Comment 29 for bug 319480

(In reply to comment #16)
> however, we also had other problems because of environ getting trashed after
> unloading libxul.so (namely, the libatk bridge made things crash when it
> tried to access environ in exit_hook); so leaking is probably still needed.

Yes, sounds like leaking would still be needed.
Here too:

http://hg.mozilla.org/mozilla-central/file/0471fa063e43/accessible/src/atk/nsAppRootAccessible.cpp#l567

I'm not a toolkit peer, but I think it would be better to change only the
instances that are not already leaked to use a function name that is not
PR_SetEnv. SaveWordToEnv() is probably fine for this.

Maybe the attempts to clear environment values could check that they are set
before needing to leak:

http://hg.mozilla.org/mozilla-central/file/f46e6aee1335/toolkit/xre/nsAppRunner.cpp#l3303