Re: BUG #16685: The ecpg thread/descriptor test fails sometimes on Windows

BTW, I searched the buildfarm logs of ECPG failures to see if I could
find anything related to this, and noticed that there were a fair
number of Windows-machine failures in the thread/prep test. They don't
all look quite alike, but this one is representative (and the most
recent):

sysname|dory
snapshot|2019-04-02 11:20:25
================== pgsql.build/src/interfaces/ecpg/test/regression.diffs ===================
diff -w -U3 c:/pgbuildfarm/pgbuildroot/HEAD/pgsql.build/src/interfaces/ecpg/test/expected/thread-prep.stderr c:/pgbuildfarm/pgbuildroot/HEAD/pgsql.build/src/interfaces/ecpg/test/results/thread-prep.stderr
--- c:/pgbuildfarm/pgbuildroot/HEAD/pgsql.build/src/interfaces/ecpg/test/expected/thread-prep.stderr 2018-04-20 16:03:08 -0400
+++ c:/pgbuildfarm/pgbuildroot/HEAD/pgsql.build/src/interfaces/ecpg/test/results/thread-prep.stderr 2019-04-02 07:43:27 -0400
@@ -0,0 +1,84 @@
+SQL error: could not connect to database "ecpg1_regression" on line 48
+SQL error: the connection to the server was lost on line 53
+SQL error: the connection to the server was lost on line 52
+SQL error: invalid statement name "i" on line 53
+SQL error: the connection to the server was lost on line 52
+SQL error: invalid statement name "i" on line 53
+SQL error: the connection to the server was lost on line 52
+SQL error: invalid statement name "i" on line 53
+SQL error: the connection to the server was lost on line 52
+SQL error: invalid statement name "i" on line 53
... quite a few repetitions of this ...
+SQL error: the connection to the server was lost on line 52
+SQL error: invalid statement name "i" on line 53
+SQL error: invalid statement name "i" on line 55
+SQL error: connection "Connection: 1" does not exist on line 56

I suspect that this might arise from a pthread_once failure
associated with the actual_connection_key variable.

regards, tom lane