Not referencing them caused them to be missing from the source packages.
This also moves all files that were in different dirs under the common
additional-files dir.
The correct solution is probably to have this package explicitly download an earlier HPKG and use it, because circular dependencies are against general notions of sanity (and the Kitchen can't deal with them)
Removed the full-stop and repeating package name in SUMMARY.
Removed hard line breaks, broke up some very long paragraphs and
use bullet lists where possible.
All done by browsing through HaikuDepot. There may be (many) more left
in the whole haikuports repo...
Included 10 extended desriptions done by soyoye14 for GCI 2014
(https://www.google-melange.com/gci/task/view/google/gci2014/5240373098053632)
Removed/insert whitespace where it seems to be the custom in recipes.
Sometimes rearranged elements of a recipe (moving license and copyright
up, for example).
* Referring the current haiku version explicitly is not needed, since
the RequiresUpdater takes care of setting the version of Haiku used
for building a package.
* Patch the build system to allow building with an external zlib
instead of using the included one. Since the included zlib symbols
would be visible also to other JNI code (e.g. translators used by the
SWT libs), this could cause problems where a newer version was
required (e.g. libpng used by the PNG translator).
* Verified ALT_JDK_IMPORT_PATH as needed to be set.
Add to the package:
* bin/java-config: When invoked with option -H it prints the java home
directory.
* data/profile.d/openjdk.sh: It is automatically sourced by
/etc/profile and sets JAVA_HOME.
We invoke make twice to work around a bug in the build system which
causes (some) generated java files not to be picked up and compiled
the first time around.
The issue is in jdk/make/common/Rules.gmk the way $(JAVA_SOURCE_LIST)
is generated. That happens as a side-effect of the */%.class rules
which are triggered due to $(FILES_class) being listed as prerequisites
of $(JAVA_SOURCE_LIST). That means $(FILES_class) -- which ultimately
is based on a shell invocation which finds .java files via find -- is
evaluated in the first pass, i.e. long before the generated source
files are made. Hence they aren't compiled and will be missing from
rt.jar. Running make a second time works around the issue, since the
source files will already be present then.
* Disable source package for the time being.
* Implement INSTALL.
* Add provides for the commands.
The package builds, but due to some missing classes it doesn't really
work yet.
Doesn't fully build yet. The build runs for a while, but eventually,
when building the jaxws/jaf sources, javac hangs. It blocks trying to
write to stderr, a pipe whose other end isn't read from (by Ant). I
haven't determined yet, whether the hang is an Ant bug and the issue
is javac simply encountering too many issues, or whether javac writing
a lot to stderr is just fine (verbose diagnostics?) and the hang is
actually a Haiku (pipe related?) bug, a combination of the two, or
something else entirely.
