A bug going into production without any need:
warning: undefined module attribute @seconds, please remove access to @seconds or explicitly set it before access
lib/api_baas/accounts_admins/admin_user_token.ex:120: ApiBaas.AccountsAdmins.AdminUserToken (module)
Yes, tests should caught it, but why doesn’t the compiler abort compilation when it finds this?
Yes, I know that warnings can be enabled to behave as errors, but some warnings need to behave as warnings, and not halt the compiler.
I am curious about the reason for this to be considered as an warning and not an error?
Elixir’s code states that it is a known issue and they just don’t want to add breaking change
# TODO: Consider raising instead of warning on v2.0 as it usually cascades
Why only in
v2.0, that may never come!?
This a breaking change to solve a serious issue that lets bugs crop into production and it’s a compiler breaking change not a runtime one.
--warnings-as-errors which is a sane option for CI
I’m curious about what these are, because I can’t remember the last project I worked on that didn’t build with
--warnings-as-errors in CI.
Compiling 1 file (.erl)
Compiling 41 files (.ex)
warning: System.stacktrace/0 is deprecated. Use __STACKTRACE__ instead
Compiling 4 files (.ex)
warning: function query_many/4 required by behaviour Ecto.Adapters.SQL.Connection is not implemented (in module Ecto.Adapters.SQLite3.Connection)
lib/ecto/adapters/sqlite3/connection.ex:1: Ecto.Adapters.SQLite3.Connection (module)
I only have 1.13 + OTP 25 right at hand to test, but on that version I still get the same warning from
plug when running
mix compile --warnings-as-errors with an empty
--warnings-as-errors doesn’t turn warnings in dependencies into errors.
Some historical digging reveals that it used to; around the time of Elixir 1.4 that was changed, by commits like:
Well IMO this is a reasonable request. An opt-in configuration to allow compilation failure if any warning is in the project’s or any of the dependencies’ code should do the trick. There are orgs that inspect and slightly modify their project deps.
I really don’t understand the rational of allowing this breaking change and not allowing the breaking change that would really fix the issue and make Elixir compilation to deliver a more robust code, instead of buggy code as it does at the moment.
The rationale of postponing for v2.0 really puzzles me, because that version may never exist as Jose keeps saying, but even if it was scheduled it doesn’t make any sense in my mind to wait, because this is a compiler bug that should be fixed, because a compiler that produces bad code output isn’t helping anyone.
I think you should ask someone from the core team about this.
I don’t think it necessarily produces bad code. It’s kind of increasing the risk of having bad code though, that’s true.