- Generate distfiles directly from pkg-plist. This list was compiled from
running "ktrace -i -tn make" on openjdk6 and openjdk7 ports.
- Re-add jre/bin/java to work around a bug in openjdk7 build infrastructure.
It is not really necessary but langtools fails to build without it because
langtools/make/tools/GenStubs/GenStubs.java executes it based on java.home
property, which points to jre subdirectory. Note this problem only exists
on recent jdk7 langtools and jdk6 and jdk8 are not affected.
Feature safe: yes
files are only installed if the option is set. In future it will likely
also mean to enable the debug and/or fastdebug build.
PR: 172627
Feature safe: yes
. Group the USE_* directives together.
. Trim the Makefile header. [2]
. Convert to optionsNG. [2]
. Remove the ABI numbers from LIB_DEPENDS. [2]
. Fix the POLICY option. [2]
. Fix permissions and ownership of jexec. [2]
. Make sure the bootstrap JDK is used by the ant bootstrap. [2]
Submitted by: kuriyama@ [1]
jkim@ [2]
Feature safe: yes
- Make sure to destroy all symlinks pointing to ${PREFIX}/bin/javavm and
${PREFIX}/etc/javavms when we deinstall the package.
Reviewed by: glewis (maintainer)
Feature safe: yes
java/jdk16 is outdated with security issues consider using java/openjdk6 [1]
java/jmp is java 1.5 only
Approved by: java (glewis) [1] \
Ronald Klop <ronald@klop.ws> (maintainer) [2]
Feature safe: yes
http://icedtea.classpath.org/hg/release/icedtea6-1.11/rev/d9564350faa6http://blog.fuseyism.com/index.php/2012/10/19/security-icedtea-1-10-10-1-11-15-2-1-3-2-2-3-2-3-3-released/
- Completely turn off parallel build by default and remove parallel build
hack for HotSpot. There were several reports that it fails to build under
certain environment, ports/162991 for example. Users can still do parallel
build by setting FORCE_MAKE_JOBS (and MAKE_JOBS_NUMBER if desired).
- Implement os::available_memory(). Now it is consistent with "vm.vmtotal"
sysctl(3) MIB rather than bogus (physical memory / 4).
- Prefer sysconf(_SC_NPROCESSORS_CONF) over HW_NCPU sysctl MIB to get the
number of installed processors. There is no functional difference except
for CURRENT, which obtains the information from ELF aux vector.
- Prefer sysconf(_SC_PHYS_PAGES) * sysconf(_SC_PAGESIZE) over HW_USERMEM
sysctl MIB to get size of physical memory. Although it looks more logical
to find currently available memory, it has an inevitable side-effect, i. e.,
it changes dynamically depending on current wired page count. Therefore,
it is unpredictable and not too useful some times. For example, launcher
uses the parameter to determine initial heap size and machine class for i386.
Now it is more consistent with other places (and Linux JDK/JREs, including
the ones we have in ports tree).
- Implement os::active_processor_count() using cpuset_getaffinity(2). For
example, Runtime.getRuntime().availableProcessors() now returns number of
available processors for the current process as it should.
- Sync. launchers (java_md.c) for HotSpot and JDK as much as possible for
maintainability. As a good side-effect, launcher for i386 can now determine
machine class based on the current hardware configuration. Previously,
client VM was always chosen by default.
- Fix CounterGet(), which is only used for debugging launcher.
- Add swap info for os::print_memory_info().
Obtained from: IcedTea project [1]
Feature safe: yes
bootstrapped with pre-built openjdk6 packages from FreeBSD build clusters
on clean 7.4-RELEASE installation, and the ports tree was checked out at
r305682.
Feature safe: yes
manually set, this dependency may not be fulfilled.
- Consistently use WRKSRC over WRKDIR where it makes sense. Although they
are actually the same directory, there is no reason to confuse users.
Feature safe: yes
- Reset maintainership. See ports/164941.
- Make it as amd64 and i386 only. Other architectures needs porting.
- Reduce Makefile headers, add licenses, and convert to optionsNG.
- Add an option to build with native JDK. When it is not selected, GCJ is
used instead. Turned on by default.
The following changes were made in the previous commit.
- Use ECJ 3.8 instead of 4.2. This simplifies and add more Eclipse mirrors.
They are actually the same compilers released with different Eclipse JDTs.
- Add an option to build gjdoc.
- Add an option to build with GMP, which enables java.math.BigInteger.
- Add an option to build with native JDK. When it is turned off, ECJ or GCJ
(when WITH_GCJ is defined) is used for glibj.zip. Note ECJ (with gij) and
GCJ can compile classpath perfectly fine on FreeBSD 9 and later but it seems
libgcj interacts badly with 7 and 8. Need more investigation.
- Reduce unnecessary dependencies. zip is not used when "--with-jar" is
specified. Use pre-built ANTLR jar file, which is only used for gjdoc.
Remove JDK run-time dependency because it is only necessary to execute tools
and these tools are actually wrappers. Now we just execute them with javavm
wrapper (or user-specified VM when JAVACMD environment variable is set).
Note you need a valid JVM to run these commands now.
- Add a patch to fix an ugly JAR updater bug. Without this, the file name
must include "/" in it, e.g., "gjar i ./glibj.zip".
- Fix conflicts with java/classpath. Note security provider configuration
is moved to ${PREFIX}/lib/security/SableVM.security. This file is always
loaded first because the short vendor name is SableVM. If it does not
exist, then classpath.security is loaded (java/security/Security.java).
- Reduce Makefile headers, add licenses, and convert to optionsNG.
