The reason was that after the big refactoring of mysqlnd at the end of
last year code that is initializing the error_info structure in the
result set was not added. It existed already for connections and PS.
The code that segfaults is hit only with MariaDB because MariaDB sends
full metadata about the EXPLAIN query + EOF packet and only then it sends
an error packet. MySQL doesn't do that but sends directly an error which
is caught (by different code path). As errors during execution (which means
after sending meta) are pretty rare there was no test case of MySQL to
catch it.
If we want to fetch into an object of a custom class that implemens
__set handler, the corstructor has to be called first. The data
passed to the constructor can be possibly required in __set handler.
Added the possibility to explicitly state that the peer certificate should not be checked.
Back to the default - checking the certificate.
Exported MYSQLI_CLIENT_SSL_DONT_VERIFY_SERVER_CERT
Usage : mysqli_real_connect( , , , , , MYSQLI_CLIENT_SSL | MYSQLI_CLIENT_SSL_DONT_VERIFY_SERVER_CERT)
If mysqli_ssl_set() is not called, but only MYSQLI_CLIENT_SSL is passed, without the (don't) very flag,
then no verification takes place.
MySQL and MariaDB of different versions have different assumptions about TIMESTAMP attributes.
Probably we need to add yet another condition to also check MariaDB version.
This implements a reduced variant of #1226 with just the following
change:
-Fatal error: Uncaught exception 'EngineException' with message 'Call to private method foo::bar() from context ''' in %s:%d
+Fatal error: Uncaught EngineException: Call to private method foo::bar() from context '' in %s:%d
The '' wrapper around messages is very weird if the exception
message itself contains ''. Futhermore having the message wrapped
in '' doesn't work for the "and defined" suffix of
TypeExceptions.
TypeException stays as-is for now because it uses messages that are
incompatible with the way exception messages are displayed.
closure_038.phpt and a few others now show that we're generating
too many exceptions for compound operations on undefined properties
-- this needs to be fixed in a followup.
