
Thanks
Andrea