Remove disabled optional weak dependencies when resolving

[Jackhr-arch]

I am not a native English user, please forgive my flaws in description.

What does this PR try to resolve?

it changes the parts of the resolver that generate Cargo.lock to not include weak dependencies.

This changes the logic when resolver checking features. Instead of add weak feature as strong one, now resolver keeps the weak dependency features until the very dependency is checked. Then it combines the weak features with the original ones.

E.g.

[features]
fa = ["da/daf", "db", "dc?/dcf", "dd?/ddf"]
fb = ["da", "db?/dbf"]
fc = ["da?/daf", "dc"]
default = ["fa", "fb", "fc"]

Now resolver acts like:

feature	dep:da	dep:db	dep:dc	dep:dd	weak dep feat
fa	with "daf"	enabled	disabled	disabled	"dc?/dcf","dd?/ddf"
fb	with "daf"	with "dbf"	disabled	disabled	"db?/dbf"
fc	with "daf"	with "dbf"	enabled	disabled	"da?/daf"

then it works on weak dependency features.

weak dep feat	dep:da	dep:db	dep:dc	dep:dd
dc?/dcf	with "daf"	with "dbf"	with "dcf" (here reslover enabled the weak feat)	disabled
dd?/dcf	with "daf"	with "dbf"	with "dcf"	disabled
db?/dcf	with "daf"	with "dbf"	with "dcf"	disabled (here reslover ignore the weak feat)
da?/dcf	with "daf"	with "dbf"	with "dcf"	disabled

Relate to #10801

How to test and review this PR?

I've enabled some tests in crates/resolver-tests relating to Weak dependencies and looks they work fine.

Here is an example repo https://github.com/Jackhr-arch/cargo-weak-dep-feat-example

[rustbot]

r? @weihanglo

rustbot has assigned @weihanglo.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[epage]

I won't be up for reviewing the core part of this, so my comments are a pre-review.

Please clean up commits for how they should be reviewed and merged.

r? @Eh2406

[epage]

Technically, this is a breaking change though we'll have to evaluate our willingness to accept this

[Jackhr-arch]

Due to the new resolver is gated, cargo now restore its old behavior. I reverted this change.

[Jackhr-arch]

The root cause is that I remove the non-activated weak dependency from root crate, too.

Now weak direct dependency is always included in Cargo.lock, with its test disabled_weak_direct_dep.

Previous change to tree and other tests is removed permanently.

[epage]

Isn't this redundant with line 512?

[Jackhr-arch]

You are right, I missed this. Now it's fixed.

[est31]

I haven't done a review of the approach, I wonder whether this supports the weak_namespaced_v4 test from my earlier PR, or if it fails similarly (see ehuss comment and discussion immediately below).

[Jackhr-arch]

I wonder whether this supports the weak_namespaced_v4 test from my earlier PR

It passed. I port 4 tests from your pull request, and deferred_v5 failed. I'm trying to fix this.