[Libevent-users] "make verify" failure in regress.c:test_methods()

[Scott Lamb]

"make verify" does not work for me under Linux (CentOS 5) with trunk 
HEAD (revision 836). It reaches the epoll section and says this:

EPOLL
  test-eof: OKAY
  test-weof: OKAY
  test-time: OKAY
  regress: Testing supported methods: epoll poll select [err] 
event_base_new_with_config: no event mechanism available
FAILED

It looks like regress.c:test_methods() is trying to disable the first 
method and instantiate a config, which is failing because the test.sh 
wrapper has set environmental variables which cause evsel->init() to 
fail on all other methods.

As for a fix...maybe the environmental variables should just be used to 
set up a default config to be used if none is specified, and possibly to 
be copied on event_config_new(). (If the latter, test_methods() would 
need to call event_config_is_avoided() before adding a method to its 
count, and this test would always be skipped on "make verify".) It seems 
odd for EVENT_NO* and event_config_avoid_method() to work by totally 
different mechanisms.

This must have been broken since the test was added on May 8th. I'm 
trying to understand how that happened - is "make verify" not what the 
cool kids use to test libevent? or do the cool kids not use Linux?

Best regards,
Scott

-- 
Scott Lamb <http://www.slamb.org/>