Refactor code server to use a single queue

[josevalim]

Prior to this patch, the code server had two
internal queues, one to track module loading
and another to track on_load callbacks. This
pull requests refactors the code to have a
single queue, in order to fix bugs and improve
maintainability.

Closes #7466.
Closes #8510.

[github-actions]

CT Test Results

    2 files     70 suites   1h 5m 36s ⏱️
1 554 tests 1 313 ✅ 241 💤 0 ❌
1 770 runs  1 497 ✅ 273 💤 0 ❌

Results for commit 72c91bb.

♻️ This comment has been updated with latest results.

To speed up review, make sure that you have read Contributing to Erlang/OTP and that all checks pass.

See the TESTING and DEVELOPMENT HowTo guides for details about how to run test locally.

Artifacts

• Complete CT logs (Download Logs)
• HTML Documentation (Download HTML Docs)
• Windows Installer

// Erlang/OTP Github Action Bot

[josevalim]

Ideally this would be merged on maint because it fixes a regression on OTP 27 but it is a conceptually large change. My suggestion is to merge it to master and, once 27.1 is out, backport it to maint, which should give us time to battle test it for 27.2.

[kikofernandez]

Thanks for your contribution. I plan to look at this after ICFP (in 10 days) approx

[josevalim]

Please don't merge this yet. I have tried this branch in a project and, while I didn't notice any bugs, compilation times for Elixir (which loads several modules) was affected negatively.

[josevalim]

False alarm, I had mixed my Erlang versions. Overall, I didn't measure any performance degradation in my usual code loading benchmarks. :)

[kikofernandez]

Looking into it this week

[kikofernandez]

This was on hold due to other investigations regarding current issues in code_server.
As soon as I am given green light, I review this PR

[kikofernandez]

Looks good to me
These changes make the code_server easier to follow.
We are running some internal tests and will merge next week if all is green

[josevalim]

Glad to hear so! 🎉 Code loading is a state machine and the previous version was encoding the states via anonymous functions and in two separate queues, which made it very hard to follow. :)

[kikofernandez]

Completely agree, it was pretty tricky to follow the logic. This was a great improvement.

Since the next OTP release 27.2 is around beginning of December, I am wondering if we can place this already on maint, instead of master. That gives us close to two months of testing

[josevalim]

As long as we are comfortable with potentially doing a 27.2.1 with this commit reverted, it sounds good to me. :D