I do not completely understand what is going on there, but I am pretty
sure dir_entry <= offset_base if not a normal situation, so we better not
to rely on such dir_entry.
`zend_string_tolower()` returns a copy (not a duplicate) of the given
string, if it is already in lower case. In this case we must not not
`zend_string_free()` both strings. The cleanest solution is to call
` zend_string_release()` on both strings, which properly handles the
refcount.
Firstly, we must not call `gdImageSetAntiAliased()` (which sets the
color to anti-alias), but rather modify the `gdImage.AA` flag.
Furthermore, we have to actually use the supplied boolean value.
We also make sure that we don't attempt to enable anti-aliasing for
palette images.
These tests are obviously meant to test successful and failing uri:
DSNs, but did not pass proper file:// URIs, so actually ended up
testing for invalid data source URIs twice. We fix this, and adjust
the expectations accordingly.
We also unfork the -win32 variant, since both test cases are almost
identical, and the expected error message may be either one.
MySQL/MariaDB version strings may have suffixes which may contain dots;
for instance, Debian stretch has 5.5.5-10.1.37-MariaDB-0+deb9u1 or
such. Therefore, we make the version extraction more liberal, and only
require that there are at least three parts separated by dot, and
ignore additional parts.
We also fix an erroneous test expectation, which would be triggered on
CI now, right away. This patch has been provided by petk@.
When actually fetching the data, bigint (unsigned) column values are
returned as integers on LLP64 architectures, so their pdo_type has to
be PDO::PARAM_INT accordingly.
When a connection is closed, we also need to remove the hash entry
from the regular_list, as it now points to freed memory. To do this
store a reverse mapping from the connection to the hash string.
It would be nicer to introduce a wrapping structure for the pgsql
link resource that could store the hash (and notices), but that would
require large changes to the extension, so I'm going for a more
minimal fix here.
Optimizing compilers have an annoying tendency to throw out
memsets over data that they think aren't used anymore. Apply secure
zero-out in cases where this has potential to happen.
