Skip to content

Feat/abi structs - #1606

Merged
sbillig merged 20 commits into
masterfrom
feat/abi-structs
Oct 1, 2026
Merged

sbillig merged 20 commits into
masterfrom
feat/abi-structs

Conversation

@cburgdorf

@cburgdorf cburgdorf commented Sep 25, 2026 •

Copy link
Copy Markdown
Collaborator

Added the #[abi] struct attribute. An #[abi] struct gets AbiSize, AbiSpan<Sol>, Encode<Sol> and Decode<Sol> implementations that encode it like the Solidity tuple of its fields, so Solidity structs such as PoolKey can be used in msg fields and return types, in arrays and inside other #[abi] structs without hand-written codecs.

DynArray::get now works for every ABI element type, including multi-word static structs and tuples and dynamic elements such as Bytes. MemVec accepts any static ABI element type. The new std::abi::dyn_array_from([..]) builds a DynArray from a fixed array of copyable ABI elements (for example a bytes[]), and std::abi::canonical_dyn_array re-encodes a decoded array in canonical layout.

@cburgdorf

Copy link
Copy Markdown
Collaborator Author

@codex review

@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Sep 28, 2026 •

Copy link
Copy Markdown

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-10-01T03:54:10.663751Z e63cf77 New commits
🔒 Security Review ✅ Completed 2026-10-01T04:15:58.171500Z e63cf77 New commits
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@chatgpt-codex-connector

Copy link
Copy Markdown

🛡️ Codex Security Review · Automatically triggered

Security review completed. No security issues were found in this pull request.

Reviewed commit: df6b13fe2b

View security finding report

Only the user who started this review can view the report in Codex.

ℹ️ About Codex security reviews in GitHub

This is an experimental Codex feature. Security reviews are triggered when:

  • You comment "@codex security review"
  • A regular code review gets triggered (for example, "@codex review" or when a PR is opened), and you’re opted in so security review runs alongside code review

Once complete, Codex will leave suggestions, or a comment if no findings are found.

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: df6b13fe2b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread crates/hir/src/core/lower/item.rs Resolved
Comment thread ingots/std/src/abi/sol/types.fe Resolved
Comment thread ingots/std/src/abi/dyn_array.fe Resolved
Comment thread ingots/std/src/abi/dyn_array.fe Resolved

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ecd65c7f7e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread ingots/std/src/abi/sol/types.fe Resolved

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0e33603aa0

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread ingots/std/src/abi/sol/types.fe Resolved
Comment thread crates/hir/src/core/lower/abi_struct.rs Resolved

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 83ec959602

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread crates/hir/src/core/lower/item.rs Resolved
Comment thread crates/hir/src/core/lower/abi_struct.rs Resolved
Comment thread crates/hir/src/core/lower/abi_struct.rs Resolved

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c1e7bd0cc0

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread crates/hir/src/analysis/ty/abi_ty.rs Resolved
Comment thread ingots/std/src/abi/dyn_array.fe Resolved
Comment thread ingots/std/src/abi/sol/types.fe Resolved
@cburgdorf
cburgdorf force-pushed the feat/abi-structs branch 2 times, most recently from c1e7bd0 to 31b3930 Compare September 29, 2026 23:19
@sbillig

sbillig commented Sep 30, 2026

Copy link
Copy Markdown
Collaborator

@cburgdorf I'm going to push a few fixes to this branch

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0cd45445b1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread ingots/std/src/abi/sol/types.fe Resolved
cburgdorf and others added 9 commits September 30, 2026 20:26
An `#[abi]` struct is lowered like a `msg` variant: it gets `AbiSize`,
`AbiSpan<Sol>`, `Encode<Sol>` and `Decode<Sol>` impls that encode it as
the Solidity tuple of its fields, plus `impl StaticAbi for S where
<fields>: StaticAbi`. Generic `#[abi]` structs and `#[abi]` combined
with `#[event]`/`#[error]` are rejected.

`DynArray::get` decodes any element type: multi-word static elements are
read at their head size and dynamic ones through their validated tail
offset. `MemVec` stores any `StaticAbi` type at its head size; the new
public `std::abi::StaticAbi` marker covers one-word values, tuples,
fixed arrays and `#[abi]` structs of static types, so a `MemVec` of a
dynamic type still fails to compile.

