Implement PrimaryKey for all &T where T implements PrimaryKey.

[0xekez]

    for<'a> &'a (K, u64): PrimaryKey<'a>,

This allows having type constraints like this for types that wrap storage-plus. In english: for all lifetimes 'a &'a (K, u64) is a valid key. I need this generic in order to implement cw-future-map.

[0xekez]

i am going to fork pending some resolution to how to land PRs into CosmWasm. no bad vibes, and i'd like to continue to contribute to and work with the CosmWasm repositories when there is a way to do so again.

relevant PRs:

• Add distribution module query type cosmwasm#1593
• Allow piecewise linear curves with non-zero y start values. wynddao/wynddao#4
• Add message SetWithdrawAddress to distribution module cw-multi-test#23
• Implement PrimaryKey for all &T where T implements PrimaryKey. #30

[webmaster128]

Thank you for this PR.

To be honest, I am not entirely sure if using references for primary keys is a good thing. What's the best practive for e.g. Map<Addr> vs. Map<&Addr>? You will not be able to deserialize into a key that is a reference.

@maurolacy do you know why both impl<'a> PrimaryKey<'a> for Addr and impl<'a> PrimaryKey<'a> for &'a Addr exist and if that's desired?

[0xekez]

@webmaster128 my understanding is that because serialization takes a reference to the type being serialized, it is more efficient to use references as keys as (1) the underlying serialization doesn't change and (2) you can avoid moves (and in turn clones) when calling methods on the map.

[chipshort]

Thanks, nice generalization.
Only change I'd require is the borrow because clippy will complain about it.

[webmaster128]

Not sure what kind of issues the CI has or had. Somehow the GitHub connection got lost. But in #31 it worked today, also for fork PRs.

When the changes requested by @chipshort are in and the CI is happy this is good to go IMO.

[maurolacy]

@webmaster128 my understanding is that because serialization takes a reference to the type being serialized, it is more efficient to use references as keys as (1) the underlying serialization doesn't change and (2) you can avoid moves (and in turn clones) when calling methods on the map.

This is correct. Using references is a good option for serialising stuff, while deserialization is always to an owned type, so, using references doesn't impact it.

Just seen this. Will take a detailed look and comment back.

[0xekez]

for some additional context, here is the map that i was able to make with these changes (we forked to include this change while the ATOM proposal was pending): cw-wormhole.

@chipshort - I've made your requested changes in the most recent commit. thank you!

[maurolacy]

LGTM.

[maurolacy]

As mentioned, the type cast is not needed.

[maurolacy]

Suggested change

	type Prefix = <T as PrimaryKey<'a>>::Prefix;
	type SubPrefix = <T as PrimaryKey<'a>>::SubPrefix;
	type Prefix = T::Prefix;
	type SubPrefix = T::SubPrefix;

[maurolacy]

Same here:

Suggested change

	<T as Prefixer<'a>>::prefix(self)
	T::prefix(self)

[maurolacy]

And here:

Suggested change

	<T as PrimaryKey<'a>>::key(self)
	T::key(self)