I ran across a problem today in which some code ran fine in regular
operation but failed in a test case. I scratched my head and thought
“why
would running from within the test harness change the behavior of my
code?”
Clearly it was the Heisenberg uncertainty principle in action! 
I discovered the root cause was that the rspec runner is setting the
magical
ruby global, $KCODE, to ‘u’. However, my application (which is not a
rails
app) had never specified $KCODE. I was relying on some default behavior
of
regular expressions under ruby 1.8 (byte-wise character matching
semantics)
that change when $KCODE is set to “u”. Specifically, all regular
expressions change their default behavior to utf-8 character-wise
semantics.
A simple example of this phenomenon: kcode_gotcha.rb · GitHub
I’ve corrected the issue on my end by adopting the $KCODE=‘u’ semantics
in
my application, but this led me to a couple comments/questions I thought
would be relevant to raise with other rspec-minded folks:
-
Is it necessary for rspec to set $KCODE or is this a bug? Wouldn’t it
be
better if it didn’t twiddle any magical globals that change
interpreter-wide
behaviors? It reminds me of the bad days of perl when some distant code
would unexpectedly change out your line terminator character on you. -
It seems like a good idea to call out all the things the rspec
environment
changes that affect natural runtime behavior of code: twiddling of
globals,
class monkey patches, etc. It’d be great to get these into a list of
publicly documented pitfalls for people to watch out for.
Thanks for listening. Rspec has brought huge value to my engineering
project
and I really appreciate the tool!
-Josh