fix race w/ *Conn.requestTimeout #112
Conversation
conn.go
Outdated
| l.requestTimeout = timeout | ||
| l.timeoutMutex.Unlock() |
There was a problem hiding this comment.
Might be simpler to make this an atomic value and do lockless load/store. We already have a shim for < go1.3 using locks
|
using the atomicValue struct triggered the race during travis build... |
that's concerning... looks like the 1.3 version needs to hold a pointer to the lock |
|
1.3 was ok, it hit 1.5: https://travis-ci.org/go-ldap/ldap/builds/229842330, the 1.3 failure was a network fetch error while setting up: https://travis-ci.org/go-ldap/ldap/builds/229846728 |
|
hmm, that's still concerning, but the atomic load/store int64 is nice and clean and sidesteps that issue. go ahead and squash and remove the unused mutex, then LGTM |
this fixes the race seen e.g. in https://travis-ci.org/go-ldap/ldap/jobs/227891683
|
hmm... https://travis-ci.org/go-ldap/ldap/jobs/230628746 ... what does the race detector find? concurrent r/w access to the *Conn struct's members? |
|
fixed by #134 |
see https://travis-ci.org/go-ldap/ldap/jobs/227891683
cherry picked from #110