I wonder if we should also send the header in some of the conditions where we early return without a content-encoding, to indicate to caches that a different response body is available when accept-encoding is set. Wdyt?

So we always send the Vary header whatever the body will be? You're right, we should do this.
I will fix that.

I have a question about branches: why has the v0.4.x branch (and the 0.4.3 version) diverged from master?
So what base branch should I target with this PR? master or v0.4.x?

So we always send the Vary header whatever the body will be? You're right, we should do this. I will fix that.

No, not in every case. If should_compress is false, the body will never be compressed no matter which headers the client sends AFAIU. However, for Encoding::Identity, the body was not compressed specifically of what headers the client sent (again, AFAIU).

I have a question about branches: why has the v0.4.x branch (and the 0.4.3 version) diverged from master? So what base branch should I target with this PR? master or v0.4.x?

It's okay to make PRs against either. I guess if you want this PR to be released (soon-ish) as part of 0.4.x, it makes sense to make the PR against that. There's (so far) only a tiny difference between the branches, so it will be trivial to cherry-pick changes such that they affect both branches.

So we should always send the Vary except when should_compress is false?

No, I think there's another early-return branch (body already compressed, i.e. content-encoding already set) in which the middleware should essentially do nothing.

cc @davidpdrsn @Nehliin wdyt about when we should set vary?

The Content-Encoding already set is tested in should_compress:

let should_compress = !res.headers().contains_key(header::CONTENT_ENCODING)
    && self.predicate.should_compress(&res);

Thus if I guard the Vary header with an if should_compress, the header will be added unless the response is already compressed or the response is never be compressed.

Ping @davidpdrsn @Nehliin any opinions about this PR / the question of when vary should be sent?

Also @Dragonink have you looked into the test failure? (MSRV job failing was fixed in #418)

sorry won't have time to review in the near future