Re: [Emacs-diffs] master fbda511: Fix cps--gensym

Re: [Emacs-diffs] master fbda511: Fix cps--gensym

[Stefan Monnier]

>    ;; Change this function to use `cl-gensym' if you want the generated
>    ;; code to be easier to read and debug.
>    ;; (cl-gensym (apply #'format fmt args))
> -  `(make-symbol ,fmt))
> +  `(make-symbol (format ,fmt . ,args)))
 
Why?  As the comment above says, if you want the more readable names,
you should use the other line of code.  The default one is aimed at
being more efficient (both cpu-use-wise and memory-use-wise),


        Stefan

Re: [Emacs-diffs] master fbda511: Fix cps--gensym

[Leo Liu]

On 2015-05-15 00:00 +0800, Stefan Monnier wrote:
> Why?  As the comment above says, if you want the more readable names,
> you should use the other line of code.  The default one is aimed at
> being more efficient (both cpu-use-wise and memory-use-wise),

I changed it because the expansion of iter-lambda etc. was difficult to
read. I don't understand the efficiency issue so the change may be a
mistake. Could you elaborate on what the issue is? Thanks.

Leo

Re: [Emacs-diffs] master fbda511: Fix cps--gensym

[Stefan Monnier]

>> Why?  As the comment above says, if you want the more readable names,
>> you should use the other line of code.  The default one is aimed at
>> being more efficient (both cpu-use-wise and memory-use-wise),
> I changed it because the expansion of iter-lambda etc. was difficult to
> read.

Right, and as the comment say, when you want to read the expansion, you
should use the other code (the one that uses cl-gensym+format).

> I don't understand the efficiency issue so the change may be a
> mistake. Could you elaborate on what the issue is? Thanks.

format is already significantly more expensive than make-symbol, so
obviously, make-symbol + format will be even worse.


        Stefan

Re: [Emacs-diffs] master fbda511: Fix cps--gensym

[Leo Liu]

On 2015-05-15 10:05 +0800, Stefan Monnier wrote:
> format is already significantly more expensive than make-symbol, so
> obviously, make-symbol + format will be even worse.

OK but this doesn't impact runtime, right? All the hard work will be
done in compile time.

Leo

Re: [Emacs-diffs] master fbda511: Fix cps--gensym

[Stefan Monnier]

>> format is already significantly more expensive than make-symbol, so
>> obviously, make-symbol + format will be even worse.
> OK but this doesn't impact runtime, right? All the hard work will be
> done in compile time.

No, it only impacts the macroexpansion time/space.  But the point is
that if you don't care about that, then you should use the
cl-gensym version.  The make-symbol version is the "fast path", so making
it fat&slow makes no sense.


        Stefan

Re: [Emacs-diffs] master fbda511: Fix cps--gensym

[Leo Liu]

On 2015-05-16 02:19 +0800, Stefan Monnier wrote:
> No, it only impacts the macroexpansion time/space.  But the point is
> that if you don't care about that, then you should use the
> cl-gensym version.  The make-symbol version is the "fast path", so making
> it fat&slow makes no sense.

OK, Stefan, I've just reverted the change.

Leo