diff options
Diffstat (limited to 'winsup/cygwin/DevNotes')
| -rw-r--r-- | winsup/cygwin/DevNotes | 26 |
1 files changed, 26 insertions, 0 deletions
diff --git a/winsup/cygwin/DevNotes b/winsup/cygwin/DevNotes index 0e86fe516..aeca33076 100644 --- a/winsup/cygwin/DevNotes +++ b/winsup/cygwin/DevNotes @@ -1,3 +1,29 @@ +2012-08-17 cgf-000016 + +While debugging another problem I finally noticed that +sigpacket::process was unconditionally calling tls->set_siginfo prior to +calling setup_handler even though setup_handler could fail. In the +event of two successive signals, that would cause the second signal's +info to overwrite the first even though the signal handler for the first +would eventually be called. Doh. + +Fixing this required passing the sigpacket si field into setup_handler. +Making setup_handler part of the sigpacket class seemed to make a lot of +sense so that's what I did. Then I passed the si element into +interrupt_setup so that the infodata structure could be filled out prior +to arming the signal. + +The other changes checked in here eliminate the ResetEvent for +signal_arrived since previous changes to cygwait should handle the +case of spurious signal_arrived detection. Since signal_arrived is +not a manual-reset event, we really should just let the appropriate +WFMO handle it. Otherwise, there is a race where a signal comes in +a "split second" after WFMO responds to some other event. Resetting +the signal_arrived would cause any subsequent WFMO to never be +triggered. My current theory is that this is what is causing: + +http://cygwin.com/ml/cygwin/2012-08/msg00310.html + 2012-08-15 cgf-000015 RIP cancelable_wait. Yay. |