abci: Add domain types #862

hdevalence · 2021-04-14T23:37:55Z

Incomplete implementation of #856

Still to do:

Referenced an issue explaining the need for the change
Updated all relevant documentation in docs
Updated all code comments where relevant
Wrote tests
Updated CHANGELOG.md

hdevalence · 2021-04-14T23:51:12Z

Update 2021-09-10, some of these are no longer relevant, so I've crossed them out.

[ ] I regenerated the protos against tendermint master to avoid having to do extra work later to align with changes that already exist upstream now. The tests seem to still pass, but if the idea is that tendermint-rs is tracking a specific release I guess this PR will have to wait.
I commented out some doctests that failed locally prior to making any changes -- not sure whether that's a local machine problem, or if the CI isn't running the doctests.
The conversions between domain types and protos have a lot of &'static str errors. I'm not sure what the crate error modeling strategy is like, so that could be changed in this PR or in a later one. After looking at Retiring anomaly #669 , I think the best thing would be to leave errors as &'static str, and have that be changed as part of a forthcoming error handling overhaul.
~~I still need to fill in the request types and arrange those nicely, and touch up the docs.~~
There's existing stub code in the abci module that seems to be used for handling some RPC calls, and it should be replaced by complete modeling of the ABCI types.
- This requires defining JSON serialization for the new domain types. This serialization format is different from the proto format for historical reasons. I think the best way to support this is not to put serde attributes on the domain types themselves, but to inject them into the generated tendermint_proto types and use TryFrom/From attributes on the domain types. This ensures that all serialization details are in one place.
- To avoid frontloading this work onto this PR, which is large enough as is, this code decouples the RPC implementation from the ABCI module by coping the old ABCI-for-RPC types into the RPC crate.
Some of the domain types could be further refined to make use of Tendermint structures, like AppHash, block::Height, etc.
[ ] There are a bunch of *Params structures that aren't properly part of ABCI, but come from other proto definitions. Presumably this means they shouldn't live in the abci module, but somewhere else. I'm not sure where. Hopefully I didn't duplicate anything, but I didn't see any existing ones.
There are no tests for the conversions. Ideally, it would be possible to use proptest to generate randomized messages and check the serialization round-trips, but I'm not sure if that infrastructure already exists. If it doesn't, I would not like to implement a bunch of one-off custom serialization tests, since I think that doing that has relatively low assurance vs effort (compared to, e.g., just using the code to talk to an ABCI client).

thanethomson

Thanks so much for all this work! 🎉 I've done a first pass here and had a few comments/questions, but it's already looking like it's in really good shape.

tendermint/src/abci/kind.rs

tendermint/src/abci/response.rs

tendermint/src/abci/types.rs

cwgoes · 2021-06-22T13:40:19Z

Bump. What's the status here / is there anything we can help with? (ref penumbra-zone#1)

thanethomson · 2021-06-23T12:13:08Z

Bump. What's the status here / is there anything we can help with? (ref hdevalence#1)

Also happy to help on this!

hdevalence · 2021-07-28T23:10:10Z

Hey, sorry for the delay -- I wasn't able to pick it up again in June after the initial review because of some other work, but I've rebased it on top of the current state of the master branch to get ready to drive it to completion.

hdevalence · 2021-08-05T06:01:53Z

Update:

I regenerated the protos against the current upstream master rather than the master as of mid-April when I opened the PR, and fixed any inconsistencies the upstream proto changes introduced relative to the rest of tendermint-rs.
I decoupled the original ABCI-for-RPC types from the new ABCI-for-ABCI types by copying them to the RPC crate. While this means the RPC crate now has duplicate definitions of ABCI types, it means that that code can be refactored with the ABCI code in chunks, rather than getting it all done in this PR (which is big enough as is, and has been unmerged for long enough to start risking problems -- sorry about that!)

Remaining work:

Another pass over the ABCI types to catch types that could be refined (e.g., sticking in Power, AppHash, etc
Aligning the original ABCI types with the new ones (without respect to the RPC code). For instance, using the response code structure, etc.

I'll post an update when I've finished those two and this is ready for re-review.

hdevalence · 2021-09-09T23:48:28Z

I've rebased this branch on top of current master to be able to get it to completion.

Currently, it's broken, because the conversions between the domain types and the proto types used &'static str errors, but it's no longer possible to use a &'static str as a source error.

