* harbour/config/dir.mk
% small simplification in used logic
* harbour/config/dirsh.mk
! use := instead of = for DIR_RULE assignment - these rules contain
dependencies macros which cannot be resolved during assignment
* config/global.mk
* Upped minimum GNU Make version requirement to 3.78 (from 3.70).
We use many 3.78-only features since very long and in many
Makefiles (like $(if)). 3.78 was released in 1999, the year
Harbour has born, so this requirement looks pretty modest.
Notice however that the "officially" documented (in INSTALL)
minimum version is 3.81, and there are a few feature we
actually use and need from this version. Some other make features
are only enabled when using 3.80 (or upper) (parallel build),
some others when using 3.81 (or upper) (build configuration
information, verbose output).
; TODO: Either up the minimum version, or try to fix the remaining
unprotected 3.80/3.81 features.
My personal opininon is that if there is any reasonable
chance we should make 3.80 or 3.81 the minimum official
requirement.
This is a *nix-only issue, as dos/win/os2 are all covered
with recent 3.81 binary builds. Question is:
How and which *nix platforms would be affected or limited
in such this case?
BTW, 3.81 was released in 2006, 3.80 in 2002.
* ChangeLog
* Updated cross-build compatibility chart.
* config/global.mk
+ Added autodetection of cegcc tools on default install location.
+ Added error messages if specified or autodetected mingw/cegcc
cross-tools couldn't be found.
! Minor cosmetic in warning/error msgs. (ending dot seems to be added
by GNU Make)
; NOTE: Currently possible cross-build scenarios (except cross-CPU only cases,
by host platform:
on win hosts:
-> wce/mingawrm
-> wce/poccarm
-> wce/msvcarm
-> dos/djgpp
-> dos/watcom (* using DOS GNU make)
-> os2/watcom
-> linux/watcom
on dos hosts:
-> win/watcom (*)
-> os2/watcom (*)
-> linux/watcom (*)
on os2 hosts:
-> win/watcom
-> dos/watcom (*)
-> linux/watcom (*)
on linux hosts:
-> wce/mingwarm
-> win/mingw
-> win/watcom
-> dos/watcom (*)
-> os2/watcom
on darwin/bsd/hpux/sunos hosts:
-> win/mingw (*)
(*): Theoretically possible, but not heard of such build yet.
* contrib/rddads/adsfunc.c
+ Added TOFIX to a very old typo in ADSDDREMOVETABLE() which
makes it require a dummy variable to be passed as 3rd parameter.
* config/wce/poccarm.mk
* config/wce/msvcarm.mk
* config/win/xcc.mk
* config/win/pocc.mk
* config/win/bcc.mk
* config/win/watcom.mk
* config/win/icc.mk
* config/win/msvc.mk
* config/rules.mk
+ Made dynamic compilation pass a little bit more generic.
Now HB_DYN_COPT variable should be set to the switch(es) which
are used in this pass. This will automatically enable the
dynamic compilation pass. F.e. for *nix/gcc compilers it can
be set to -fPIC.
* config/global.mk
+ Implemented win/mingw cross-tool autodetection.
Based on the logic found in make_gnu_xmingwce.sh. It's not
exactly the same, because I used HB_CCPATH and HB_CCPREFIX
only. There is no fiddling with the PATH f.e. This may cause
problems on some systems, but I couldn't understand the intent
behind every logic (and host environments) in the old script.
F.e. I left out dealing with 'TARGET' variable completely,
including just recently added manual override. If they are
needed with the new system, hopefully we can implement it in
some ways.
Please test. If this works out good, the two launcher scripts
are no longer necessary.
* Changed the way wce cross-builds are initiated. From now on
HB_ARCHITECTURE needs to be set to wce in this case.
For win targets, it needs to be set to win.
So basically, to launch a win build on Linux you need to type:
make HB_ARCHITECTURE=win
(optionally you may set HB_CCPREFIX and HB_CCPATH (before make)
to your liking, if the tool autodetection doesn't work)
For wce builds, use:
export HB_CCPATH=...
make HB_ARCHITECTURE=wce
* config/global.mk
! Fixed previous commit after actual testing.
Now wce cross build on linux works by simply setting HB_CCPATH
and issuing 'make'.
; TODO: Extend above logic to other *nix platforms.
* config/global.mk
* config/wce/mingwarm.mk
* config/win/mingw.mk
+ Added logic to support linux wince cross-builds.
