0.8.29 should be a major bump

[l4l]

Commit cecda92 introduces an alloc feature and adds it to the defaults. Unfortunately this breaks semver, in particularly for those crates that use cargo parametr default-features = false. As an example you may easily reproduce it with rust-fontconfig = "0.1.5" dependency, its output:

error[E0599]: no method named `decode_to_string_without_replacement` found for struct `Decoder` in the current scope
  --> /home/kitsu/.cargo/registry/src/github.com-1ecc6299db9ec823/allsorts_no_std-0.5.2/src/get_name.rs:94:36
   |
94 |         let (res, _read) = decoder.decode_to_string_without_replacement(data, &mut s, true);
   |                                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: there is an associated function with a similar name: `decode_to_str_without_replacement`

error: aborting due to previous error

and manually downgrading encoding_rs = "0.8.28" in Cargo.lock solves the problem.

[hsivonen]

That's sad. Since there were no default features previously, my assumption was that no one would have had a reason to do default-features = false and a one-time addition of default features would be semver-compatible.

Do I understand correctly that default-features = false is transitive so that rust-fontconfig setting default-features = false for allsorts_no_std ends up causing allsorts_no_std to propagate default-features = false to encoding_rs despite allsorts_no_std not saying so?

[l4l]

Actually there's something strange published for allsorts_no_std, it's cargo.toml differs from the github sources. It directly disables features of encoding_rs. Apparently rust-fontconfig is redundant for the example.

[hsivonen]

I've examined quite a number of crates the depend on encoding_rs, and allsorts_no_std is an exception by disabling default features before there were any to disable. Instead of yanking the latest encoding_rs right away, I'm going to see if a new release of allsorts_no_std could be made.