1. On OpenBSD, and check is made for only bison or byacc. yacc is
present, but not usable apparently. So - need to install bison on the
BSD's.
2. the libjpeg.h check fails as it is located in a non-standard
location (/usr/local) on BSD systems. Just remove the check for now
until we have a better way to check that stuff (like X11 and Xm
headers too).
A list of major things like ksh, cpp, etc are saved in a list if they
are not found. If this list is non-empty when configure is nearly
done, an error message is displayed listing the missing programs.
This is less annoying than stopping after every missing programs.
Also, removed the X11/Xm header checks for now. Those need to take
into account X_CFLAGS in some way since these files are located in
dirfferent areas on different OS's (obsd puts them in
/usr/X11R6/include for example).
... and move the PROG_CC and PROG_CXX checks toward the beginning.
Without the move the compiler search arguments are ignored. We search
for 'cc gcc clang' and 'c++ g++ clang++', in that order for CC and CXX
respectively.
This means there should be no need to override the CC
and CXX variables on the configure command line for BSD systems.
On OpenBSD for example, cc links to clang. On Linux, cc links to
gcc. You can still override these if you like with CC=... and
CXX=... on the configure command line.
gmake (MAKE=gmake) still needs to be used on the BSD's though.
With these and previous changes, CDE builds, installs, and runs
correctly on OpenBSD 6.5 now.
Certain programs like dtlogin and tt_type_comp need access to the
system's cpp command to process certain files at runtime. tradcpp is
not installed on most systems, and is intended for use as a imake-cpp
replacement which is all we are using it for. We still need an
honest cpp for the CDE components to use.
Now, we look in various places for a 'cpp' command and set CPP_COMMAND
to it's value. This way tt_type_comp and other CDE programs can do
required processing at runtime.
There are still some issues here - specifically with the C
dtbuilder/dtcodegen catalogs. They may be broken due to dtbuilder not
being built yet.
Also, once dtinfo can be built, the msgs and app-defaults for it
should be re-enabled.
I also see gencat coredumping in localization/ while doing a multicore
(-jX) make. Regular single core builds seem to work fine.
Make -lXinerama a dep on libDtXinerama so it doen't need to be
specified in Makefile.am files or in LIBS as it was being done
previously.
This still needs a little work, ideally Xinerama should be completely
optional and only enabled if present. But we can save that for
later.
Previously we would detect whether TIRPC is installed or not, and set
TIRPCINC to the include directory with the -DOPT_TIRPC macro defined.
Then, pretty much every Makefile.am needs to be sure that $(TIRPCINC) is
specified in the cpp/c/cxx flags.
Since we can never be sure that an RPC header file might be indirectly
included, a better approach is to simply add TIRPCINC to the global
list of CXXFLAGS and CFLAGS in configure.ac for everybody. This way,
it is always specified properly on tirpc systems, and we don't need to
always add it to every individual Makefile.am since everyone will get
it by default.
TIRPCLIB is still marked as a dep in libtt, so as long as libtt is
linked, you should automatically get the tirpc library too. This is
still unchanged.
For ksh, we need a full pathname. AC_CHECK_PROGS only sets the name,
so we can't use that (think of a "#!" in a shell script.
We use some shell scripting to locate the ksh pathname.
While on that subject, the current use of CPP (gcc -E) as a general
preprocessor does not work very well. I messes up whitespace,
adds/translates random whitespace, and complains bitterly about single
quotes (') in various places like comments. It's not usable for what
CDE needs.
So, now we use GENCPP. Using shell scripting like that used for ksh,
we locate the cpp program, and set GENCPP to "/full/path/to/cpp
-traditional -nostdinc". This is what Linux uses now in an Imake
build, and it works fine. We'll have to see what the BSD/Solari do.
We might need to just include BSD's "tradcpp" into the build and use
that. It too works well in limited testing, but eats blank lines. We
can live with that if we have to.