My goal was to implement make_gnu_xmingwce.sh functionality
inside make files. I've yet to install cegcc to my Linux,
so I didn't test it yet, but please do. It's experimental
yet and ATM the wce target can be enabled by pointing
HB_CCPATH to cegcc install dir. This may change in the future.
+ In cross build scenarios the default HB_INSTALL_PREFIX is
now /usr/local/harbour-<arch>-<comp>
Somewhat replicating current .sh behaviour. Not exactly
because I wanted to stay with conventions found in other parts
of Harbour. With .sh the dir was:
/usr/local/arm-wince-mingwce-harbour
With new system it's:
/usr/local/harbour-wce-mingwarm
; QUESTION: I'm not familiar with .spec files, but it looks there are
some redunant settings inside harbour-wce-spec.
F.e. HB_CONTRIBLIBS should be left to default and local
Makefiles should do the ground work of deciding whether
a given contrib needs to be built.
CC_HB_USER_PRGFLAGS also seems redundant now, although
there may still be some .sh logic which needs it.
'make -i' seems dangerous option when creating a release
package.
* make_gnu_xmingwce.sh
! Typo in comment.
* ChangeLog
; Added missing change to one the recent entries:
---
* config/global.cf
+ Added 'SHELL := $(COMSPEC)' to OS/2 shells.
I'm not sure this is needed, please test and confirm
whether it solves possible interference with sh.exe in
PATH.
---
* harbour/source/rtl/hbproces.c
! fixed hb_fsProcessRun() (__RUN() and HB_RUN()) in WinCE builds
to pass application name as 1-st parameters
+ added parameters passing to hb_fsProcessRun() (__RUN() and HB_RUN())
in WinCE builds - please test.
* utils/hbmk2/hbmk2.prg
+ Added support for HB_INSTALL_PREFIX macros (recently introduced
in global.cf). Maybe it'd be better to not use this build-time
envvar at all.
Now that autoconfiguration is working pretty well, I think it may
be time to drop support for some envvars in hbmk2. In case of
HB_INSTALL_PREFIX it rather makes things more complicated.
* include/hbsetup.h
% Optimized out HB_OS_UNIX_COMPATIBLE. HB_OS_UNIX should be used
instead. Since the definition logic for HB_OS_UNIX was completely
different from HB_OS_UNIX_COMPATIBLE, it may be possible that
it's not 100% the same as before, although their meaning was
the same.
HB_OS_UNIX_COMPATIBLE is still available for compatibility, but
it's now marked with HB_LEGACY_LEVEL2 and will be removed in next
major release.
* include/hbapifs.h
* include/hb_io.h
* include/hbsocket.h
* source/vm/set.c
* source/debug/dbgentry.c
* source/common/hbffind.c
* source/rtl/fscopy.c
* source/rtl/fstemp.c
* source/rtl/gtstd/gtstd.c
* source/rtl/gttrm/gttrm.c
* source/rtl/version.c
* source/rtl/hbsocket.c
* source/rtl/copyfile.c
* source/rtl/filesys.c
* source/rtl/console.c
* source/rtl/cputime.c
* source/rtl/gtpca/gtpca.c
* source/rtl/fssize.c
* source/rtl/memofile.c
* source/compiler/hbcomp.c
* source/compiler/gencobj.c
* contrib/hbct/files.c
* contrib/xhb/xhbcopyf.c
* contrib/xhb/hbserv.c
* contrib/hbnf/getenvrn.c
* HB_OS_UNIX_COMPATIBLE -> HB_OS_UNIX
* config/global.cf
* HB_UNIX_COMPATIBLE -> HB_OS_UNIX
* contrib/hbxbp/tests/demoxbp.prg
! Opening test.dbf in RO mode to avoid it being update.
Couldn't spot the code which touched it.
* config/global.cf
- Deleted unused and commented code section.
* INSTALL
- Deleted HB_INSTALL_PREFIX from quickstart instructions.
! Modified hbmk2 hello.prg command to work on *nixes.
(I never tried this default scenario though)
* Modified non-*nix hbmk2 commands to look more natural
and similar to *nix.
- Deleted mingw32-make as suggested tool for dos builds done
on NT hosts. It chokes with too long cmdline with mingw32-make.
* bin/hb-mkdyn.sh
- Deleted wininet from wce liblist.
* make_gnu_xmingw.sh
+ Allow user to override the default TARGET.
