Speed up extraction of global props, headlines, and title

[kisaragi-hiu]

Motivation for this change

Org-roam takes a long time to process files. In addition, big files (10000+ lines) can take several seconds to process after it is modified. This becomes unbearable for eg. habit (org-habit) tracking files.

This PR introduces some optimizations:

• Replace unnecessary uses of org-element-parse-buffer
  • org-roam--extract-headlines: Use org-map-region and org-entry-get instead of org-element-parse-buffer.
  • org-roam--extract-titles-headline: Find the first headline with org-outline-regexp-bol instead of using org-element-parse-buffer.
• If it's available, use org-collect-keywords which is a lot faster as it also avoids org-element-parse-buffer. (As it is a relatively new function in Org, added in April 2020, I have kept the previous implementation if it's not present.)

Performance comparison

vocab.org is an Org file with 14795 lines (1052 headings).

I'm running this on a Ryzen 1400 CPU.

(with-current-buffer "vocab.org"
  (list
   (list 'total 'gc-count 'gc-time)
   (benchmark-run 1
     (org-roam--extract-headlines))
   (benchmark-run 1
     (org-roam--extract-titles-title))
   (benchmark-run 1
     (org-roam--extract-titles-headline))
   (benchmark-run 1
     (org-roam--extract-titles-alias))))

Before:

total	gc-count	gc-time
1.771265479	1	0.7655653740000048
0.932756371	0	0.0
1.645399532	1	0.7050031620000254
1.701283592	1	0.742148186999998

After:

total	gc-count	gc-time
0.149613758	0	0.0
0.00223786	0	0.0
8.2539e-05	0	0.0
0.001961503	0	0.0

From 5 seconds down to 0.15 seconds.

[jethrokuan]

The changes LGTM.

It looks from the tests like the headline extraction now returns the candidates in reverse order, so the tests also need to be fixed here.

[kisaragi-hiu]

It looks from the tests like the headline extraction now returns the candidates in reverse order, so the tests also need to be fixed here.

Ah, yes, because push adds to the front. I've made it so that the test uses :to-have-same-items-as to ignore the ordering of its items.

[jethrokuan]

LGTM, very nice 😄

[jethrokuan]

Related to #834 #940