-
Notifications
You must be signed in to change notification settings - Fork 235
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
fix: "Types in a binary operation should match, but found T and T" #4648
Conversation
I just found a simpler scenario that also triggers this bug, in case you want to test against it or add it as a regression: struct Wrapper {
inner: Field,
}
trait FromField {
fn from_field(value: Field) -> Self;
}
trait ToField {
fn to_field(self) -> Field;
}
impl ToField for Wrapper {
fn to_field(self) -> Field {
self.inner
}
}
impl FromField for Wrapper {
fn from_field(value: Field) -> Wrapper {
Wrapper { inner: value }
}
}
#[test]
fn test() {
let wapper = Wrapper { inner: 37 };
assert_eq(FromField::from_field(wapper.to_field()), wapper);
} This errors out with |
@nventuro I think the original example is sufficient. I'll mention though that the new error message with this PR for your snippet is:
|
Ah yes, I didn't bother adding that impl for the minimum reproducible example 😅 |
…#4662) # Description ## Problem\* Resolves #4653 ## Summary\* If the object type is an unbound type variable, impl search currently just chooses the first available matching impl instead of erroring that type annotations are needed or similar. This can lead to confusing situations where the type chosen is chosen just because it was the first impl in the stdlib to be defined. This PR prevents that. ## Additional Context This PR depends on #4648 and should be merged after. ~~The `hashmap` test is currently failing because the `H: Hasher` constraint on its `Eq` implementation ~~is actually unsolvable since `H` isn't mentioned in the hashmap type at all~~. It is solvable since it is meantioned in `B`, although solving it so far has lead to errors during monomorphization. Previously it was fine since H was unbound and the first/only impl was just always chosen. Now I'll need to change or remove H to fix it.~~ This PR is now ready to review. I've been debugging a bug with monomorphizing some HashMap code for a little while but it turns out the bug is also present in master. So I'm considering it a separate issue: #4663 ## Documentation\* Check one: - [x] No documentation needed. - [ ] Documentation included in this PR. - [ ] **[For Experimental Features]** Documentation to be submitted in a separate PR. # PR Checklist\* - [x] I have tested the changes locally. - [x] I have formatted the changes with [Prettier](https://prettier.io/) and/or `cargo fmt` on default settings.
…4648) # Description ## Problem\* Resolves #4635 Resolves #4502 ## Summary\* This was more difficult to fix than it originally seemed. The main issue was between interactions with unbound type variables, type aliases, type rules for operators, and operator traits. Removing the "infer unbound type variables to be numeric" rule on operators causes a lot of stdlib code to break where it'd be unreasonable to have type annotations. This caused unbound type variables to be bound to the first object type whose impl it was checked against when calling verify trait impl. I eventually settled on just delaying the verify trait impl check for operators until the end of a function when more types are known. ## Additional Context ## Documentation\* Check one: - [x] No documentation needed. - [ ] Documentation included in this PR. - [ ] **[For Experimental Features]** Documentation to be submitted in a separate PR. # PR Checklist\* - [x] I have tested the changes locally. - [x] I have formatted the changes with [Prettier](https://prettier.io/) and/or `cargo fmt` on default settings.
…#4662) # Description ## Problem\* Resolves #4653 ## Summary\* If the object type is an unbound type variable, impl search currently just chooses the first available matching impl instead of erroring that type annotations are needed or similar. This can lead to confusing situations where the type chosen is chosen just because it was the first impl in the stdlib to be defined. This PR prevents that. ## Additional Context This PR depends on #4648 and should be merged after. ~~The `hashmap` test is currently failing because the `H: Hasher` constraint on its `Eq` implementation ~~is actually unsolvable since `H` isn't mentioned in the hashmap type at all~~. It is solvable since it is meantioned in `B`, although solving it so far has lead to errors during monomorphization. Previously it was fine since H was unbound and the first/only impl was just always chosen. Now I'll need to change or remove H to fix it.~~ This PR is now ready to review. I've been debugging a bug with monomorphizing some HashMap code for a little while but it turns out the bug is also present in master. So I'm considering it a separate issue: #4663 ## Documentation\* Check one: - [x] No documentation needed. - [ ] Documentation included in this PR. - [ ] **[For Experimental Features]** Documentation to be submitted in a separate PR. # PR Checklist\* - [x] I have tested the changes locally. - [x] I have formatted the changes with [Prettier](https://prettier.io/) and/or `cargo fmt` on default settings.