Before trying to correct this, I'd like to make sure I understand how the error modeling is supposed to work. I looked at the flex-error docs, but it seems to be an abstraction over other error crates, so I'm not sure what the intended goal is or how to use it. The change to flex-error was in PR #923, which references discussion about an error overhaul in #669, but that issue just discusses some possibilities, and doesn't have any discussion motivating the creation of flex-error.

thanethomson · 2021-09-10T12:28:09Z

Before trying to correct this, I'd like to make sure I understand how the error modeling is supposed to work. I looked at the flex-error docs, but it seems to be an abstraction over other error crates, so I'm not sure what the intended goal is or how to use it. The change to flex-error was in PR #923, which references discussion about an error overhaul in #669, but that issue just discusses some possibilities, and doesn't have any discussion motivating the creation of flex-error.

Sure I can give some context there! flex-error does, as you say, abstract over other error handling crates, and our main goals there are:

Harmonizing the error handling approach across both tendermint-rs and ibc-rs
no_std support for as many crates in tendermint-rs and ibc-rs as possible/necessary

This is useful until informalsystems/tendermint-rs#862 is merged upstream.

The BlockParams and ConsensusParams structs moved out of the ABCI protos.

This fixes a compile error introduced by upstream proto changes that add an Sr25519 variant.

The previous data modeling allowed a user to construct an `Err(0)` value that would be serialized and deserialized as `Ok`.

This changes the Serialize/Deserialize implementations for Block to convert to/from the proto-generated `RawBlock`, and use the derived serialization for that type. This is much cleaner and more maintainable than keeping Serde annotations for each sub-member of the data structure, because all of the serialization code is kept in one place, and there's only one validation path (the TryFrom conversion from RawBlock to Block) that applies to both kinds of serialization.

This changes the Block type to hold the transactions as a plain `Vec<Vec<u8>>`, rather than as a custom `abci::transaction::Data` type (which is itself a wrapper for an `Option<Vec<Transaction>>`, the `Transaction` type being a wrapper for a `Vec<u8>`). This also updates the `Block` getter functions; it's not clear (to me) why they're there, since they access the public fields and aren't used anywhere else.

The existing contents of the `tendermint::abci` module note that they're only for the purpose of implementing the RPC endpoints, not a general ABCI implementation. Moving that code from the `tendermint` crate to the `tendermint-rpc` crate decouples the RPC interface from improvements to the ABCI domain modeling. Eventually, it would be good to eliminate this code and align it with the new ABCI domain types.

These types mirror the generated types in tendermint_proto, but have better naming. The documentation for each request type is adapted from the ABCI Methods and Types spec document. However, the same logical request may appear three times, as a struct with the request data, as a Request variant, and as a CategoryRequest variant. To avoid duplication, this PR uses the `#[doc = include_str!(...)]` functionality stabilized in Rust 1.54 to keep common definitions of the documentation.

The code in the `abci` module had more complete documentation from the ABCI docs, so I copied it onto the existing structure.

hdevalence · 2021-11-16T00:23:26Z

@thanethomson I've rebased this again and deduplicated types per your suggestions (thanks for pointing those out!). I also aligned all the errors, and then rebased again to eliminate the placeholder commits labeled "drop me".

So, this should be good to go.

thanethomson · 2021-11-27T19:00:07Z

Superseded by (and included in) #1022.

hdevalence · 2021-11-29T17:27:34Z

🎊 Thanks so much for getting this over the finish line!!

Now that informalsystems/tendermint-rs#862 is merged, we can stop using our own branch.

hdevalence force-pushed the abci-domain-types branch 6 times, most recently from 29bf6b3 to 1175d4f Compare April 15, 2021 22:24

hdevalence mentioned this pull request Apr 19, 2021

What's the purpose of tendermint::Time? #865

Open

cwgoes mentioned this pull request May 24, 2021

WIP: Implement ABCI++ #833

Closed

5 tasks

tzemanovic mentioned this pull request May 26, 2021

integrate tower-rs for ABCI anoma/anoma#240

Closed

hdevalence mentioned this pull request Jun 2, 2021

WIP: Implement ABCI++ penumbra-zone/tendermint-rs#1

Closed

5 tasks

thanethomson reviewed Jun 7, 2021

View reviewed changes

xla changed the title ~~[WIP] Add ABCI domain types to tendermint::abci.~~ [WIP] abci: Add domain types Jun 8, 2021

thanethomson mentioned this pull request Jun 23, 2021

prost 0.7 uses core types instead of std #871

Closed

hdevalence force-pushed the abci-domain-types branch from 1175d4f to e6d14ac Compare July 28, 2021 23:08

hdevalence force-pushed the abci-domain-types branch 2 times, most recently from 36069f3 to 2a78e5f Compare August 5, 2021 05:54

hdevalence mentioned this pull request Sep 5, 2021

Finish ABCI domain types work and use it for Penumbra penumbra-zone/penumbra#11

Closed

hdevalence force-pushed the abci-domain-types branch 3 times, most recently from 26f9ef5 to dd59660 Compare September 9, 2021 23:40

hdevalence force-pushed the abci-domain-types branch 2 times, most recently from 726beaa to 45c3946 Compare September 10, 2021 01:05

hdevalence added a commit to penumbra-zone/penumbra that referenced this pull request Sep 10, 2021

docs: render tendermint docs

c109627

This is useful until informalsystems/tendermint-rs#862 is merged upstream.

hdevalence and others added 18 commits November 15, 2021 16:20

Regenerate protobuf types using tools/proto-compiler.

c43a78e

Regenerate protos against tendermint v0.35.0-rc3.

