Stefan Monnier <monnier@iro.umontreal.ca> writes:

> Sounds good.  I rarely use large CSV files, but I know the operation is slow.
>
> I'm OK with the patch, tho please see my comment below.

Thanks for reviewing it.

> 40s is still slow, but a factor of 2 is good, thanks.

Yes (though 40s is the time for all three benchmark runs, so one
alignment is 40s/3).

> If you're interested in this line, I think there are two avenues to
> improve the behavior further:
> - align lazily via jit-lock (this way the time is determined by the
>   amount of text displayed rather than the total file size).

Wouldn't that still depend on knowing the column widths? I find that the
column width computation is taking about 80% of the time when calling
'csv-align-fields' (after the patch).

> - make align-fields' into a mode, where fields are kept aligned even while
>   the buffer is modified.

That sounds nice.

>>  (defun csv--column-widths ()
>> -  (let ((widths '()))
>> +  (let ((column-widths '())
>> +        (field-widths '()))
>
> I think the return value is now sufficiently complex that the function
> deserves a docstring describing it.

Agreed, I'll add one before I install the patch.

I've also attached a new suggestion for speeding up the column width
computation itself by eliminating another 'current-column'-call. I'm not
too sure about its correctness yet, but it seems to work in a few tests
I've done, and it sped up 'csv--column-widths' by a factor of 1.3–1.4.