When asked to enable libedit in a system that ALSO has readline,
the headers of the former would be found and used by the earlier.
While that would mostly work, some functions will be missing
definitions (which is forbidden in C99), so instead abort the
configuration and let the user provide for them.
Using cmake to configure and enable linking pcre2test with libedit,
could result in a broken build, because the header used was instead
pointing to readline.
In cases were the build will succeed (because both libraries were
available), it would likely show warnings, because several history
functions were being used without declarations, since readline
requires including "history.h" for those.
Additionally, since PCRE2_SUPPORT_READLINE is ON by default (unlike
configure), turning PCRE2_SUPPORT_LIBEDIT=ON, would require setting
that other option to OFF explicitly (even if readline wasn't available)
or the setup would abort.
Lastly, in systems with no default sysroot (ex: macOS), the use of
absolute paths for searching for libedit's readline.h could fail so
use instead relative PATH_SUFFIXES.
When `./configure --enable-pcre2test-libedit` is used in FreeBSD,
the resulting test will succeed but won't set the necessary flag
to distinguish between libedit and readline header files, therefore
using readline's at built time (if installed)
Consolidate all header tests into one and use instead the corresponding
autogenerated defines to check for each possibility.
eb42305f (jit: avoid integer wraparound in stack size definition (#42),
2021-11-19) introduces a check to avoid an integer overflow when
allocating stack size for JIT.
Unfortunately the maximum value was using PCRE2_SIZE_MAX, eventhough
the variable is of type size_t, so correct it.
Practically; the issue shouldn't affect the most common configurations
where both values are the same, and it will be unlikely that there would
be a configuration where PCRE2_SIZE_MAX > SIZE_MAX, hence the mistake
is unlikely to have reintroduced the original bug and this change should
be therefore mostly equivalent.
Signed-off-by: Carlo Marcelo Arenas Belón <carenas@gmail.com>
On CHERI, and thus Arm's Morello prototype, pointers are represented as
hardware capabilities, which consist of both an integer address and
additional metadata, meaning they are twice the size of the platform's
size_t type, i.e. 16 bytes on a 64-bit system. The ovector member of
heapframe happens to only be 8 byte aligned, and so computing frame_size
ends up with a multiple of 8 but not 16. Whilst the first frame is
always suitably aligned, this then misaligns the frame that follows it,
resulting in an alignment fault when storing a pointer to Fecode at the
start of match.
Thus, round up frame_size to a multiple of heapframe's alignment to
ensure alignment is preserved. This can be completely optimised away on
traditional architectures and, since CHERI's capabilities are in fact
2 * sizeof(PCRE2_SIZE) bytes in size, the variable part of the
expression is also proven to be a multiple of the alignment and so the
aligning gets folded into the offsetof part by adding an additional 8,
so no dynamic alignment code is needed even on CHERI architectures.
Notably, running the script directly from a build subdirectory will
infer srcdir as .. if not otherwise set, but doesn't work for these.
With this commit sh pcre2_grep_test.sh works as expected.
Carlo Marcelo Arenas Belón
Zoltan Herczeg
Philip Hazel
Adam
Zoltan Herczeg
Jessica Clarke
Michael Kaufmann