* config/global.cf
! Don't add /harbour (LIB|INC)POSTFIX if HB_INSTALL_PREFIX contains
/usr/home (synonym of /home under FreeBSD).
! Fixed INCPOSTFIX usage so it works now.
* config/linux/global.cf
! Added x86_64 to -fPIC machines.
* config/global.cf
+ Package creation now only works if run from Harbour source root.
+ 'clean' of pkg dir will only happen now if run from Harbour source root.
* ChangeLog
* config/globsh.cf
+ 'clean' now deletes pkg/<arch>/<comp> dir. This ensures that
package creation always starts with a fresh empty dir.
Resolves existing TOFIX (marked DONE).
* config/global.cf
! Fixed to process HB_BUILD_NAME when forming package install dir
inside /pkg dir.
% Delete an unnecessary line.
* package/mpkg_win.nsi
* config/global.cf
* bin/postinst.bat
* Little bit changed the way dirs/names are communicated from
GNU Make to postinst package creation code.
The goal was to have an internal PKG_DIR var which is similar
to other *_DIR vars. It was also possible to delete some
exported vars this way.
- Disabled package creation for non-NT shells (= pure MS-DOS).
I found no way to make this work cleanly on pure MS-DOS.
+ Documented weird requirements of zip building. Also to better
clarify why abspath is needed in GNU Make.
* config/instsh.cf
* Updated OS/2 comment to clarify some issues according to
information from Maurilio.
* ChangeLog
* source/rtl/Makefile
* source/rdd/Makefile
* config/bin.cf
* config/wce/global.cf
* config/rules.cf
% Using HB_PRGFLAGS instead of HB_GCMODE. Logic is now simple
and standard.
* config/global.cf
* config/rules.cf
* config/dos/watcom.cf
* config/win/watcom.cf
* config/os2/watcom.cf
% Using HB_PRGFLAGS instead of HB_FLAGSEXTRA.
% Using HB_CFLAGS and HB_PRGFLAGS instead of HB_CDBG.
% Using HB_CFLAGS instead of HB_CUNICODE.
% Moved all flags from HB_RULE to HB_FLAGS (except HB_USER_PRGFLAGS).
This only affects dos/watcom, more cmdlines will be put in HARBOURCMD.
* external/sqlite3/Makefile
* external/libhpdf/Makefile
* external/libpng/Makefile
* source/vm/maindllh/Makefile
* source/vm/maindllp/Makefile
* source/vm/vmmt/Makefile
* source/rtl/gtxwc/Makefile
* source/rtl/gtcrs/Makefile
* source/rtl/gttrm/Makefile
* source/rtl/gtsln/Makefile
* contrib/gtalleg/Makefile
* contrib/hbmysql/Makefile
* contrib/hbodbc/Makefile
* contrib/hbsqlit3/Makefile
* contrib/hbqt/Makefile
* contrib/hbfbird/Makefile
* contrib/hbcurl/hbcurls/Makefile
* contrib/hbcurl/Makefile
* contrib/gtqtc/Makefile
* contrib/rddsql/sddmy/Makefile
* contrib/rddsql/sddpg/Makefile
* contrib/rddsql/sddfb/Makefile
* contrib/rddsql/sddodbc/Makefile
* contrib/hbhpdf/Makefile
* contrib/hbpgsql/Makefile
* contrib/rddads/Makefile
* contrib/hbfimage/Makefile
* contrib/hbgd/Makefile
* contrib/hbtip/hbtipssl/Makefile
* contrib/hbssl/Makefile
! Using HB_CFLAGS/HB_PRGFLAGS to specify Makefile local
compiler settings instead of modifying HB_USER_*FLAGS.
This way the double defines are gone.
+ Included global.cf to all Makefiles which needs to modify
HB_CFLAGS/HB_PRGFLAGS. Probably we should move global.cf
inclusion to the top of all Makefiles as a convention.
Not terribly elegant, but solves a few problems and
more efficient.
% Changed ', -I$(d))' to ',-I$(d))' along the way. This avoids
wasting an extra space in command line.
; NOTE: It'd still be beneficial to find a way to get rid of
'descend' rule. It could make cmdlines cleaner/shorter
and whole build process simpler/faster. I didn't manage
so far.
* source/vm/Makefile
* Deleted 'HB_USER_CFLAGS := $(subst -DHB_MT_VM,,$(HB_USER_CFLAGS))'
(since it's no longer supported practice to modify user supplied flags)
I'm not sure if it's important enough to have. AFAIK there is no
reason users use HB_MT_VM define, it's not a public setting, so we
shouldn't go too much far in trying to protect against its presence.
