make linkChild idempotent

[stefanpenner]

• find related tests
• add a test to ensure we don't regress

while looking at @sandstrom https://github.com/sandstrom/initial-render-components I noticed, copious amounts of u needed changes events where happening.

That example although synthetic, does indicate death a a thousand papercuts scenario. With a large 15 -> 20% GC cycle. GC is just the result of us allocating to much.

Anyways, back on topic. to many change events where happening.These changes have additional side-affects, but the main one is sendEvent itself, which seems to have drop by 15% -> 25% with this change.

I really wouldn't call home about this one, but it helps a teeny tiny bit.

[stefanpenner]

@tomdale / @wycats

[wycats]

r+

[sandstrom]

Great! I've updated the repo to use this change.

I really appreciate that you are looking into this! It'll make the views-to-components paradigm better in apps with a lot of components.

(for people stumbling upon this commit, more details can be found here: #11502 (comment))

[stefanpenner]

I really appreciate that you are looking into this! It'll make the views-to-components paradigm better in apps with a lot of components.

np. The large chunk of this type of waste still needs to be addressed (the attrs object, and related backwards compat) I have been noodling on it, and think there is a way to sort it out.