feat(hackney)!: support hackney 4.x by tomekowal · Pull Request #880 · elixir-tesla/tesla

tomekowal · 2026-05-31T10:17:27Z

Adds support for Hackney 4.x.

There are two workarounds in this PR that are further explained in comments:

I kept Hackney 1.x in mix.lock because Hackney 4.x depends on quic which requires Erlang 26 which breaks the CI matrix; AFAIU, the only way to upgrade Hackney to 4.x in mix.lock would be to drop Erlang 25 from the CI matrix
Second commit introduces a function that does nothing except correcting wrong typespec in hackney for dialyzer to pass.

I made comments in places that I believe require attention during review.

tomekowal · 2026-05-31T10:54:33Z

I've changed the PR title, but the action did not rerun. @yordis Could you rerun it?

yordis · 2026-05-31T21:41:25Z

@tomekowal thank you so much, I wonder, should we leave TODO comments with the proper context in the code to later remove 1.0 support?

tomekowal · 2026-06-01T13:31:17Z

Do we want to remove 1.0 support? I don't believe keeping both in the same adapter adds that much complexity. There is not a single if statement necessary. The only small thing is checking if hackney reference is either Erlang's reference or PID.

It would be totally fine to leave both 1.x and 4.x support indefinitely.

halostatue · 2026-06-01T14:37:49Z

As someone adjacent to security…I don't think keeping 1.0 is a good idea. It's not getting any more updates and has three unresolved CVEs.

But that changes a number of things for consumers including the minimum supported OTP.

yordis · 2026-06-01T17:12:05Z

I am in a tricky situation here, it is effectively a breaking change for Tesla ... 😭 What do I do? Maybe I shouldn't feel so emotionally attach to cut a v2 or put a big disclaimer in the release note?

tomekowal · 2026-06-01T17:26:07Z

I think that makes sense for a major upgrade:

cut the support for OTP 25
cut support for Hackney 1.x
any other stuff that you would like to do but compatibility held you back?

yordis · 2026-06-01T17:28:54Z

use feat(hackney)! the ! is the key here, lets break changes 😢

tomekowal · 2026-06-01T19:33:20Z

I've change the title. Would you like to me to make other changes that drop Hackney 1.x support completely?

yordis · 2026-06-01T19:35:27Z

yeah, drop it completely by now

BREAKING CHANGE: drops hackney 1.x. Tesla now requires hackney >= 4.0.2 and OTP 26+ (transitively via :quic). - Adapter targets hackney 4.x conn() = pid() everywhere - Stream upload path calls finish_send_body before start_response (4.x requires explicit chunked terminator) - :max_body honored only on streaming response read; 4.x sync requests always return body inline - start_stream_request wrapper restores {:ok, pid()} return shape that hackney's request_ret/0 omits, keeping dialyzer happy Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Hackney 4.x pulls in :quic which requires OTP 26+. Bump the gun1 lockfile job to the next pair (1.16.3 / 26.2.5) to keep gun 1.x coverage. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

excoveralls 0.16.1 hard-required hackney ~> 1.16, conflicting with the new hackney ~> 4.0 constraint. 0.18.5 dropped the mandatory hackney dep (castore optional instead). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

tomekowal · 2026-06-01T20:15:17Z

OK, so:

I've removed traces of Hackney v1
Removed Erlang 25 from test matrix
updated excoveralls in gun lockfile (because updating hackney caused errors)
the Semantic PR Title failure seems to be not from the title itself but from how the job works

yordis · 2026-06-02T17:07:03Z

@tomekowal thank you so much for the help!

tomekowal · 2026-06-02T19:08:17Z

Happy to help :)