Summary

Struct() (and the module-level struct.pack / unpack / calcsize) raised the wrong exception type when the format argument contained non-ASCII characters:

Both error type and message diverge from CPython.

Fix

Restructure IntoStructFormatBytes::try_from_object in crates/stdlib/src/pystruct.rs:

• str input with non-ASCII char → raise UnicodeEncodeError("ascii", s, start, start+1, "ordinal not in range(128)"). The position is the first non-ASCII code point, matching what format.encode('ascii') would naturally produce in CPython.
• bytes input with non-ASCII byte → raise struct.error("bad char in struct format") via the existing new_struct_error helper. CPython gets here because it passes the bytes through to the format parser, which then errors on the bad char; we emit the equivalent error directly without the indirection.

Other paths (str ASCII, bytes ASCII, non-str/bytes argument) are unchanged.

Verification

Targeted probes (CPython 3.14.4 byte-identical)

$ rustpython -c 'import struct; struct.pack("é", 42)'
UnicodeEncodeError: 'ascii' codec can't encode character '\xe9' in position 0: ordinal not in range(128)

$ rustpython -c 'import struct; struct.pack(b"\xe9", 42)'
struct.error: bad char in struct format

Identical to CPython 3.14.4 output for both.

Test suite

$ rustpython -m unittest test.test_struct
Ran 43 tests in 0.6s
OK (skipped=6, expected failures=8)

No regressions; expected-failures unchanged (none of the 8 expected failures cover this code path).

Sanity check on valid usage

Verified that standard struct.pack / unpack / calcsize with valid str or bytes format strings, and the 's' / 'p' byte-string formats, all behave as before. No behaviour change on success paths.

Pre-push

• cargo fmt --all --check clean
• cargo clippy -p rustpython-stdlib --all-targets -- -D warnings clean
• prek run --all-files — all 8 hooks pass

Scope

1 file, +30/−8 lines. Single function. The only behaviour change is the exception type and message for non-ASCII format input, which previously produced an incorrect UnicodeDecodeError with an empty message.

Out of scope (separate PRs)

While probing this fix, two further struct divergences were observed:

• struct.pack('5s', 'hello') raises TypeError in RustPython but struct.error in CPython. This is a separate code path in the format-spec packer.
• struct.pack('Q@', 42) is silently accepted in RustPython but raises struct.error: bad char in struct format in CPython. The shared FormatSpec::parse (also used by memoryview) silently skips endianness markers mid-format; addressing this requires deciding a parser-mode strategy.

Both deferred to follow-up PRs.

Summary by CodeRabbit

• Bug Fixes

  • struct format parsing now yields more accurate errors: string format inputs raise a UnicodeEncodeError with CPython-like message/position; bytes format inputs raise a clearer struct.error for invalid characters.

• Tests

  • Added tests verifying non-ASCII behavior for both string and bytes struct format inputs.