EXIT traps that print to stdout shouldn't break process substitution

[onlynone]

OS X sets a trap on EXIT to call shell_session_update, this calls shell_session_save which will print to stdout. Normally this is not a problem for subshells instantiated like $(...), because the traps are apparently not inherited by (at least these types of) subshells. However, rvm re-installs this trap, (even from within subshells) when you call something like cd (which rvm overrides).

Basically:

trap 'echo trapped' EXIT
a=$(echo a)

Will not have the EXIT trap called inside the $(...), and will result in a=a. However:

a=$(trap 'echo trapped' EXIT; echo a)

Will have the trap called, and will result in a=a trapped.

The latter is what rvm ends up doing when it's loaded by its version of cd, and you do something like:

a=$(cd somewhere && pwd)

Perhaps the overridden shell commands should never install the exit trap? Or rvm could detect when it's being run inside a subshell where it's helpers aren't needed and not run any of the init code?

[onlynone]

This test failed, only because there's a bug with rvm.