`std::abi::dyn_array_from([..])` builds a `DynArray` of any element
type from a fixed array, and `canonical_dyn_array` re-encodes a decoded
array in canonical layout.
An #[abi] struct can now be an #[event] or #[error] field. The lowering
generates impl SolCompat with the Solidity tuple type of the fields,
built from std::abi::SolPunctuation literals and each field's
<F as SolCompat>::S / SOL_TYPE, nested in chunks like event signatures.
Previously such an event failed at MIR lowering with an internal CTFE
error for TOPIC0.

Generated impl consts must reify to a runtime value; SOL_TYPE is only
consumed by CTFE for signature hashes, so the generated one is checked
like hand-written impls instead.
A SolCompat value from the generic DynArray impl carries its type as
(Item::S, String<2>) with the projection unnormalized, while the
aggregate of an #[abi] struct containing DynArray<Item> names the
normalized type. The shape check already skips expected types with
projections; skip actual ones the same way.
Only the struct validator knew about the new `abi` attribute, so
`#[abi]` on functions, enums, traits, modules and other items was
silently accepted. Add an unsupported-target rule next to the `event`
and `error` ones.
The StaticAbi markers stopped at eight-element tuples while core provides
AbiSize and Encode through arity 16, so e.g. a MemVec of a fully static
nine-element tuple was rejected. Cover one-element tuples and arities 9
to 16 as well.
…elds

The sizing loop in canonical_dyn_array decoded every element only to
measure it, and the encoding loop then decoded it again. Static elements
now skip the sizing pass, as their size is fixed. Dynamic elements are
decoded once and encoded into a scratch tail while sizing; the tails are
then copied into place. Keeping the decoded values instead would need
`T: Copy`, since the ownership checker cannot see that a loop moves each
slot out only once, and `canonical_dyn_array` is meant to accept
non-`Copy` elements too.

For eight 32-byte `bytes` elements this lowers the gas of the call from
9686 to 8333, and from 1990 to 1683 for eight `uint256` elements.

Zero-length fixed arrays are `StaticAbi`, so their decoders now return
early instead of decoding a first element.

The generated `Decode<Sol>` for `#[abi]` structs only defined
`decode_payload`, so `decode_from_bounded` and
`decode_from_prechecked_head` fell back to `decode_from`, which checks
dynamic fields against the whole input rather than `input_len`. A
dynamic field could thus read a tail outside the declared ABI region.
The struct impls now pass `input_len` to every field, like the tuple and
fixed-array impls.

`StaticAbi` cannot be sealed, because `#[abi]` structs in any ingot
implement it. A downstream impl for a dynamic type would make `MemVec`
under-allocate and encode tails over neighbouring memory, so the
`MemVec` constructors now revert with `Panic(0x01)` when
`AbiSize::IS_DYNAMIC` is set. The check is constant and folds away for
static types.

`#![abi]` inside a module body is rejected like `#[abi] mod m`.

An indexed fixed array, tuple or `#[abi]` struct event field needs
Solidity's hashed topic of its encoding, which `#[event]` does not
generate, so it failed with a type error in the generated emit body.
Such fields, including ones hidden behind type aliases, are now reported
at the field (10-0009), like indexed dynamic fields.

Tuples implement `SolCompat` for arities 0 through 16, so `#[abi]`
structs with tuple fields no longer fail their generated `SolCompat`
bound.

`canonical_dyn_array` returns right after the length word for zero-size
elements such as `()` or an empty `#[abi]` struct, instead of looping
over a length that costs the sender nothing to inflate.

The fixed-array ABI codecs need `Copy` elements, as an array element
cannot be moved out by index. `StaticAbi` for `[T; N]` now requires
`T: Copy` too, and an `#[abi]`, `#[event]` or `#[error]` field holding a
fixed array of a non-`Copy` type is reported once at the field (16-0008)
instead of as trait errors in the generated code.
Semantic bodies of generic instances are substituted but not normalized,
so a local of `impl<T> Tr2 for W<T>` can be typed `(<u8 as Tr>::A, u16)`.
CTFE labelled aggregate values with that raw local type, and a concrete
aggregate embedding one failed the child-type check with "constant child
type differs from its aggregate field type". This was reachable on master
without `#[abi]`.

