Create a serde1-alloc feature flag

[TethysSvensson]

Currently the bstring_serde module is hidden behind the serde1 feature flag, however this flag also pulls in the std flag, which is unnecessary.

This PR adds an additional feature flag to enable that impl without depending on the entire std flag.

Alternatives considered

• Change the serde1 feature flag to only depend on ["alloc", "serde1-nostd", "serde/alloc"]. I do not know if you consider this backwards incompatible (or if you care).
• Simplify the feature flags by bumping the MSRV to 1.60 and using the new cargo features.

[BurntSushi]

Hmmm yeah, good catch. I think this is the right path. To respond to your alternatives:

Change the serde1 feature flag to only depend on ["alloc", "serde1-nostd", "serde/alloc"]. I do not know if you consider this backwards incompatible (or if you care).

I'm just about to release 1.0, so this is the time to make backwards incompatible changes.

However, I think making serde1 depend on alloc like this probably doesn't work. Because then the question arises: how do you enable serde/std? I can't put it in as a dependency of the std feature, because then that will make serde an unconditional dependency on bstr whenever std is enabled. That's no good.

Simplify the feature flags by bumping the MSRV to 1.60 and using the new cargo features.

Yeah I think Rust 1.60 is just too new. This is a little unfortunate because I suspect moving to this will be a breaking change and I do not expect to release a bstr 2.0 any time soon (if ever).

[TethysSvensson]

how do you enable serde/std?

Since none of the bstr code depends on serde/std feature, one option would be to punt the problem to the user: If they require the serde/std feature they have to enable it themselves.

[BurntSushi]

That is an option, but one that seems higher risk with respect to backcompat. e.g., If bstr ever does care about serde's std support.

[TethysSvensson]

Then perhaps exposing three features like in this PR might be the best option?

Though if you don't care about breaking changes right now, then they could be renamed to serde-std, serde-alloc and serde-nostd?

[BurntSushi]

Right, that's what I plan to do: #40 (comment)

But serde1-core instead of serde1-nostd, since nostd can refer to either alloc-only or core-only.

Sorry if I wasn't clear about that. I was mostly just trying to write out a justification for why the alternatives were considered, but ultimately that 3 different features seems like the right way to go.

[TethysSvensson]

Since you are renaming anyways, why not get rid of the 1?

[BurntSushi]

In case serde has a major version bump.