🤖 I have created a release *beep* *boop* --- <details><summary>0.27.0</summary> ## [0.27.0](v0.26.0...v0.27.0) (2024-04-10) ### ⚠ BREAKING CHANGES * Brillig typed memory (AztecProtocol/aztec-packages#5395) ### Features * **acir_gen:** Fold attribute at compile-time and initial non inlined ACIR (AztecProtocol/aztec-packages#5341) ([a0f7474](a0f7474)) * **acvm_js:** Execute program ([#4694](#4694)) ([386f6d0](386f6d0)) * **acvm:** Execute multiple circuits (AztecProtocol/aztec-packages#5380) ([a0f7474](a0f7474)) * Add `remove_enable_side_effects` SSA pass ([#4224](#4224)) ([94952db](94952db)) * Allow slices to brillig entry points ([#4713](#4713)) ([62423d5](62423d5)) * Brillig typed memory (AztecProtocol/aztec-packages#5395) ([0bc18c4](0bc18c4)) * **docs:** Documenting noir codegen ([#4454](#4454)) ([24f6d85](24f6d85)) * Improve nargo check cli with --override flag and feedback for existing files ([#4575](#4575)) ([5e7fbd4](5e7fbd4)) * Improve optimisations on range constraints ([#4690](#4690)) ([96b8110](96b8110)) * Improve SSA type-awareness in EQ and MUL instructions ([#4691](#4691)) ([669f1a0](669f1a0)) * **nargo:** Multiple circuits info for binary programs ([#4719](#4719)) ([50d2735](50d2735)) ### Bug Fixes * "Types in a binary operation should match, but found T and T" ([#4648](#4648)) ([30c9f31](30c9f31)) * **acvm:** Mark outputs of Opcode::Call solvable ([#4708](#4708)) ([8fea405](8fea405)) * Correct ICE panic messages in brillig `convert_black_box_call` ([#4761](#4761)) ([f3eee6c](f3eee6c)) * Error when a type variable is unbound during monomorphization instead of defaulting to Field ([#4674](#4674)) ([03cdba4](03cdba4)) * Field comparisons ([#4704](#4704)) ([079cb2a](079cb2a)) * Impl search no longer selects an impl if multiple are applicable ([#4662](#4662)) ([0150600](0150600)) * Last use analysis & make it an SSA pass ([#4686](#4686)) ([0d3d5fd](0d3d5fd)) * Slice coercions ([#4640](#4640)) ([c0bae17](c0bae17)) * **ssa:** Accurate constant type for slice dummy data in flattening ([#4661](#4661)) ([b87654e](b87654e)) * **ssa:** Do not use get_value_max_num_bits when we want pure type information ([#4700](#4700)) ([b55a580](b55a580)) * **ssa:** Fix slice intrinsic handling in the capacity tracker ([#4643](#4643)) ([1b50ce1](1b50ce1)) * Unknown slice lengths coming from as_slice ([#4725](#4725)) ([f21129e](f21129e)) * Update commit for noir-gates-diff ([#4773](#4773)) ([a9766c5](a9766c5)) </details> <details><summary>0.43.0</summary> ## [0.43.0](v0.42.0...v0.43.0) (2024-04-10) ### ⚠ BREAKING CHANGES * Brillig typed memory (AztecProtocol/aztec-packages#5395) * **acir:** Program and witness stack structure (AztecProtocol/aztec-packages#5149) * automatic NoteInterface and NoteGetterOptions auto select (AztecProtocol/aztec-packages#4508) * Acir call opcode (AztecProtocol/aztec-packages#4773) * Support contracts with no constructor (AztecProtocol/aztec-packages#5175) * Internal as a macro (AztecProtocol/aztec-packages#4898) * move noir out of yarn-project (AztecProtocol/aztec-packages#4479) * note type ids (AztecProtocol/aztec-packages#4500) * rename bigint_neg into bigint_sub (AztecProtocol/aztec-packages#4420) * Add expression width into acir (AztecProtocol/aztec-packages#4014) * init storage macro (AztecProtocol/aztec-packages#4200) * **acir:** Move `is_recursive` flag to be part of the circuit definition (AztecProtocol/aztec-packages#4221) * Sync commits from `aztec-packages` ([#4144](#4144)) * Breaking changes from aztec-packages ([#3955](#3955)) * Rename Arithmetic opcode to AssertZero ([#3840](#3840)) * Remove unused methods on ACIR opcodes ([#3841](#3841)) ### Features * Acir call opcode (AztecProtocol/aztec-packages#4773) ([c3c9e19](c3c9e19)) * **acir_gen:** Fold attribute at compile-time and initial non inlined ACIR (AztecProtocol/aztec-packages#5341) ([a0f7474](a0f7474)) * **acir:** Program and witness stack structure (AztecProtocol/aztec-packages#5149) ([13eb71b](13eb71b)) * **acvm_js:** Execute program ([#4694](#4694)) ([386f6d0](386f6d0)) * **acvm:** Execute multiple circuits (AztecProtocol/aztec-packages#5380) ([a0f7474](a0f7474)) * Add bit size to const opcode (AztecProtocol/aztec-packages#4385) ([158c8ce](158c8ce)) * Add CMOV instruction to brillig and brillig gen (AztecProtocol/aztec-packages#5308) ([13eb71b](13eb71b)) * Add expression width into acir (AztecProtocol/aztec-packages#4014) ([158c8ce](158c8ce)) * Add instrumentation for tracking variables in debugging ([#4122](#4122)) ([c58d691](c58d691)) * Add poseidon2 opcode implementation for acvm/brillig, and Noir ([#4398](#4398)) ([10e8292](10e8292)) * Add support for overriding expression width ([#4117](#4117)) ([c8026d5](c8026d5)) * Added cast opcode and cast calldata (AztecProtocol/aztec-packages#4423) ([78ef013](78ef013)) * Allow brillig to read arrays directly from memory (AztecProtocol/aztec-packages#4460) ([158c8ce](158c8ce)) * Allow nested arrays and vectors in Brillig foreign calls (AztecProtocol/aztec-packages#4478) ([158c8ce](158c8ce)) * Allow variables and stack trace inspection in the debugger ([#4184](#4184)) ([bf263fc](bf263fc)) * Automatic NoteInterface and NoteGetterOptions auto select (AztecProtocol/aztec-packages#4508) ([13eb71b](13eb71b)) * **avm:** Back in avm context with macro - refactor context (AztecProtocol/aztec-packages#4438) ([158c8ce](158c8ce)) * **avm:** Brillig CONST of size > u128 (AztecProtocol/aztec-packages#5217) ([c3c9e19](c3c9e19)) * **aztec-nr:** Initial work for aztec public vm macro (AztecProtocol/aztec-packages#4400) ([158c8ce](158c8ce)) * Backpropagate constants in ACIR during optimization ([#3926](#3926)) ([aad0da0](aad0da0)) * Breaking changes from aztec-packages ([#3955](#3955)) ([5be049e](5be049e)) * Brillig IR refactor (AztecProtocol/aztec-packages#5233) ([c3c9e19](c3c9e19)) * Brillig typed memory (AztecProtocol/aztec-packages#5395) ([0bc18c4](0bc18c4)) * Check initializer msg.sender matches deployer from address preimage (AztecProtocol/aztec-packages#5222) ([c3c9e19](c3c9e19)) * Evaluation of dynamic assert messages ([#4101](#4101)) ([c284e01](c284e01)) * Init storage macro (AztecProtocol/aztec-packages#4200) ([158c8ce](158c8ce)) * Initial