bug#65251: 30.0.50; Duration in compilation buffer

[Helmut Eller <eller.helmut@gmail.com>]

On Wed, Aug 16 2023, Mattias Engdegård wrote:

>> +;; The time when the compilation started.
>> +(defvar compilation--start-time nil)
>
> What about using defvar-local? Then...
>
>> +        (setq-local compilation--start-time (current-time))
>
> can use plain setq.

Seems to be a matter of taste.  I don't know what the Official Style
Guide says about it.

> And if you use (float-time) here, then... 
>
>> +     (let* ((secs (float-time (time-since compilation--start-time))))
>
> ...this becomes a simple subtraction: (- (float-time) compilation--start-time)
>

But then we couldn't use bignums.  And representing time values as a
pair of bignums is cool.  So cool that it was worth to require libgmp
for Emacs.  Oh wait, current-time still doesn't use bignums.  But when
it switches, it will be sooo cool.

Anyway, ERT uses current-time for ert--stats-start-time and I followed
that example.

>> +       (cond ((< secs 1) (format "%.0fms" (* secs 1000)))
>> +	     ((< secs 10) (format "%.2fs" secs))
>> +	     ((< secs 60) (format "%.1fs" secs))
>> +	     (t (format-seconds "%hh%mm%z%ss" secs)))))
>
> First of all, proper style is to separate the number and unit by a space.
> The 'ms' case isn't very important -- 745 ms is no more readable than
> 0.745 s, probably less so.
> The last case is also less than readable. Something like 3:45:58 would
> be better.

Seems to be a matter of taste.  I copied the style used by Go's Duration
type: https://pkg.go.dev/time#Duration.String

> The reader would also like to know what this new time indication
> means. What about
>
>    ..., duration 34.5 s
>
> or
>
>    ..., 34.5 s elapsed
>
> ?

Seems to be a matter of taste.  ERT prints it like

Ran 10 tests, 10 results as expected, 0 unexpected (2023-08-17 00:29:48+0200, 0.813428 sec) 

and nobody seems to complain.

Helmut