`eval_expr` now normalizes the result type against the frame's instance
before building a value, and the verifier's actual-side projection skip
is removed, since it disabled the check for whole subtrees.
Tuples implemented `AbiSpan` only up to arity 4 and `Decode` up to 12,
while `AbiSize`, `Encode`, `SolCompat` and `StaticAbi` go up to 16. The
generated `#[abi]` codec needs `AbiSpan` for every field, so a struct
with a five-element tuple field failed to compile, and a `DynArray` of
such tuples could not be decoded.
`msg`, `#[abi]` and `#[event]` lowering declared their generated type
parameters as plain `D`, `I` and `L`. A field type spelled the same, a
struct or an alias, then resolved to the parameter inside the generated
bodies: an `#[abi]` struct with a field of type `D` failed trait
resolution, and a `msg` field aliased to `D` produced internal borrow
checking errors. The parameters now use `__fe_` generated names, like the
generated value bindings.
The composite-elements fragment claimed `dyn_array_from` takes a fixed
array of any element type; it requires `Copy` elements.
`payload_end` now delegates to `payload_end_with_input_len` with
`input.len()` instead of repeating the whole field fold, which also drops
the duplicate diagnostics of the second body. Generic argument lists are
built with a new `GenericArgListId::given_types`, positional calls use
`call_expr`, `u256` is built once, the two `Decode` bodies are generated
explicitly instead of through a flag loop, and the `StaticAbi` and
`SolCompat` impls share one field-bound helper.

A desugar snapshot now covers the generated impls, including fields
named like the generated parameters, and round-trips them through the
analyzer.
`#[abi]` diagnostics were `#[error]` diagnostics (`ErrorLower`), and the
non-`Copy` fixed-array field error of `#[event]` fields was reported under
`#[error]` as well.

`#[abi]` structs now have an `AbiStructLower` pass (18): generic structs
are 18-0001, conflicting attributes 18-0002 and non-`Copy` fixed-array
fields 18-0003. The fixed-array field error of an `#[event]` or `#[error]`
struct is reported by that struct's pass, as 10-0010 or 13-0006. The field
scan behind it is now a tracked query shared by the three passes and the
body pass.
`abi_struct_event_emits` emitted two events and asserted nothing. The
emits move to a test-runner fixture whose snapshot records the logs;
their data matches `cast abi-encode` of the same Solidity values. The
TOPIC0 comments of `Pair` and `OrdersEvent` are back next to their
assertions.
`ErrorLower` and `SemanticBorrowck` once shared pass code 16, so one
printed code could name two unrelated errors. A unit test now fails when
two passes share a code.
`usize` and `isize` have ABI codecs and map to `uint256` and `int256` in
the JSON ABI, but had no `SolCompat`, so an `#[abi]` struct with such a
field failed its generated `SolCompat` impl, and they could not be event
or error fields. They also lacked `TopicValue` for indexed event fields.
`reify_runtime_const_impl` matched a value's shape against the expected
type as given. A trait constant's expected type can be an unnormalized
projection, such as `<Key as SolCompat>::S` for `SolCompat::SOL_TYPE`,
which is no tuple, so the generated `SOL_TYPE` of an `#[abi]` struct could
not be reified. `check_impl_trait_const_bodies` worked around this by
exempting any trait named `SolCompat` from the required-value check.

The expected type is now normalized against the instance first, and the
name-based exemption is removed.
A field whose type lacks an ABI trait made every generated impl report
its own failure at the whole struct: ten errors for one field. The
`AbiStructLower` pass now checks each field for `AbiSize`,
`AbiSpan<Sol>`, `Encode<Sol>` and `Decode<Sol>` and reports the missing
traits once at the field (18-0004). The generated impls of such a struct,
or of a recursive one, are no longer diagnosed.

A field without `SolCompat` no longer rejects the struct: it keeps its
codec and just does not implement `SolCompat`, whose generated impl is
not diagnosed. Where a Solidity type name is needed, an `#[event]` or
`#[error]` field of an ABI-encodable type without `SolCompat` is now
reported at the field (10-0011, 13-0007). Before, such an event or error
passed checking and failed during MIR lowering with a CTFE error for its
TOPIC0 or selector.
Generated names were `__fe_`-prefixed but still valid source identifiers,
so a type spelled `__fe_D` or `__fe_I` in a field was captured by the
generated type parameter, and a constant spelled like a generated local
by that local. `HirBuilder` now collects the identifiers of the `msg`,
`#[event]`, `#[error]` or `#[abi]` item it desugars, and generated names
take a numeric suffix when the source mentions them. The encode locals
`__tail`, `__head_pos`, `__field_N` and the `#[abi]` span fold's `__end`
go through the same helper.
Where clauses now also hold const predicates, so the generated
`StaticAbi` and `SolCompat` field bounds are `WhereClausePredicate::Type`.
@sbillig
sbillig merged commit ee2199f into master Oct 1, 2026
9 checks passed
@sbillig
sbillig deleted the feat/abi-structs branch October 1, 2026 17:31
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants