@mjpieters do you want me to reject the release build, wait and include this PR?

Not sure why the changelog bot is failing to notice the towncrier fragment... Not sure why but it's safe to ignore for now.

[towncrier-fragments]:13: : Spell check: hostnames: Started rejecting ASCII hostnames with invalid characters. For host strings that.

@mjpieters this can be fixed by adding "hostnames" to docs/spelling_wordlist.txt. Though, I hope to integrate codespell later that would replace sphinxcontrib-spelling and seems to have less false positives.

While implementing my PR I noticed that if you used a non-ASCII hostname with @ or any other invalid character (so not unreserved, percent encoded or sub-delims), then idna.encode() would raise an exception. But when I tried to reproduce this behaviour, just now, I noticed we swallow that exception, because it is a UnicodeError subclass 🤦 .

I'm not sure why the encoder uses except UnicodeError with a fallback to host.encode('idna').decode('ascii'), because that completely defeats the valid codepoint check.

I can hoist the validation to apply to non-ASCII hostnames too but that means I need to explicitly test for all disallowed ASCII codepoint (instead of the negative character classes I use now).

I'll go search through the commit history to see if I cannot find out more here.

@mjpieters this can be fixed by adding "hostnames" to docs/spelling_wordlist.txt. Though, I hope to integrate codespell later that would replace sphinxcontrib-spelling and seems to have less false positives.

:-D I already have that change staged locally :-P

@mjpieters so do you think it's worth putting into a separate release? The automation is there so it wouldn't take as long to make new releases from now on...

Not sure why the changelog bot is failing to notice the towncrier fragment... Not sure why but it's safe to ignore for now.

It is looking for the files in the wrong location, news/ instead of CHANGES/.

@mjpieters do you want me to reject the release build, wait and include this PR?

I think it's fine to put this in a next release. I don't see this as urgent.

@mjpieters do you want me to reject the release build, wait and include this PR?

I think it's fine to put this in a next release. I don't see this as urgent.

Okay, will proceed with releasing, then. The build is half-way ready: https://github.com/aio-libs/yarl/actions/runs/6931722870.

I'll raise a separate issue for IDNA host encoding. I can see why the UnicodeError exception handler is there but I think it needs some reworking.

This PR can focus on ASCII hostnames only, as a first step.

Okay, will proceed with releasing, then. The build is half-way ready: https://github.com/aio-libs/yarl/actions/runs/6931722870.

What a glorious deployment pipeline we have! 😁

Not sure why the changelog bot is failing to notice the towncrier fragment... Not sure why but it's safe to ignore for now.

It is looking for the files in the wrong location, news/ instead of CHANGES/.

Oh, that makes sense — this is because I moved the config to a dedicated file and haven't updates the bot to account for that. Thanks for pointing it out!

Okay, will proceed with releasing, then. The build is half-way ready: https://github.com/aio-libs/yarl/actions/runs/6931722870.

What a glorious deployment pipeline we have! 😁

Hopefully, I'll make it scalable and reuse for other projects. I have similar things in many other repos but they all have tiny differences which makes them hard to reuse as is.
I made the CI run tests from the exact sdists we publish and install the wheels instead of rebuilding from Git source in each job. Though, I broke codecov reporting along the way so ignore it for now, I'll fix it later, hopefully.

Okay, will proceed with releasing, then. The build is half-way ready: https://github.com/aio-libs/yarl/actions/runs/6931722870.

It took a few reattempts to get the last bit right... But now we have v1.9.3:

• https://pypi.org/project/yarl/1.9.3/
• v1.9.3 #957
• https://github.com/aio-libs/yarl/releases/tag/v1.9.3

Though, I broke codecov reporting along the way so ignore it for now, I'll fix it later, hopefully.

Fixed it. The coverage metrics are back up!

I'm not sure why the encoder uses except UnicodeError with a fallback to host.encode('idna').decode('ascii'), because that completely defeats the valid codepoint check.

I checked the coverage reported before I broke it temporarily, and turns out that the line you're talking about is the only one uncovered in the entire project: https://app.codecov.io/gh/aio-libs/yarl/commit/65333fab5f5b2bcce9fbb6aa810e9618fa10c6f0/blob/yarl/_url.py#L1176.

So I guess this might also need some extra extensive testing...