bug#44273: "total used in directory 19 available 5.2 GiB"

[Michael Albinus]

Eli Zaretskii <eliz@gnu.org> writes:

>> True.  If `ls-lisp' (or something similar) isn't fast enough, then it
>> should be pretty easy to add a C-level function for efficient directory
>> reading.
>
> Are you sure it's reading the directory that takes the lion's share of
> the time?  Maybe it's inserting the file information, one file at a
> time, into a buffer?  Or something else?
>
> That's why I suggested to profile and benchmark this thing.
>
>>  (But we'd still need the parser for Tramp.)
>
> Tramp could always use what it does now, or use ls-lisp, or something
> else.

Counting in the sources, there are already 4 different implementations
of insert-directory in Tramp. One of it uses ls-lisp, another uses a
remote "ls" call.

If there shall be a reimplementation, I would even consider to add
another implementation, based on Perl or Python. This shall minimize
performance penalties.

Best regards, Michael.