2/2: Add enumerate() to DSLX stdlib

[joannabrozek]

The second of two PRs responsible for migrating from the built-in enumerate, which cannot be lowered to IR.

This PR includes: removing the build-in enumerate.

[mikex-oss]

Aside from obvious updates to use std:enumerate, I'm seeing the following error in a bunch of our targets.

Not sure yet how to repro this in a small standalone test.

xls/dslx/stdlib/std.x:20:13-20:19
0018: 
0019: pub fn enumerate<T: type, N: u32>(x: T[N]) -> (u32, T)[N] {
0020:     for (i, result) in 0..N {
~~~~~~~~~~~~~~~~~~^----^ TypeInferenceError: TypeInferenceError: xls/dslx/stdlib/std.x:20:13-20:19 Annotated array size is too small for explicit element count.. The body of function `enumerate` does not actually return the function's declared return type, which is `(u32, u32)[N]`
0021:         update(result, i, (i, x[i]))
0022:     }([(u32:0, zero!<T>()), ...])
Error: INVALID_ARGUMENT: TypeInferenceError: xls/dslx/stdlib/std.x:20:13-20:19 TypeInferenceError: xls/dslx/stdlib/std.x:20:13-20:19 Annotated array size is too small for explicit element count.. The body of function `enumerate` does not actually return the function's declared return type, which is `(u32, u32)[N]`
=== Source Location Trace: === 
xls/dslx/errors.cc:48
xls/dslx/type_system_v2/type_annotation_resolver.cc:325
xls/dslx/type_system_v2/inference_table_converter_impl.cc:1077
xls/dslx/type_system_v2/inference_table_converter_impl.cc:269
xls/dslx/type_system_v2/inference_table_converter_impl.cc:884
xls/dslx/type_system_v2/inference_table_converter_impl.cc:245
xls/dslx/type_system_v2/inference_table_converter_impl.cc:884
xls/dslx/type_system_v2/inference_table_converter_impl.cc:245
xls/dslx/type_system_v2/inference_table_converter_impl.cc:2007
xls/dslx/type_system_v2/inference_table_converter_impl.cc:1750
xls/dslx/type_system_v2/inference_table_converter_impl.cc:1835
xls/dslx/type_system_v2/inference_table_converter_impl.cc:737
xls/dslx/type_system_v2/inference_table_converter_impl.cc:245
xls/dslx/import_routines.cc:247
xls/dslx/type_system_v2/populate_table_visitor.cc:136
xls/dslx/type_system_v2/populate_table_visitor.cc:1939
xls/dslx/type_system_v2/typecheck_module_v2.cc:89
xls/dslx/import_routines.cc:247
xls/dslx/type_system_v2/populate_table_visitor.cc:136
xls/dslx/type_system_v2/populate_table_visitor.cc:1939
xls/dslx/type_system_v2/typecheck_module_v2.cc:89
xls/dslx/import_routines.cc:247
xls/dslx/type_system_v2/populate_table_visitor.cc:136
xls/dslx/type_system_v2/populate_table_visitor.cc:1939
xls/dslx/type_system_v2/typecheck_module_v2.cc:89
xls/dslx/parse_and_typecheck.cc:126
xls/dslx/ir_convert/ir_converter.cc:697
xls/dslx/ir_convert/ir_converter_main.cc:159

[erinzmoore]

I believe I was able to reproduce the issue by testing with an empty array, for example:

#[test]
fn emumerate_test() {
    ...

    let empty_array : u32[0] = [];
    let enumerated_empty = enumerate(empty_array);
    assert_eq(enumerated_empty, []);
}

I tried to see if I could fix this with a const if in the enumerate implementation and it also failed, which may be a separate issue:

pub fn enumerate<T: type, N: u32>(x: T[N]) -> (u32, T)[N] {
    const if N == 0 {
      (u32, T)[0]:[]
    } else {
      for (i, result) in 0..N {
          update(result, i, (i, x[i]))
      }((u32, T)[N]:[(u32:0, zero!<T>()), ...])
    }
}

[joannabrozek]

@erinzmoore, thank you for the test to reproduce the issue.

I've added more commits to this PR to provide support for empty array enumeration:

• Updated std::enumerate to use the zero macro
• Added two more test cases (enumeration of struct and empty array)
• Allowed empty range for loops (mentioned here: [enhancement] dslx should support empty for loops #3965)
• Allowed indexing into zero-sized arrays for constexpr

Please let me know if this solved the issue on your side as well.

[richmckeever]

It seems like here and in other test files, we should keep empty range/array tests but change the assertions to expect success, and add a const_assert! in the DSLX that expects it to be empty.

We should also add tests for an empty range with the start being the max of the type, for both signed and unsigned, because there is some logic which normalizes exclusive ranges by converting them into inclusive with end + 1.

[joannabrozek]

I restored RangeExprEmptyRange test and added two more to the same file.

[joannabrozek]

@richmckeever, thank you for the review. I've rebased the code and added the suggested tests. Let me know if there's still something missing.

[proppy]

do we also need to update documentation and samples?

[magancarz]

@proppy I've opened a pull request with enumerate docs update in #4326 and also one for removal of obsolete builtin test in #4327.