Closed campbellC closed 3 years ago
https://github.com/campbellC/third-wheel/commit/939b7515ff0e3292ec4e40b1e0e5cb41ed370c59
I've added the harness needed for testing third wheel. Its nearly all rust although I call out to openssl for one bit since I still haven't written the rust code to create a CA. Anyway, this harness:
I've added a single test 'foo' that verifies the whole chain of trust is working. There's plenty of room for improvement here but this means this task can just become: write a whole bunch of tests verifying behaviour that we want. The warp server and the mitm need to be more configurable to allow testing of in-stream modifications, and testing against different http sessions.
The second set of tests I think would be ideal here would be having one client that uses the mitm and one that does not, and verifying that they see identical traffic. Likewise on the server side. Unfortunately, reqwest does not currently allow dns bypass which I believe would be required for this. I will take a swing at implementing that for reqwest and see if I can unblock this better kind of testing.
I'm unlinking this from beta since the basic testing is up and running and will be more fleshed out as features are figured out.
Hopefully this PR will be merged and this will unblock more full featured testing. If not, we'll need to switch clients in the test package.
Apparently crates allows dev-dependencies to depend on github forks so there's no need to wait for the PR. I've added some tests that compare proxied and non-proxied requests and responses to verify the proxy is doing its thing. I think this suffices.
Currently we have a test that loads 'https://example.com' in chrome, firefox and curl and verifies that the content of the page is as expected. This is fine but doesn't exercise basic functionality (as seen in this issue) where it wasn't passing through query parameters.
The testing should cover all the HTTP verbs for a standard site, and should verify connection persistence (e.g. using 206 responses for large bodies).
Possible approach:
I think it is acceptable for this to be done using curl as the basis for the verifier as an initial improvement, but ideally we would be able to override the DNS for the Chrome and Firefox tests as well.