encoding_rs_compat 0.2.33

encoding_rs_compat provides the rust-encoding 0.2.33 API implemented on top of encoding_rs. Technically, encoding_rs_compat is a fork of rust-encoding 0.2.32 with the internals replaced. The use case was to allow Gecko to use crates that depend on rust-encoding without having to include duplicate data tables and converter functionality, but Gecko ended up not needing this compatibility crate after all.

Usage

Put this in the Cargo.toml file of your top-level Cargo artifact:

[replace]
"encoding:0.2.33" = { git = 'https://github.com/hsivonen/encoding_rs_compat' }

Upon cargo build, ensure you see don't see the encoding-index-* crates being built.

Differences from rust-encoding

• The bugs in the converters and the spec snapshot they implement are those of encoding_rs.

• ISO-8859-1 as an encoding distinct from windows-1252, HZ, and the error encoding are not supported.

• Attempting to encode to UTF-16LE or UTF-16BE panics.

• The types of the constants in all::* differ.

• The "constants" in all::* are static instead of const, because they need to refer to statics and Rust, very annoyingly, doesn't allow const to refer to even an address of a static.

• There is no direct access to the indices.

• The codec module isn't visible.

• RawDecoder.raw_feed() always identifies a zero-length byte sequence as being the erroneous one. (This is due to encoding_rs not backtracking when a byte sequence that might turn out to be erroneous later is split across a buffer boundary. The entry points that take a DecoderTrap pass the erroneous bytes to the DecoderTrapFunc correctly, however.)

• While RawEncoder.raw_feed() signals unmappable characters the same way as rust-encoding, which cannot represent the current spec requiring certain unmappables in ISO-2022-JP to be reported as U+FFFD, unmappable characters passed to EncoderTrap are reported as U+FFFD where required by the spec for ISO-2022-JP.

• The performance profile of custom ByteWriter and StringWriter differs from the performance profile of the default Vec<u8> and String. The former involve an extra intermediate copy of the output while the latter run at the native speed of encoding_rs.