This should be enough to fix the main bug as reported in #314.
Other cross-compile limitations may still exist.
Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
This reverts commit c34590c4bf067d4e749e5298a8ce9f90072d6332
and all search code patches that followed it.
The execvp() would do the $PATH searching of m4 for flex. There's no
need to duplicate the effort.
Another reason for the removal is to remove dependency on stat()
system call, which would cause a problem in a 32-bit OS accessing a
filesystem with 64-bit inode number (see #413).
Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
A build test is done to determine the minimum versions of the configure
tools (Autoconf, Automake, Libtool & gettext) that can successfully
build flex in the current state of the code. This commit would help
builders to figure out whether their build tools are still compatible
for building flex, or they need tools upgrade, especially when they
patch flex and need to regenerate configure script for any reason.
This commit resolves version requirement debates as in #374.
A summary of version requirements: Autoconf 2.62, Libtool 2.2,
Automake 1.11.3 and gettext 0.19.6.
Signed-off-by: Kang-che Sung <explorer09@gmail.com>
Due to a bug, autoheader (2.69) will treat M4 dnl comments in a quoted
argument of AC_CHECK_FUNCS as function tokens and generate a lot of
redundant and useless HAVE_* macros in config.h.in.
(Examples: HAVE_DNL, HAVE_AVAILABLE_, HAVE_BY)
It seems to be this commit dbb4e94dc7bacbcfd4acef4f085ef752fe1aa03f of
mine that revealed this autoheader bug, and the affected config.h.in
had been shipped within flex-2.6.4 release tarball.
I have reported the autoheader bug here:
<https://lists.gnu.org/archive/html/bug-autoconf/2018-02/msg00005.html>
As a workaround, let's move comments out of AC_CHECK_FUNCS.
Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
This would, e.g. define _GNU_SOURCE in config.h, enabling the
reallocarray() prototype in glibc 2.26+ on Linux systems with that
version of glibc.
Fixes#241.
NetBSD had a crash during build. Since the provided substitute for
reallocarray() wasn't working, use NetBSD's reallocarr(). Let
configure choose that function whenever it is available. Use
reallocarray if available. Still fallback if neither is available.
Fixes#219
Files in lib/ are picked up and built using makefile directives in
src/Makefile.am. Remove the need to generate lib/Makefile.in and the
stub lib/Makefile.am.
Check for cross compiling. If cross compiling, build stage1flex using a
custom link command.
We also override LDADD since that adds the replacement implementations
that are cross compiled, and instead always use the replacement library
implementations.
We don't use BUILD_OBJEXT and BUILD_EXEEXT since it seems that automake
does not support these.
Fixes#78.
strings.h (not string.h) is the standard-defined include header for
strcasecmp(). Include <strings.h> for portability (even though glibc
exposes strcasecmp() declaration also in <string.h> by default).
`configure` will now error if a required header or function is not
found on the system.
Also add comments on optional functions checks.
Add sys/stats.h and sys/wait.h to list of required headers in
configure.
This fixes issue #180.
"Unlisting" the pthread test when libpthread is not available is never
a good idea. Should let it compile into a stub program that returns
the "skip" status.
configure.ac and tests/Makefile.am are edited so that pthread test can
be built anyway, but only works (not as the stub) when both pthread.h
and libpthread are detected on the system.
Also fix a small typo in tests/pthread.l comments.
AC_PROG_YACC already detects the bison program. We can just use it,
and make notice if [ $YACC != 'bison -y' ]
Remove unused AC_SUBST([BISON], ...).
Add AM_CONDITIONAL([HAVE_BISON], ...) so that we can skip bison tests
in testsuite when bison is absent.
We don't make any attempt to detect other yacc implementations since
flex is developed with bison. While it would be possible, the need
isn't there right now. It would be possible to add a --with-parser
option that defaulted to bison if such a need cropped up.
Use reallocarray() when we have it (i.e. in OpenBSD system). When we
don't, use equivalent overflow detection for our allocate_array and
reallocate_array functions.
Remove lib/reallocarray.c from our LIBOBJS as we no longer need it.
Provide a fallback SIZE_MAX macro definition in flexint.h (not
preprocessor friendly, but enough for our reallocate_array use case).
If configure is run with '--disable-bootstrap', then stage1flex won't
be built and stage1scan.c will be generated by sed'ing scan.c.
This option is intended to workaround bootstrap bugs rather than to
fix the bootstrapping issues which are proving subtler and harder to
fix than anyone would like.
For various reasons, we may wish to build libfl explicitly even when
configure has been run with the --disable-libfl option. This is
possible, now, via 'make -C src libfl.la'.
The libcompat.la library was small and less friendly to bootstrapping
and cross compilation. Now, we will simply link individual object
files as needed, which is simpler.
HELP2MAN and TEXI2DVI (or the corresponding ac_prog variables) will
never be zero length as they fall back to the missing script; check for
the fall back and warn on that instead of always warning.
`WARNINGFLAGS' can be passed when invoking `configure'
and when invoking `make'
if configure switch `--enable-warnings' was given
then default to something useful if we have `GCC'
`WARNINGFLAGS' is not used when compiling `stage1flex'
to avoid unnecessary clutter
The [] characters are used for quoting in m4, so the attempt to use
them in place of `test` fails yielding the warning at build time:
.../flex/configure: line 20222: no: command not found
This is taken from OpenSSH Portable, which in turn takes it from
OpenBSD.
reallocarray wraps the stdlib's realloc function. It takes two size
arguments and checks for overflow, like calloc, but doesn't zero the
memory. Therefore, it allows us to do overflow-safe array reallocations
and overflow-safe unzeroed array allocations, which the stdlib
allocation functions don't.
We have a bunch of specific array allocation macros, none of
which check for overflow. reallocarray should be able to replace them.