If it's needed, we should add 'HB_CFLAGS += -DHB_ST_VM' and handle
HB_ST_VM as the reverse of HB_MT_VM inside the source.
* source/rdd/dbffpt/dbffpt1.c
* source/rdd/dbffpt/Makefile
* Optimizations are now disabled using #pragma for __XCC__ instead of
deleting them from HB_USER_CFLAGS. Please test, and see above on
modifying HB_USER_CFLAGS in our make files.
* bin/postinst.cmd
! Deleted double quote from target dir in xcopy commands.
Blind change.
* utils/hbtest/rt_main.ch
* utils/hbtest/hbtest.hbp
* utils/hbtest/Makefile
+ Changed to use #pragma to turn on line numbers instead of
using HB_USER_PRGFLAGS (or HB_PRGFLAGS). This way there is
no need for Makefile hack/tweaks and it works with all make methods
automatically.
* config/instsh.cf
! Partially restored a few things which were lost without comments
in prev OS/2 commit:
- Formatting
- '@' prefixes
- '-' prefixes
* config/global.cf
+ Added new plans for further setting cleanup.
* config/global.cf
* config/rules.cf
* config/dos/watcom.cf
* config/win/watcom.cf
* config/os2/watcom.cf
+ Added HB_CFLAGS, HB_PRGFLAGS. These will now replace all
Makefile level manipulation of flags. Should eventually
replace all HB_USER_*FLAGS, HB_GCMODE, HB_INC_DEPEND,
HB_CUNICODE, HB_CDBG and HB_FLAGSEXTRA variables, if everything
goes well.
* contrib/hbwin/wapi_winuser.c
+ Added WAPI_GETKEYSTATE().
* config/none.cf
* config/global.cf
* config/instsh.cf
* config/dirsh.cf
! Fixed minor visual glitch on non-sh shells where some echoed messages
were shown inside double quotes.
There is only one such but very rare message left in global.cf.
* ChangeLog
! Typo in prev entry.
* config/dos/djgpp.cf
* Changed 'echo.' to 'echo' to make it shell independent just like
all other echo commands in the make config files.
Pre-2.02 DJGPP distros had an echo.exe, originally I may have
added to dots to force using the shell echo instead of this external
one. Quick tests made with djecho.exe (the new name of echo.exe)
showed it working. Maybe old echo.exe had some bad side effects,
I don't remember and ChangeLog entry doesn't have the reason.
* utils/hbmk2/hbmk2.prg
! Minor fix for darwin and automatic entry function detection using 'nm'.
nm on Darwin doesn't seem to support '--defined-only -C' option.
Przemek, please check this, maybe this hurts functionality.
- make_gnu.sh
* config/linux/global.cf
- Deleted. The -fPIC logic is moved to linux/global.cf but it's
incomplete yet, some comments were added there.
* contrib/hbcurl/hbcurl.hbc
+ contrib/hbcurl/hbcurls.hbc
+ Added separate .hbc file for static mode hbcurl lib.
This replaces former HB_CURL_STATIC setting.
* contrib/hbcurl/Makefile
* external/Makefile
* contrib/Makefile
% Minor optimization to DIRS var.
(in few cases it wasn't init to empty with := operator)
+ contrib/hbtip/hbtipssl.hbc
+ contrib/hbtip/hbtipssl
+ contrib/hbtip/hbtipssl/Makefile
* contrib/hbtip/Makefile
+ Now building separate SSL enabled hbtip lib in case HB_HAS_OPENSSL.
+ Added separate .hbc file for SSL enabled hbtip lib, which also
pulls hbssl libs and dependencies.
* config/prg.cf
* config/lib.cf
* config/bin.cf
* config/globsh.cf
* config/dir.cf
* config/c.cf
* config/header.cf
* config/doc.cf
* config/instsh.cf
* config/dirsh.cf
% Optimized remaining (more delicate) logic to use :=
instead of = assigment operator.
At this point there is nothing left to optimize in this
regard.
; I didn't restest parallel build and BISON related rules,
please do so, I hope they aren't broken. Or in case you
notice anything strange please shout.
* config/global.cf
+ Added link to GNU Make NEWS files which describes at which release
some features were added.
* config/dos/watcom.cf
* config/win/watcom.cf
* config/os2/watcom.cf
! Synced special watcom CC_RULEs with recent changes to global CC_RULE.