57dd21b

Remove SetOption-related code in tendermint-abci.

d6c8395

Update imports to reflect protobuf movements.

39a9bd4

The BlockParams and ConsensusParams structs moved out of the ABCI protos.

Update tendermint-abci to reflect upstream proto changes.

6e48efa

p2p: treat all non-Ed25519 keys as unsupported

ee4c84a

This fixes a compile error introduced by upstream proto changes that add an Sr25519 variant.

Improve ABCI response code modeling with NonZeroU32

d9ed69a

The previous data modeling allowed a user to construct an `Err(0)` value that would be serialized and deserialized as `Ok`.

Use the Bytes type in ABCI protos.

4f47712

Add conversions between protobuf and chrono types.

156cce9

tendermint: eliminate &'static str errors in ABCI domain types.

0dcdc6f

Merge abci::params::ConsensusParams with consensus::Params.

4dbbbd0

The code in the `abci` module had more complete documentation from the ABCI docs, so I copied it onto the existing structure.

Add hex encoding Serde attribute to Sr25519 keys.

3c5eff5

Replace integers with block::Height, vote::Power

1f1a4c8

Replace integer with block::Round

9660feb

hdevalence force-pushed the abci-domain-types branch from fde6080 to 9660feb Compare November 16, 2021 00:21

This was referenced Nov 16, 2021

Update git dependency to abci-domain-types_r14 penumbra-zone/tower-abci#10

Closed

Update ABCI to abci-domain-types_r14 penumbra-zone/penumbra#152

Closed

thanethomson mentioned this pull request Nov 19, 2021

Add domain types (updated) and fix integration tests for Tendermint v0.35.0 #1022

Merged

5 tasks

thanethomson closed this Nov 27, 2021

hdevalence added a commit to penumbra-zone/tower-abci that referenced this pull request Dec 8, 2021

Update tendermint-rs now that the ABCI changes are merged.

6d47a4b

Now that informalsystems/tendermint-rs#862 is merged, we can stop using our own branch.

hdevalence mentioned this pull request Dec 8, 2021

Update tendermint-rs now that the ABCI changes are merged. penumbra-zone/tower-abci#13

Merged

hdevalence added a commit to penumbra-zone/tower-abci that referenced this pull request Dec 8, 2021

Update tendermint-rs now that the ABCI changes are merged. (#13)

8465938

Now that informalsystems/tendermint-rs#862 is merged, we can stop using our own branch.

hdevalence mentioned this pull request Dec 8, 2021

Switch to upstream tendermint-rs penumbra-zone/penumbra#239

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

abci: Add domain types #862

abci: Add domain types #862

hdevalence commented Apr 14, 2021 •

edited

Loading

hdevalence commented Apr 14, 2021 •

edited

Loading

thanethomson left a comment

cwgoes commented Jun 22, 2021

thanethomson commented Jun 23, 2021

hdevalence commented Jul 28, 2021

hdevalence commented Aug 5, 2021 •

edited

Loading

hdevalence commented Sep 9, 2021

thanethomson commented Sep 10, 2021

hdevalence commented Nov 16, 2021

thanethomson commented Nov 27, 2021

hdevalence commented Nov 29, 2021

abci: Add domain types #862

abci: Add domain types #862

Conversation

hdevalence commented Apr 14, 2021 • edited Loading

hdevalence commented Apr 14, 2021 • edited Loading

thanethomson left a comment

Choose a reason for hiding this comment

cwgoes commented Jun 22, 2021

thanethomson commented Jun 23, 2021

hdevalence commented Jul 28, 2021

hdevalence commented Aug 5, 2021 • edited Loading

hdevalence commented Sep 9, 2021

thanethomson commented Sep 10, 2021

hdevalence commented Nov 16, 2021

thanethomson commented Nov 27, 2021

hdevalence commented Nov 29, 2021

hdevalence commented Apr 14, 2021 •

edited

Loading

hdevalence commented Apr 14, 2021 •

edited

Loading

hdevalence commented Aug 5, 2021 •

edited

Loading