`org-table-align' may benefit from recent `org-table-to-lisp'
speed-up.
The current code contains the following lines in org-table.el. They
parse the table into a Lisp structure, which is precisely what
`org-table-to-lisp' is supposed to do.
(fields (mapcar
(lambda (l)
(and (not (string-match-p org-table-hline-regexp l))
(org-split-string l "[ \t]*|[ \t]*")))
(split-string (buffer-substring beg end) "\n" t)))
Just replace them by a call to `org-table-to-lisp':
(fields (cl-loop for row in (org-table-to-lisp)
collect (and (listp row) row)))
I made a benchmark on a large table (10000 rows, 6 columns). Adding a
column, then removing it, M-S-right M-S-left. I measured time and
memory with profiler-*, taking care to call `garbage-collect'
beforehand. Conclusions:
- `org-table-align' is faster, overall operation is faster
- there is much less garbage collection
- memory allocation is significantly lower
| | current | new | % |
|----------------------------------+-----------+-----------+-----|
| org-table-align CPU samples | 1313 | 1102 | 16. |
| overall CPU samples | 2040 | 1628 | 20. |
| GC CPU samples | 4283 | 2609 | 39. |
| overall allocated bytes | 622087730 | 272436087 | 56. |
| org-table-align allocated bytes | 445910318 | 146646065 | 67. |
#+TBLFM: $4=100*($-2-$-1)/$-2;f0
Note: running twice the same bench yields slightly different results.
Have fun
Thierry