* INSTALL
* make_gnu.sh
* config/global.cf
* config/lib.cf
* config/bin.cf
% Deleted HB_USER_MAKEFLAGS envvar.
Simply pass these flags to make executable (or .sh scripts)
* config/global.cf
+ Added some TODO/NOTE/TOFIX about GNU Make version compatibility
of some feature we're using.
+ Showing MAKEFLAGS.
% Using MAKEFLAGS for --no-print-directory trick.
* INSTALL
+ Separated instructions for linux/darwin and rest of *nixes.
The only difference is that the latter needs 'gmake' command
to start.
+ MSYS/Cygwin section extended a bit. (It's now possible to
build for any compilers from MSYS/Cygwin)
* make_gnu.sh
% Deleted unused HB_ARCH64 variable.
* config/global.cf
- Deleted svnversion trick since it's not guaranteed this
command is present on builder's system.
+ Added doc links from make_gnu.sh. (deleted from there)
* Changed to default HB_INSTALL_PREFIX to /usr/local on
*nix systems. I very much don't like this (even though I
understand why it's done that way).
+ Added {HB_TOP} macro for HB_INSTALL_PREFIX, so if someone
wants to install Harbour in its own dir, it enough to do:
export/set HB_INSTALL_PREFIX={HB_TOP}.
This is crucial if someone doesn't want to "pollute" system
areas with Harbour builds, doesn't want to use admin rights
to make the install, or if someone wants to use multiple
Harbour builds in parallel for any reason. hbmk2 works
perfectly without Harbour being installed to system location.
+ Whenever HB_INSTALL_PREFIX is modified by GNU Make (defaulting
or macro replacement), the new value will be presented for the
user.
% Minor optimization.
* Minor cleanups to pkg dir assembling logic.
* config/linux/global.cf
+ Added -fPIC detection logic. This is just a start, it doesn't
replicate full logic found in make_gnu.sh. Przemek, if you have
some spare time, please look into it.
* source/pp/Makefile
! Fixed to not throw harmless but still confusing error
messages when cleaning hbpp binary parts.
* bin/hb-mkdyn.bat
% Minor optimization in initialization.
- make_gnu_os2.cmd
- make_gnu.bat
- Deleted these starter batch files.
; Please use 'mingw32-make' (if installed in PATH) or
'config/mingw32-make' (if not installed in PATH) instead.
OS/2 users should use 'make'.
; OS/2 users will have to add:
set HB_USER_CFLAGS=-DTCPV40HDRS
to local configuration unless they used:
set HB_OS2_TCP32=yes
I'd like to ask OS/2 users to add (now deleted) related
documentation to one of our docs: INSTALL or include/hbsetup.h.
Since this was a setting to disable some default setup,
we can readd this setting with another name to
config/global.cf if needed. Pls tell me about it.
; TODO: Delete make_gnu.sh and/or integrate make_gnu_xmingw*.sh.
There is still two things which is done by make_gnu.sh:
- Detection for -fPIC option.
- Setting HB_INSTALL_PREFIX to system location.
(I don't like this, but it seems to be standard
behaviour on Linux, so we may want to change GNU Make
default logic according to that)
; TODO: Move mpkg_*.sh files with related support files to /package.
Best would be to integrate them to postinst.sh.
* INSTALL
+ Added TOFIX to one example: djgpp with NT shell will choke on
too long cmdline when building external/libhpdf.
; TOFIX: see above. (if possible. If it's too complicated we
should simply drop this scenario, after all DJGPP GNU
Make also works well.
* contrib/gtalleg/Makefile
* contrib/hbmysql/Makefile
* contrib/hbodbc/Makefile
* contrib/hbsqlit3/Makefile
* contrib/hbqt/Makefile
* contrib/hbfbird/Makefile
* contrib/hbxbp/Makefile
* contrib/hbcurl/Makefile
* contrib/gtqtc/Makefile
* contrib/rddsql/sddmy/Makefile
* contrib/rddsql/sddpg/Makefile
* contrib/rddsql/sddfb/Makefile
* contrib/rddsql/sddodbc/Makefile
* contrib/hbhpdf/Makefile
* contrib/gtwvg/Makefile
* contrib/hbpgsql/Makefile
* contrib/rddads/Makefile
* contrib/hbfimage/Makefile
* contrib/hbgd/Makefile
* contrib/hbwin/Makefile
+ Indentation.
