classfile: decode CONSTANT_Utf8 as modified UTF-8

[Jun025]

CONSTANT_Utf8 entries are modified UTF-8 (JVMS §4.4.7), but parse_utf8 decodes them with String::from_utf8. Two encodings javac emits are rejected:

• NUL is written as C0 80
• a supplementary character is written as its UTF-16 surrogate pair, each half as a 3-byte sequence (ED A0..AF xx ED B0..BF xx)

Any class with "\0" or, for example, U+1F40D in a string constant fails to parse, and loading it throws ClassFormatError.

Repro: compile class A { String s = "MTR\0"; } with javac and parse the resulting class file.

Fix: decode_modified_utf8 keeps the str::from_utf8 fast path for input that is already valid UTF-8. Otherwise it decodes 1–4 byte sequences into UTF-16 units and pairs them with char::decode_utf16. So the change only widens what is accepted:

• standard 4-byte UTF-8 still decodes, as it did before
• a lone surrogate, which is legal in Java but cannot be stored in a Rust String, becomes U+FFFD
• truncated sequences, stray continuation bytes and invalid lead bytes still fail

Tests: four unit tests in constant_pool.rs:

• C0 80
• a surrogate pair
• ASCII and non-ASCII BMP text unchanged
• edge cases of the wider decoder

With String::from_utf8 restored, three of the four fail. cargo test --workspace, cargo fmt --check and cargo clippy -p classfile --all-targets -D warnings pass.

🤖 Generated with Claude Code

[dlunch]

Thanks!

[dlunch]

Thanks!