ArgumentError in Sibyl.Handlers.attach_all_events/2 in docker

[florius0]

I was evaluating Sybil and came across this error:

** (Mix) Could not start application fish_finder: exited in: FF.Application.start(:normal, [])
     ** (EXIT) an exception was raised:
         ** (ArgumentError) errors were found at the given arguments:

   * 1st argument: invalid destination

             :erlang.send(:undefined, {:shell_state, #PID<0.3858.0>})
             (stdlib 5.0.1) shell.erl:590: :shell.get_state/0
             (stdlib 5.0.1) shell.erl:597: :shell.get_function/2
             (stdlib 5.0.1) shell_default.erl:159: :shell_default."$handle_undefined_function"/2
             (sibyl 0.1.2) lib/sibyl/events.ex:105: Sibyl.Events.reflect/1
             (elixir 1.15.0) lib/enum.ex:4317: Enum.flat_map_list/2
             (elixir 1.15.0) lib/enum.ex:4318: Enum.flat_map_list/2
             (sibyl 0.1.2) lib/sibyl/handlers.ex:20: Sibyl.Handlers.attach_all_events/2
             (fish_finder 0.1.0) lib/fish_finder/application.ex:29: FF.Application.start/2
             (kernel 9.0.1) application_master.erl:293: :application_master.start_it_old/4

It was not reproduceable on my local elixir, but only in docker.

My repo using which the bug may be reproduced.

From what I'm gathering, :shell.__info__ results not in exception, but in message sending to :undefied pid.

Versions

Local:

• OS: macOS 13.4
• Erlang/Elixir:

Erlang/OTP 25 [erts-13.2.2.1] [source] [64-bit] [smp:12:12] [ds:12:12:10] [async-threads:1] [jit:ns] [dtrace]

Elixir 1.15.0 (compiled with Erlang/OTP 25)

Docker:

• Runtime: Orbstack 0.16.1
• bitwalker/alpine-elixir-phoenix:1.15.0

[vereis]

Beautiful, thanks for the reproduction steps.

Will take a look and get back to you soon 🤞

[florius0]

I might be able to fix it today or narrow the problem at least

[florius0]

Using module_info/2 instead of __info__/2 in Sibyl.Events.reflect/1 seems have fixed the issue.

In my fork I've updated to Elixir 1.15.0 (tho I'm not sure yet that it didn't caused any problems), and renamed guards and predicates to match elixir conventions (is_... for guards, ...? for predicates).

If you would like I can include those changes in PR.

PS
I would like to suggest to run your CI on all minor versions from lowest supported up to latest

[vereis]

@florius0 that sounds awesome! Looking forward to seeing a PR 🙌 Thank you!

And yes, running CI against a matrix of supported versions (defining what those are explicitly) will be great too, thank you for the suggestion!

[florius0]

Tried it on elixir:1.15.4-otp-25-alpine, same issue. Perhaps it has to do smth with alpine. My fix works tho

[rob-vetspire]

I can confirm issue #30 also occurs when running an application with elixir 1.15.7-otp-26 and erlang 26.1.2 directly on macos.

[florius0]

I guess it is related to new stripping of unused functions by elixir compiler maybe...

[vereis]

@florius0 after investigating, I think we should just skip all Erlang modules when doing reflection, as reflection will only discover events defined by Sibyl, and that can only happen in Elixir applications.

I think this is probably a better fix than relying on module.module_info, as at the very least, the __info__/1 callback is documented: https://hexdocs.pm/elixir/1.7.3/Module.html

Have the tentative fix in #34, will just need to do some regression testing but I believe this fixes the problem!

Apologies for the delay!

[florius0]

I don't think that restricting Erlang compatibility is a good option especially when there is no cost of keeping it.

If my memory serves me, Sybil defines events as module attributes (with persist: true), and getting such attribute with module_info/1 is ok.
It is not documented in Elixir docs since it is Erlang callback: https://www.erlang.org/doc/reference_manual/modules#module_info-0-and-module_info-1-functions.

As for delay, do not worry, we use my fork for the time being.

[vereis]

Ah I wasn't aware of that. Will revisit my change, thanks ❤️

[vereis]

Resolved in #34