Why does String.to_integer/1 give strange results with with some values?

congrats on reaching this milestone - and completing a major step/rite of passage on your elixir journey - it’s a charlist Binaries, strings, and charlists — Elixir v1.21.0-dev

we have all been there - and gone through the same wtf etc.

Try this to force lists of integers to display as lists of integers (reference: Inspect.Opts — Elixir v1.12.3):

["41", "92", "73", "84", "69"] 
|> Enum.map(&(String.to_integer(&1)))
|> IO.inspect(charlists: :as_lists)

That this is not the default is a little crazy to me.

It is a tribute to the Erlang community

Yeah, you are basically seeing the runtime trying to be a bit too helpful – if a list of integers looks like a list of printable characters, then the REPL will instead show them as a printable string. The data is still there and is still the same.

If you want to get rid of that behavior in your project, just create an .iex.exs file at the root of your project and put this in there:

IEx.configure(inspect: [charlists: :as_lists])

Restart your iex session and, et voila:

iex(1)> ["41"] |> Enum.map(&String.to_integer/1)
[41]

The problem I run into is that it does not apply to any IO.inspect calls within the code. You have to remember to do IO.inspect(number_list_that_isnt_supposed_to_be_chars, charlists: :as_lists) every time you might be faced with this weirdness.

I think it’s time to make charlists: :as_lists to be a default option, and deprecate the charlists: :infer option.

I think that would just surface how often we actually run into charlists when interacting with erlang libraries. E.g. hardly any elixir app doesn‘t use gen_tcp somewhere.

From my personal experience I‘ve been hitting the case of needing to see the underlying list of integers mostly in tests or puzzles, where there is time and opportunity to deal with the printed formatting. Not optimal, but fine.

Interacting with actual charlists on the other hand comes up when you‘re debugging some production issue and trying to dig deeper into some failure, where you don‘t want to deal with first setting up useful formatting rules to be able to understand what the system is trying to provide in information.

Hehe
I was right here

At least now there is the chance that someone can look up the documentation for the ~c sigil. Just because it didn’t happen in this case it does not mean it’s not an improvement.