A had a bit of a vexing experience with Elixir the other day that I thought I’d document here. This was in the process of writing a Braid email bot that lets people view & manage their Gmail inbox in Braid.
I had noticed a few days ago that non-ASCII codepoints were being displayed improperly in the messages that were sent by the email bot. I was able to see that the bot was printing out the correctly-encoded strings to the iex repl before sending the message to Braid, so I assumed the error was on Braid’s end. This let to a great deal of confusion, as the problem only seemed to manifest when receiving the request: I could copy-paste the bytes that the Elixir-powered bot retrieved into a Clojure REPL and see that it decoded to the expected MessagePack-encoded transit, so it seemed like something strange was happening in the Ring handler. However, after lots of banging my head against the wall, I had no idea what that strange thing could be.
Doing some pair-programming with my co-founder Raf however, got me to look more closely at the bot.
While I could print out the information as retrieved from the Gmail API properly, when I tried to print out the message right before MessagePack-ing and sending, it was garbled.
It transpires that I was using Erlang’s
:io_lib.format when building the message to send and, as the Elixir documentation very helpfully says
Also note that Erlang’s formatting functions require special attention to Unicode handling.
By searching for the Erlang io_lib documentation, I was able to eventually figure out that the solution was to use the
~ts format specifier, instead of
I’m not entirely sure what the
t stands for – “don’T break things”, I suppose.
Kinda annoyed that the Elixir linters I tried were screaming about the most minor formatting thing, but said nothing about how using this format specifier or
:io_lib would seem to work until they were fed non-ASCII data.
Summary: If you’re going to use
:io_lib.format/2 in your Elixir code, you probably want to use
~s, or it will break when formatting non-ASCII codepoints.