PR #4147 <https://github.com/leo-editor/leo-editor/pull/4147> fixes a
medium-scale performance bug. While doing so, I discovered two ways of
collapsing the complexity of the colorizer.
*Naive expectations*
We have the right to expect that the qsh will work as follows:
- When changing nodes, the qsh should call _redraw for *every *line of p.b.
- Otherwise, we expect the qsh to issue highly optimized calls to _redraw
when users edit text.
In fact, the qsh called _redraw for *all *lines of p.b whenever the user
changed p.b. Ouch!
The PR identifies two culprits as the cause of the performance bug.
*jedit.colorizer*
Previously, this method *always* forced a full redraw. That makes no sense!
Now it contains the following new code:
# #4146: Fully recolor p.b *only* if necessary.
old_language = self.language
self.updateSyntaxColorer(p)
if p.v == self.old_v and self.language == old_language:
return
*jedit._redraw*
Imagine my surprise when I found the following code at the start of
jedit._redraw:
block_n = self.currentBlockNumber()
n = self.prevState()
if p.v == self.old_v:
new_language = self.n2languageDict.get(n)
if new_language != self.language:
self.language = new_language
self.init()
else:
self.updateSyntaxColorer(p) # Force a full recolor
assert self.language
self.init_all_state(p.v)
self.init()
Huh??? My intense study of the code showed me that jedit._redraw was a
callback for qsh.highlightBlock. *Nothing else should ever be calling
jedit._redraw.* That's why I added the leading underscore.
Forcing the qsh to do a full redraw inside _redraw is utterly misguided. It
is a gross interference with the qsh!
I immediately removed this bizarre code. All of a sudden, for the first
time in Leo's history, the qsh worked as one would naively expect!
*Summary*
How did Leo's legacy syntax colorer have *any* chance of working? It's a
mystery.
I fixed the two culprits as follows:
- jedit.colorize forces a full redraw only when Leo changes nodes.
- jedit._redraw *never* forces a full redraw.
After these changes, the qsh operates exactly as one would naively expect.
Edward
--
You received this message because you are subscribed to the Google Groups
"leo-editor" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion visit
https://groups.google.com/d/msgid/leo-editor/29de1dbd-94b8-4d7f-b760-a2787874a727n%40googlegroups.com.
