Speed Up Creates & Updates By Memoizing on encode

[vintagepenguin]

Serialization is one of the slowest things that you do in Protobuf. If we already have an encoded message, then we don't need to encode it again so why not memoize it.

@abrandoned @liveh2o @mmmries @brianstien

[abrandoned]

why would we need the attr_accessor? (either of them)

[film42]

I think we might want to say @has_change = true. It becomes more clear to me what we're trying to keep track of. That's just my opinion though.

[vintagepenguin]

@film42 commenting here since the diff you commented on got deleted.

But yeah, I think @encode = nil it is confusing if you're looking at this for sure.

But the goal here it to clear out that instance variable so that when we hit the memoization in the encode method we know that we need to encode the message again because there has been a change and we know that because the setter was called.

[film42]

@vintagepenguin It would be more clear set a flag, or call a bust_encode_cache! method or something. But this seems 👍 .

Looks like you have a 👮 to fix, but then .

[claygoddard]

It seems that with this change, performing operations on repeated fields is no longer handled properly after the memoized encoding has been computed.

a.encode
a.some_repeated_field << "new-value"
a.encode # doesn't contain "new-value"

[liveh2o]

That's definitely a problem, @goddardc. Mind writing up an issue?

[claygoddard]

@liveh2o Sure thing!