Here is some sort of notable changes:

1. response_content no longer support chunk_size since httpx does not support this. (https://github.com/encode/httpx/issues/394)
2. gitlab clients are moved to client.py
3. 3 implementations of gitlab clients: gitlab.BaseGitlab abstract implementation that share common methods and store all docsrings, gitlab.Gitlab is sync implementation with httpx.Client, gitlab.AsyncGitlab is async implementation with httpx.AsyncClient.
4. on_http_error make nested since we can get coroutine and need to catch same error in awaiting response
5. RESTObject must be created (with create()), not initialized, since arguments could be coroutines, and need to be awaited, and __init__ does not support that interface
6. Everything should be returned, since it can be couroutin
7. every postprocess of response should be a function and be decorated with gitlab.utils.awaitable_postprocess, as it's done with _update_attrs and other edge cases
8. unit tests are run by pytest and run against async and sync client implementations
9. since unit tests are run with async gitlab, tests are required to be async, but fetching result is done through fixture that knows which client is used and return value if that's sync version or return await value if that's async version.
10. GitlabList support two versions of iteration
11. requests are completely removed

1. `response_content` no longer support `chunk_size` since `httpx` does not support
    this. ([encode/httpx#394](https://github.com/encode/httpx/issues/394))

I just noticed something: chunk_size is a configurable argument in several public methods (for blobs/snippet content, export download, etc) so it's possible people are using it and this would be a breaking change, right? Since you're replacing it with aiter_bytes and using chunks there, is there any way to preserve backward compatibility (I haven't checked the implementation)? Or wait for the upstream issue to resolve?

3. 3 implementations of gitlab clients: `gitlab.BaseGitlab` abstract implementation that share common methods and store all docsrings, `gitlab.Gitlab` is sync implementation with `httpx.Client`, `gitlab.AsyncGitlab` is async implementation with `httpx.AsyncClient`.
9. since unit tests are run with async gitlab, tests are required to be async, but fetching result is done through fixture that knows which client is used and `return value` if that's _sync_ version or `return await value` if that's `async` version.

That's really cool :)

@vishes-shell Thanks for all the work that you're doing!

People could have been using chunk_size, by providing a different requests instance, right?
(Which is also not supported anymore then?)

But we have nothing about this officially in the docs, so I would say we can solve this by mentioning it in the release notes. We anyway need to bump the version to 3.X.

@max-wittig I tried earlier and at least in some cases it can even be just:

project = gl.projects.get(1)
export = project.exports.create({})
dl = export.download(chunk_size=512)

From what I see in the current solution by @vishes-shell this would still work and wouldn't break anything, the argument would just be ignored by response_content(). But if anyone relies on chunk_size for some specific reason, it might break their behavior. I'm not sure of all the use cases but it seems useful especially for project exports: https://stackoverflow.com/questions/46205586/why-to-use-iter-content-and-chunk-size-in-python-requests/46205745

What's the current state?

When using your library, I get the following warning:

httpx-0.18.2-py3.9.egg/httpx/_client.py:1978: UserWarning: Unclosed <httpx.AsyncClient object at 0x7ffabf61e280>. See https://www.python-httpx.org/async/#opening-and-closing-clients for details.

Apart from that, it works like a charm 😍

Btw: chunk_size is supported by httpx now.

When using your library, I get the following warning:

httpx-0.18.2-py3.9.egg/httpx/_client.py:1978: UserWarning: Unclosed <httpx.AsyncClient object at 0x7ffabf61e280>. See https://www.python-httpx.org/async/#opening-and-closing-clients for details.

Apart from that, it works like a charm 😍

You'll need to use the client with async with AsyncGitlab() as gitlab: ...

I am using it since a few months without any problems. chunk_size is implemented by now as well. Any other blockers? Or would it be accepted when updated to implement the latest features in the main branch?

As part of onboarding myself into this project, I stumbled into this PR. Could someone provide a use case when an async wrapper is being used?

@lmilbaum We are evaluating the jobs from the last day via script such that our QA team has a good overview about possible regressions or problems with our hardware text boxes (e.g. when the same code runs on one runner but not the other one). As network is slow, waiting for the answer of the previous request before submitting the next one takes quite a lot of time. If we can submit all requests asynchronously and just wait for all of them to finish in parallel, we have reduced the run time of our script by a multitude. Unfortunately, this MR is outdated and newer features are not available yet...

The wrapper here https://github.com/pan-net-security/aio-gitlab/ has a few points similar to what @darkdragon-001 described above. But true, this PR would need quite a bit of rework, or be used as starting point for a new one, since the library has changed a bit since then. I personally have not yet had enough need for this to work on it.

Another option would be to first switch from requests to httpx for the sync client (with a breaking change, because we're currently allowing people to bring their own requests sessions) and smooth out any kinks, then add async later, to make the transition less painful.

@darkdragon-001 and @nejch thanks for your feedback. Now, I can understand better the need for an async capability.
Would it make sense to move this discussion to Discussions while closing this PR?
The first discussion point might be the migration from requests to httpx as suggested by @nejch.

@darkdragon-001 and @nejch thanks for your feedback. Now, I can understand better the need for an async capability. Would it make sense to move this discussion to Discussions while closing this PR? The first discussion point might be the migration from requests to httpx as suggested by @nejch.

@lmilbaum thanks for looking into this! I'd keep this PR open even if as draft, just so it stays on our radar as there's a lot of useful discussion and work done here. And just so we remember to credit @vishes-shell as author whether they want to still work on this or not.

I'd maybe open a new issue to plan the switch to httpx as that would be for v4.0.0. We could then get rid of types-requests and requests-toolbelt I assume, as httpx is fully typed and does multipart encoding well IIRC.

This Pull Request (PR) was marked stale because it has been open 90 days with no activity. Please remove the stale label or comment on this PR. Otherwise, it will be closed in 15 days.

This PR was closed because it has been marked stale for 15 days with no activity. If this PR is still valid, please re-open.