summaryrefslogtreecommitdiff
path: root/build.conf
Commit message (Collapse)AuthorAgeFilesLines
* Building PPC and linking on darwin 9 in sequence of _all, _atomic, _(others) wrowe2007-10-261-2/+2
| | | | | | | | | | | | | | | | | | | .o images results in the following; ld: bc out of range (79008 max is +/-64K) from _apr_atomic_add32 in atomic/unix/.libs/ppc.o to _apr_atomic_add32$stub in .libs/libapr-1.0.dylib in _apr_atomic_add32 from atomic/unix/.libs/ppc.o Apparently (and I don't know why we are coupled to the $stub's in the first place) moving atomic to the end of the code segment places ppc.o in proximity to the $stub relocations. Of course this commit may inversely clobber another platform. But not Linux, at least. Committing before I look for various exceptions to this solution. git-svn-id: http://svn.apache.org/repos/asf/apr/apr/trunk@588793 13f79535-47bb-0310-9956-ffa450edef68
* * build/gen-build.py: Make location of DSP file configurable and itsjorton2006-07-121-0/+2
| | | | | | | | | use optional (hopefully fixes apr-util build). * build.conf: Define DSP file location. git-svn-id: http://svn.apache.org/repos/asf/apr/apr/trunk@421276 13f79535-47bb-0310-9956-ffa450edef68
* Fix the generation of the build-outputs.mk file. That file must remaingstein2004-02-211-15/+8
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | platform independent since it gets included with the tarball, and the tarball is targeted for all platforms. In this new scheme, the build-outputs.mk includes symbols named OBJECTS_<plat> which specifies the necessary object files for each platform. The apr_rules.mk(.in) file then pulls in the right symbol for the configured platform. Since apr-util directly uses apr_rules.mk, then it picks up the same logic with no change. The build.conf was altered to include a set of platform-independent files, and subdirectories which contain per-platform subdirectories. This was needed to help out the logic which selects object files based on whether a platform is present in each source subdir. (and if a platform subdir is not present, then "parent" gets used instead; e.g. use 'unix' if 'beos' is not in the subdir) Lastly, configure.in was updated to select the appropriate set of objects for the build. * build.conf: (paths): remove per-platform symbols (platform_dirs): new option to list the dirs which have platform subdirs * configure.in: - define and substitute OBJECTS_PLATFORM to reference the platform specific set of objects, which get defined in build-outputs.mk * build/apr_rules.mk.in: - define OBJECTS in terms of the substituted OBJECTS_PLATFORM symbol * build/gen-build.py: (PLATFORMS): the various platforms found in APR. unused, actually. (MAKE_PLATFORMS): the platforms which use a Makefile, and what default platform should be used if SUBDIR/<platform> is not present. (main): update logic to generate multiple OBJECTS_* symbols. this now includes logic to deal with subdirs which may not have a platform subdir for the platform in question, thus needing to default to another. (write_objects): some code factored out of main() to process a list of source files into dependencies lines in the makefile. (get_files): no more need to substitute for {platform}, and simplified some of the .split() stuff. (get_platform): no longer required. we generate for all platforms. git-svn-id: http://svn.apache.org/repos/asf/apr/apr/trunk@64914 13f79535-47bb-0310-9956-ffa450edef68
* First whack at switching to a single top-level make. This adds a dependencygstein2004-02-051-0/+33
upon Python at packaging time, but not at end-user config/build time. As far as I can tell, the build continues to function properly. (out-of-dir config/make not tested, and apr-iconv prolly needs some work) The buildconf scripts now generate a build-outputs.mk file which is included by the root's Makefile (via the build/gen-build.py script). bulid-outputs.mk specifies all of the various files present in the distribution. The top-level Makefiles were simplified to use an $(OBJECTS) symbol rather than 'find'ing them. Similarly, a $(HEADERS) symbol is used for the exports. The corresponding delete-* targets were eliminated since we have a precise set of inputs. The subdirs' Makefiles were removed since they are no longer called/used. The apr-util/uri Makefile was responsible for compiling a C program to generate the uri_delims.h file. That process was replaced by a Python script to generate the header (called by buildconf). The .c and .dsp were left for the Windows build to continue, but that should be revamped. build/apr_rules.mk was revamped somewhat to avoid recursion, but a lot of cleanup is still needed. Much of the recursive/local/x- logic is no longer needed and can be elimianated. rules.mk was created for inclusion by N makefiles, but that isn't really true any more, so it could probably be tossed (caveat: test/Makefile). Saved for a phase 2. Some additional work was added to properly clean up files in */build/, rather than relying on a makefile in there. git-svn-id: http://svn.apache.org/repos/asf/apr/apr/trunk@64891 13f79535-47bb-0310-9956-ffa450edef68