Some features were removed instead of deprecated/stabilized

[jethrogb]

From @varkor in #52644 (comment):

it seems some past features were simply removed outright rather than being deprecated as they should be. ... I think we'll need to retroactively make these features deprecated.

This applies at least to the following features:

• collections
• str_char
• unicode
• alloc

There might be others.

[Centril]

If they were stabilized they should be added to the accepted list and otherwise to the removed list.

[rasendubi]

1. All of the mentioned flags are lib flags. Is it OK to add them to feature_gate.rs? (I am not sure as these are "lang" features and might be handled differently.)

2. e.g., fs_read_write was stabilized as fs_read_write_bytes (0600d0f). Should we add the old name? as accepted?

3. A part of alloc was renamed to alloc_prelude (without stabilization) (5d1022a). The rest was stabilized as alloc (fc928a1). @jethrogb are you sure you get an error when using alloc?

(If anyone feels like doing code archeology, a simple git log -p -G '"str_char"' seems to give pretty good results.)

[jethrogb]

2. e.g., fs_read_write was stabilized as fs_read_write_bytes (0600d0f). Should we add the old name? as accepted?

I think they should be added as removed

3. ... @jethrogb are you sure you get an error when using alloc?

I thought I was with some nightly version between nightly-2018-08-14 and nightly-2019-04-27, but I can't reproduce this right now.

[varkor]

1. All of the mentioned flags are lib flags. Is it OK to add them to feature_gate.rs? (I am not sure as these are "lang" features and might be handled differently.)

Yeah, I'm not sure whether we want to flag removed libs features differently. It'd probably be nice to keep a record, but I'm not sure if the difference is valuable.

2. e.g., fs_read_write was stabilized as fs_read_write_bytes (0600d0f). Should we add the old name? as accepted?

I agree with @jethrogb: we should probably mark it as removed. (Alternatively, we could track "renamings of feature flags", like we do for lints, so we could produce better diagnostics.)

[Centril]

It's rather weird to add libs features to the language feature list. I also think it's not the end of the world to just throw away the removed feature gate list entirely (dump it in a document somewhere outside of the code); it's essentially technical debt that we are monotonically increasing at the moment. The only think that we'd need to do is to not error. Adding these features, which seemingly hasn't bothered anyone for a long time, only for the purposes of consistency seems to be a weak motivation.

[Enselic]

Triage: Since many years has passed now the risk of messing with feature flags seems greater than the benefit of fixing them (and I'm unsure exactly what that means atm tbh). I'll go ahead and close this with this motivation from above comment:

Adding these features, which seemingly hasn't bothered anyone for a long time, only for the purposes of consistency seems to be a weak motivation.