Consolidate and update Borsh for TS/JS

[MaksymZavershynskyi]

Currently we have three different implementation of TS/JS that have slightly different functionality and bugs:

• https://github.com/nearprotocol/borsh
• https://github.com/nearprotocol/nearlib/blob/master/src.ts/utils/serialize.ts
• https://github.com/nearprotocol/near-bridge/blob/master/ethrelay/index.js

We should unify them into one implementation that lives in https://github.com/nearprotocol/borsh .

Also, near/borsh#57 and near/borsh#56 introduced a slightly modified Borsh schema format which now allows serializing any Rust types, even with generics, and it is not Rust-specific. We should make sure we follow this new schema format.

Note, as requested by @vgrichina Borsh Schema is in Borsh format rather than JSON.

[kcole16]

@nearmax Is this the correct list of TODOs?

• Update borsh-ts and borsh-js to the new schema
• Replace usage of nearlib/utils/serialize with borsh-ts
• Replace custom serializer in near-bridge/eth-relay with borsh-js

[vgrichina]

There are few important points that shouldn't be missed:

• borsh-ts is most out of date and untested implementation, but it ultimately should be destination where new implementation lands
• https://github.com/nearprotocol/nearlib/blob/master/src.ts/utils/serialize.ts likely has the best error handling (e.g. keeping track of field path so that you can see where exactly stuff fails:

  near-api-js/src.ts/utils/serialize.ts

  Line 287 in 092d647

  error.addToFieldPath(fieldName);

  ). It also uses js-fuzz which helps to find bugs with implementation. Likely nearlib is the best starting point for new implementation.
• https://github.com/nearprotocol/near-bridge/blob/master/ethrelay/index.js improves one thing which is important to fix elsewhere – Borsh schema doesn't depend on JS constructors anymore (fragile and breaks in many weird environments). Compare https://github.com/nearprotocol/near-bridge/blob/75fab60c92b672a9f86e633a2ea66bdad5af8bfb/ethrelay/index.js#L301 and

  near-api-js/src.ts/transaction.ts

  Line 126 in 092d647

  export const SCHEMA = new Map<Function, any>([

[MaksymZavershynskyi]

@kcole16 I don't know how I missed your comment.

@nearmax Is this the correct list of TODOs?

• Update borsh-ts and borsh-js to the new schema
• Replace usage of nearlib/utils/serialize with borsh-ts
• Replace custom serializer in near-bridge/eth-relay with borsh-js

Yes, correct.

@vgrichina

borsh-ts is most out of date and untested implementation, but it ultimately should be destination where new implementation lands

I agree.

improves one thing which is important to fix elsewhere – Borsh schema doesn't depend on JS constructors anymore

This is an important improvement, which will allow us in the future to directly convert JSON to Borsh using Borsh schema

---

@vgrichina Please also note that this is not blocking our Mainnet launch, so we have removed it from Mainnet release in Zenhub last week. If you have other items that are blocking the Mainnet launch then you might consider them more important.

[vgrichina]

@nearmax re mainnnet priorities we likely need to do smth about RPCs. Looks like long polling on nearcore side works bad (e. g. tx-status for non existing hash). This will require changes both in nearcore and in nearlib, though should be pretty localized.

[MaksymZavershynskyi]

@nearmax re mainnnet priorities we likely need to do smth about RPCs. Looks like long polling on nearcore side works bad (e. g. tx-status for non existing hash). This will require changes both in nearcore and in nearlib, though should be pretty localized.

Yes, but these changes look orthogonal to Borsh changes. CC @frol

[frol]

1. This issue indeed looks orthogonal to the RPC.
2. I agree that RPC and the whole node need benchmarks () and load testing. We have tracking issues for those and will work on those items in the near future.
3. RPC polling is never a good idea, and we will need to implement some streaming APIs, but this can be done after the mainnet, I believe.

[MaksymZavershynskyi]

@kcole16 , @vgrichina We need to bump priority on this. Flux has been asking examples on JS-side Borsh serialization, which we cannot do until we consolidate different implementations.

[MaksymZavershynskyi]

Requires: near/borsh#71

[volovyks]

@vgrichina @nearmax this issues is a bit outdated

• Update borsh-ts and borsh-js to the new schema
• Replace usage of nearlib/utils/serialize with borsh-ts
• Replace custom serializer in near-bridge/eth-relay with borsh-js (has it's own issue here: RB should use borsh-js library instead of internal util. Near-One/rainbow-bridge#438)

should we consider Update borsh-ts and borsh-js to the new schema as a P1?

[vgrichina]

@volovyk-s not sure it’s a P1 but it’s relatively important as would solve the problem with using multiple near-api-js instances in your app. Basically the idea is to avoid relying on constructor instances to identify schema.

[volovyks]

Closing this issue since it's not relevant to near-api-js anymore. borsh-js schema should be fixed in near/borsh-js#10