Michal Sojka [Fri, 25 Feb 2011 07:25:23 +0000 (08:25 +0100)]
wvtest: Output "Test function header"
Previously, when something in the test failed before the test output
its header, it looked like the error occurred in the previous test. Now,
the header is the very first thing generated by make so the user see the
proper name of the failed test.
Michal Sojka [Tue, 22 Feb 2011 16:07:11 +0000 (17:07 +0100)]
Fix compilation of manual
The manual includes files from test (to ensure that the documentation
matches the real behavior). Since we converted the tests to a different
format, we have to change the way how the files are included.
Michal Sojka [Tue, 22 Feb 2011 15:44:44 +0000 (16:44 +0100)]
Call submakes with --no-builtin-rules switch
This fixes the bug described in the previous commit. Additionally
we also get a performance gain of about 20% (measured on the total
execution time of the testsuite).
Michal Sojka [Tue, 22 Feb 2011 15:18:38 +0000 (16:18 +0100)]
Add test for bug in bin_SCRIPTS handling
when the name of a script is 'x' and there is a newer file 'x.c' in the
same directory then make applies built-in implicit rules and attempts
to recreate x from x.c. This should not happen.
Michal Sojka [Mon, 21 Feb 2011 19:21:12 +0000 (20:21 +0100)]
Remove support for testing from snippets
As we are migrating the testing framework to wvtest we will not use these.
The idea is that condition similat to those deleted by this patch will
be present in the test scripts.
Michal Sojka [Thu, 17 Feb 2011 23:27:13 +0000 (00:27 +0100)]
wvtest: Set PATH and LD_LIBRARY_PATH as documented in manual
This change also implements quite and verbose variants of the make
output. The quite variant prints RUN and the short name of the
program/script to be run.
Michal Sojka [Thu, 17 Feb 2011 20:26:07 +0000 (21:26 +0100)]
doc: Change how is the version number included
Now, we use @include command, whereas earlier we have edited the source
with sed. The current method has the advantage that error messages refer
directly to the source file instead of the automatically generated one.
Michal Sojka [Fri, 11 Feb 2011 13:59:08 +0000 (14:59 +0100)]
qt: Set DEPENDPATH to generate dependencies on header files
Without this, when some OMK managed hadere has changed, qmake didn't
recompile the QT application and it caused problems. Now, qmake
correctly generate dependencies to headers in _compiled/include.
Michal Sojka [Thu, 29 Jul 2010 22:02:32 +0000 (00:02 +0200)]
tester.py: Do not import nonexistent rulesdef
rulesdef.py is not used for some time, but tester.py still depended on it.
This change recreates the data structuere provided by rulesdef.py on the
fly from the compiled Makefile.rules files.
I had a stale rulesdef.pyc on my laptop so I didn't discover this bug
until I recently upgraded to newer python version (2.6).
Michal Sojka [Thu, 29 Jul 2010 21:06:07 +0000 (23:06 +0200)]
Remove -C flag from install command
This flag is now useless, as the comparison must be (and is) done
before calling install. The -C flag was not available in older
versions of install (e.g. on Debian Lenny) so this change allows OMK
to be run on older systems.
Michal Sojka [Fri, 2 Apr 2010 15:29:52 +0000 (17:29 +0200)]
Comment out BACK2TOP_DIR
This variable is never used in Makefile.rules. It was assigned
in every make instance and its assignment involved sed invocation. This
way, sed was uselessly invoked several hundred times during compilation
of bigger projects.
Michal Sojka [Fri, 2 Apr 2010 12:33:56 +0000 (14:33 +0200)]
include-pass: Move the comment to separate line
When the comment was on the command line, shell has often been invoked
to just "interpret" the comment (when there were no commands before
the comment). Now, if there are no commands, shell is not invoked.
Michal Sojka [Fri, 2 Apr 2010 12:24:41 +0000 (14:24 +0200)]
Remove unnecessary subshell invocation from include pass
When we copied include files there were several places where a subshell
was invoked without a real reason. These places were modified to not use
subshell.
Michal Sojka [Sun, 21 Mar 2010 20:42:01 +0000 (21:42 +0100)]
Allow sources in subdirectories
When you have *_SOURCES=subdir/file.c the file.o should go to subdir
directory under _build, but this directory does not exist. It was
possible to workaround this limitation by adding SUBDIRS=subdir to
Makfile.omk and putting empty Makefile.omk to subdir.
With this change, the workaround is no longer necessary.
Michal Sojka [Mon, 15 Mar 2010 08:15:38 +0000 (09:15 +0100)]
sysless: Switch order of LOADLIBES and *_MOREOBJS
This change was proposed by Tran Duy Khanh. He describes it as:
Pozmenil jsem poradi linkovani *_MOREOBJS a LOADLIBES, jinak se mi
slinkuje system_stub veci z libc driv nez system_stub syslessu. Myslim
si, ze i tak je to spravnejsi, protoze jinak se neda "prepsat" veci ze
standardni knihovny.
Michal Sojka [Wed, 20 Jan 2010 16:18:03 +0000 (17:18 +0100)]
Fixes to support overriding SUBDIRS from command line
It is now possible to run 'make SUBDIRS=xxx' to compile only subdirectory
xxx and the current directory. It might be useful for speeding up
compilation of large projects if it is know that only small part of the
tree changed.
Michal Sojka [Thu, 21 May 2009 10:43:13 +0000 (12:43 +0200)]
Added dependency of IDL generated files on IDL compiler
It is expected that the IDL compiler executable is the first word in the
IDL_COMPILER variable. If the first word is not a file (e.g. shell
variable assignment), it will work correctly as $(wildcard ) is used and
it will return empty string in that case.
Michal Sojka [Thu, 21 May 2009 10:43:13 +0000 (12:43 +0200)]
Added dependency of IDL generated files on IDL compiler
It is expected that the IDL compiler executable is the first word in the
IDL_COMPILER variable. If the first word is not a file (e.g. shell
variable assignment), it will work correctly as $(wildcard ) is used and
it will return empty string in that case.
Michal Sojka [Tue, 12 May 2009 09:49:09 +0000 (11:49 +0200)]
Changed ambiguous CONFGEN filename
CONFGEN message included the name of the stamp file without the
extension. This was ambiguous because in the case of named (not
LOCAL_CONFIG_H) config files, the final header file is located in
different directory than stamp file and therefore the message mentioned
a non-existent file path. Now, the CONFGEN message only includes
filename without directory. The full path is shown in update message.