For element nodes with a single text child, directly set textContent
#3939
+119
−118
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
📊 Tachometer Benchmark ResultsSummaryduration
usedJSHeapSize
Results02_replace1k
duration
usedJSHeapSize
run-warmup-0
run-warmup-1
run-warmup-2
run-warmup-3
run-warmup-4
run-final
03_update10th1k_x16
duration
usedJSHeapSize
07_create10k
duration
usedJSHeapSize
filter_list
duration
usedJSHeapSize
hydrate1k
duration
usedJSHeapSize
many_updates
duration
usedJSHeapSize
text_update
duration
usedJSHeapSize
todo
duration
usedJSHeapSize
|
|
Size Change: +374 B (0%) Total Size: 54.7 kB
ℹ️ View Unchanged
|
andrewiggins
commented
Mar 17, 2023
Comment on lines
-1674
to
-1677
| '<div>.appendChild(#text)', | ||
| '<div>1.insertBefore(<div>3, #text)', | ||
| '<div>.appendChild(#text)', | ||
| '<div>31.insertBefore(<div>4, #text)', |
There was a problem hiding this comment.
Most of the test changes here are just removing calls to appendChild(#text). But this test case invokes something more interesting I think it worth examining.
Here, we are transitioning from a tree like:
<div>1</div>
<div>2</div>to
<div class="wrapper">
<div>3</div>
<div>4</div>
</div>Our diff matches <div>1 with the new div.wrapper. So it sees a div with a single text child (1) transitioning to a div with multiple children <div>3 and <div>4. So with this update, instead of creating and inserting the new children before the old children, it removes the text node and appends the new children to the now empty div.
Comment on lines
-1689
to
-1695
| [ | ||
| '<div>34.insertBefore(#text, <div>3)', | ||
| '<div>4.remove()', | ||
| '<div>3.remove()', | ||
| '<div>.appendChild(#text)', | ||
| '<div>1.appendChild(<div>2)' | ||
| ], |
There was a problem hiding this comment.
Here, we are transition from a tree like:
<div class="wrapper">
<div>3</div>
<div>4</div>
</div>to
<div>1</div>
<div>2</div>Our diff matches the old div.wrapper with the new <div>1 node. So it sees a div with multiple children transitioning to a div with a single text child. To handle this, it clears away the old children by setting the textContent of div.wrapper to the new single text child. That change is why you no longer see the calls to .remove(). Also since we set textContent to the new text child, you don't see the relevant insertBefore call of the text node. In one operation we unmount old children and set new children.
We do still call unmount on old children so any lifecycle methods are called, but no DOM is removed in those calls since setting textContent will take care of that.
|
I think the regression on 02_replace1k is due to changes in how V8 is optimizing the code. Check out the trace logs below: preact-local: The differences in colors under the red underlined area show that V8's optimization heuristics are treating each of these differently. I don't entirely understand the difference but running this benchmark locally and comparing the average duration of all the runs (warmups + final test run) shows that there isn't really a significant difference between on average between the two: |
JoviDeCroock
approved these changes
Mar 18, 2023
andrewiggins
force-pushed
the
skip-append-for-element-text-child
branch
from
5242ac0
to
2a3bba9
Compare
marvinhagemeister
approved these changes
Apr 4, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds a (hopefully) fast path for element VNodes with a single text child. Given how common this kind of tree structure (after all, UI is ultimately about fancy ways to display text lol), I hypothesize this should improve our rendering runtime.
Two significant internal changes that comes out of this are:
Single text children no longer are represented as VNodes
With this change, single text children no longer get a VNode created representing them. Hopefully this will help contribute to better runtime and memory performance by removing lots of allocations. But it does have implications for the VNode tree (which leads to the next point).
Elements with a single text child have a
null_childrenpropertySince single text childs no longer have a VNode created, the
_childrenproperty of their parents is empty (akanull). Inspecting the tree VNode tree just using_childrenwould then lead to an incomplete tree. We could fix this by adding the string to the_childrenarray, but then that breaks the assumption the codebase has had for a while that_childrenis alwaysArray<VNode | null>. But perhaps that is a better approach?For now, I'm going to try this implementation and see if any performance improvements are observable. If so, then we can further pursue what the right thing do here is.