This is very ugly: Bison provides a yynerrs variable, which is
usually not actually used, but also not annotated with
YY_MAYBE_UNUSED. Suppress this warning by adding a (void)yynerrs
into the top-level reduction action. The alternative would be to
disable the warning for these generated files.
Generalises tests for all the different drivers.
Run the general test for the various drivers.
This allows support for parallel testing as the tests now do not rely on the same DB file.
Despite commit 69c3f8c[1] claiming otherwise, there is no need for any
Windows specific implementation here. Users can use random_bytes(), if
they desire so.
[1] <69c3f8cbba>
Closes GH-9153.
This fixes an issue where a namespaced class beginning with "U" or "u"
would yield an invalid arginfo file due to the occurrence of a unicode
escape sequence, causing a compile error.
Co-authored-by: Guilliam Xavier <guilliamxavier@users.noreply.github.com>
Closes GH-9154.
Having a stack allocated zend_function may cause crashes if the stack is polluted between bailout and the actual unwinding in zend_observer_fcall_end_all.
Signed-off-by: Bob Weinand <bobwei9@hotmail.com>
This PR changes the glob stream wrapper so it impacts "glob://"
streamsas well. The idea is to do a check for each found path instead
of the pattern which was not working correctly.
The arguments 3 to 6 of the authorizer callback may be `NULL`[1], and
we have to properly deal with that. Instead of causing a segfault, we
deny authorization, which is still better than a crash, and apparently,
we cannot do better anyway.
[1] <https://www.sqlite.org/c3ref/set_authorizer.html>
Closes GH-9040.
One may argue that `DROP_NEW_LINE` does not make sense in combination
with `READ_CSV`, but without `DROP_NEW_LINE`, `SKIP_EMPTY` does not
skip empty lines at all. We could fix that, but do not for BC reasons.
Instead we no longer drop newlines in `spl_filesystem_file_read_ex()`
when reading CSV, but handle that in `spl_filesystem_file_read_csv()`
by treating lines with only (CR)LF as being empty as well.
Closes GH-7618.
This has only been done for Windows systems so far, and there was a
TODO comment about looping for larger values; that appears to be
overkill, though, since 2 million bytes should be sufficient for all
use cases, and if there is really the need for more, users can still
loop manually. Anyhow, checking the range upfront on all platforms
is clearer then silently casting to `int`.
We split the error message for the least possible BC break.
Closes GH-9126.
