Add import/export for client, connection, channel states

[cwgoes]

Then chains can be import-export upgraded & retain clients, connections, channels.

For light client verification to work, the import/export will have to retain the height (e.g. start at n + 1) and the chain ID.

cc @fedekunze

[fedekunze]

which are the types we need to import and export?

[cwgoes]

All of the state in x/ibc, essentially, I don't think there is anything we shouldn't export.

[fedekunze]

so if there are connections are channels open already should we keep them open or recreate the handshakes?

[fedekunze]

ok so we should also write simulations for genesis state too

[cwgoes]

If we end up doing upgrades at height n with the same chain ID I think we should keep the historical info as well.

[fedekunze]

If we end up doing upgrades at height n with the same chain ID I think we should keep the historical info as well.

AFAIK you can't export state or not depending on a condition of the importer chain

[cwgoes]

I just mean "if we end up supporting those kinds of upgrades" (which are strictly better than our current kind of upgrade & so should replace them completely). As I understand this still requires changes in Tendermint.

[cwgoes]

Ref #5935 which might block on this.

[fedekunze]

kk, I'll start working on this. @cwgoes just to confirm, we should only export open connections and channels, right? I also assume all clients will need to be exported and then updated with the new chain ID on InitGenesis.

[fedekunze]

@AdityaSripal suggested to not export frozen clients 👍

[cwgoes]

Hmm, I'm pretty sure we want to export everything - if we don't export a closed channel, for example, the identifier will be freed up, and previously sent packets could possibly be re-played. We don't yet have a particular mechanism for "recovering" closed channels safely (it will require governance intervention), but in the meantime I think it's safest to export/import the entire state.

[fedekunze]

the identifier will be freed up, and previously sent packets could possibly be re-played

how? the timestamp would timeout, right?

[cwgoes]

the identifier will be freed up, and previously sent packets could possibly be re-played

how? the timestamp would timeout, right?

Possibly, but it depends on the timing of the export & the specific heights/timestamps involved.