Earthly CI (AztecProtocol/aztec-packages#5069) ([c3c9e19](c3c9e19)) * Internal as a macro (AztecProtocol/aztec-packages#4898) ([5f57ebb](5f57ebb)) * New brillig field operations and refactor of binary operations (AztecProtocol/aztec-packages#5208) ([c3c9e19](c3c9e19)) * Note type ids (AztecProtocol/aztec-packages#4500) ([78ef013](78ef013)) * Remove range constraints from witnesses which are constrained to be constants ([#3928](#3928)) ([afe9c7a](afe9c7a)) * Remove replacement of boolean range opcodes with `AssertZero` opcodes ([#4107](#4107)) ([dac0e87](dac0e87)) * Signed integer division and modulus in brillig gen (AztecProtocol/aztec-packages#5279) ([c3c9e19](c3c9e19)) * Support contracts with no constructor (AztecProtocol/aztec-packages#5175) ([c3c9e19](c3c9e19)) * Sync `aztec-packages` ([#4011](#4011)) ([fee2452](fee2452)) * Sync commits from `aztec-packages` ([#4068](#4068)) ([7a8f3a3](7a8f3a3)) * Sync commits from `aztec-packages` ([#4144](#4144)) ([0205d3b](0205d3b)) * Sync from aztec-packages ([#4483](#4483)) ([fe8f277](fe8f277)) * Sync from noir (AztecProtocol/aztec-packages#5234) ([c3c9e19](c3c9e19)) * Sync from noir (AztecProtocol/aztec-packages#5286) ([c3c9e19](c3c9e19)) ### Bug Fixes * **acvm:** Mark outputs of Opcode::Call solvable ([#4708](#4708)) ([8fea405](8fea405)) * Noir test incorrect reporting (AztecProtocol/aztec-packages#4925) ([5f57ebb](5f57ebb)) * Remove panic from `init_log_level` in `acvm_js` ([#4195](#4195)) ([2e26530](2e26530)) * Return error rather instead of panicking on invalid circuit ([#3976](#3976)) ([67201bf](67201bf)) ### Miscellaneous Chores * **acir:** Move `is_recursive` flag to be part of the circuit definition (AztecProtocol/aztec-packages#4221) ([158c8ce](158c8ce)) * Move noir out of yarn-project (AztecProtocol/aztec-packages#4479) ([78ef013](78ef013)) * Remove unused methods on ACIR opcodes ([#3841](#3841)) ([9e5d0e8](9e5d0e8)) * Rename Arithmetic opcode to AssertZero ([#3840](#3840)) ([836f171](836f171)) * Rename bigint_neg into bigint_sub (AztecProtocol/aztec-packages#4420) ([158c8ce](158c8ce)) </details> --- This PR was generated with [Release Please](https://github.com/googleapis/release-please). See [documentation](https://github.com/googleapis/release-please#release-please).
Description
Problem*
Resolves #4635
Resolves #4502
Summary*
This was more difficult to fix than it originally seemed. The main issue was between interactions with unbound type variables, type aliases, type rules for operators, and operator traits.
Removing the "infer unbound type variables to be numeric" rule on operators causes a lot of stdlib code to break where it'd be unreasonable to have type annotations. This caused unbound type variables to be bound to the first object type whose impl it was checked against when calling verify trait impl.
I eventually settled on just delaying the verify trait impl check for operators until the end of a function when more types are known.
Additional Context
Documentation*
Check one:
PR Checklist*
cargo fmt
on default settings.