Update Dependencies and test on Elixir 1.14 and 1.15 #738
Conversation
|
I opened up a PR on my fork so I could run CI and it seems to be failing for the same reason as the Selenium v4 macOS pipeline. It is pulling a newer version of elixir and there are deps that are incompatible with elixir 1.14+ |
.github/workflows/ci.yml
Outdated
| elixir-otp: | ||
| - otp: 25.x | ||
| elixir: 1.14.x | ||
| - otp: 26.x | ||
| elixir: 1.15.x |
There was a problem hiding this comment.
why did you remove 1.12 and 1.13?
There was a problem hiding this comment.
Because they are not backwards compatible with the changed dependencies.
There was a problem hiding this comment.
This will likely require a major version bump to signify this change.
There was a problem hiding this comment.
I don't want to drop support for older versions of Elixir just for a dev dependency that I have never even used since taking over the project.
you can probably add --only test to mix deps.get in CI to avoid having to update benchee.
There was a problem hiding this comment.
I am down to drop benchee as a dependency altogether. However there are other changes around ExUnit.Server.modules_loaded() that are not backwards compatible.
There was a problem hiding this comment.
that is a private elixir API, and is in a unit test.
I would just add a conditional that checks if the function is exported and call the one that is
function_exported?(ExUnit.Server, :modules_loaded, 0), etc
| @@ -73,7 +73,7 @@ defmodule Wallaby.HTTPClientTest do | |||
|
|
|||
| test "includes the original HTTPoison error when there is one", %{bypass: bypass} do | |||
| expected_message = | |||
| "Wallaby had an internal issue with HTTPoison:\n%HTTPoison.Error{id: nil, reason: :econnrefused}" | |||
| "Wallaby had an internal issue with HTTPoison:\n%HTTPoison.Error{reason: :econnrefused, id: nil}" | |||
There was a problem hiding this comment.
if this is flaky, i would do multiple =~ asserts instead
There was a problem hiding this comment.
I'm not confident that it is flaky just yet. I want to understand why this change was necessary for this pass.
There was a problem hiding this comment.
otp changed how maps work for keys < 32 so now key order is not deterministic like it used to be
There was a problem hiding this comment.
Agreed. But if maps < 32 keys are treated like keyword lists in memory, they are linked and therefore have a deterministic order yes? So my inference would be that {:id, nil} was added before {:reason, :econnrefused} in the internals of HTTPoison, thereby putting :reason at the head of the linked list.
There was a problem hiding this comment.
I'm trying to say that in OTP26 that is no longer the case. the order is nondeterministic
https://www.erlang.org/blog/otp-26-highlights/#changed-ordering-of-atom-keys
You can pass custom_options: [sort_maps: true] to as an Inspect.Opt tho, which will sort it and it should be deterministic.
There was a problem hiding this comment.
Ah. I see! Thanks for explaining 😃 I'll make some more adjustments and see if I can get things backwards compatible as well. Thanks for the feedback!
kyleboe
changed the title
No description provided.