OpenJDK Build README
This README file contains build instructions for the OpenJDK. Building the source code for the OpenJDK requires a certain degree of technical expertise.
!!!!!!!!!!!!!!! THIS IS A MAJOR RE-WRITE of this document. !!!!!!!!!!!!!Some Headlines:
- The build is now a "
configure && make" style build
- Any GNU make 3.81 or newer should work
- The build should scale, i.e. more processors should cause the build to be done in less wall-clock time
- Nested or recursive make invocations have been significantly reduced, as has the total fork/exec or spawning of sub processes during the build
- Windows MKS usage is no longer supported
- Windows Visual Studio
vcvars*.batfiles are run automatically
- Ant is no longer used when building the OpenJDK
- Use of ALT_* environment variables for configuring the build is no longer supported
The OpenJDK sources are maintained with the revision control system Mercurial. If you are new to Mercurial, please see the Beginner Guides or refer to the Mercurial Book. The first few chapters of the book provide an excellent overview of Mercurial, what it is and how it works.
For using Mercurial with the OpenJDK refer to the Developer Guide: Installing and Configuring Mercurial section for more information.
Getting the SourceTo get the entire set of OpenJDK Mercurial repositories use the script
get_source.shlocated in the root repository:Once you have all the repositories, keep in mind that each repository is it's own independent repository. You can also re-run
hg clone http://hg.openjdk.java.net/jdk8/jdk8 YourOpenJDK
./get_source.shanytime to pull over all the latest changesets in all the repositories. This set of nested repositories has been given the term "forest" and there are various ways to apply the same
hgcommand to each of the repositories. For example, the script
make/scripts/hgforest.shcan be used to repeat the same
hgcommand on every repository, e.g.
bash ./make/scripts/hgforest.sh status
The set of repositories and what they contain:
Repository Contains . (root) common configure and makefile logic hotspot source code and make files for building the OpenJDK Hotspot Virtual Machine langtools source code for the OpenJDK javac and language tools jdk source code and make files for building the OpenJDK runtime libraries and misc files jaxp source code for the OpenJDK JAXP functionality jaxws source code for the OpenJDK JAX-WS functionality corba source code for the OpenJDK Corba functionality
Repository Source GuidelinesThere are some very basic guidelines:
- Use of whitespace in source files (.java, .c, .h, .cpp, and .hpp files) is restricted. No TABs, no trailing whitespace on lines, and files should not terminate in more than one blank line.
- Files with execute permissions should not be added to the source repositories.
- All generated files need to be kept isolated from the files maintained or managed by the source control system. The standard area for generated files is the top level
- The default build process should be to build the product and nothing else, in one form, e.g. a product (optimized), debug (non-optimized, -g plus assert logic), or fastdebug (optimized, -g plus assert logic).
- The .hgignore file in each repository must exist and should include ^build/, ^dist/ and optionally any nbproject/private directories. It should NEVER include anything in the src/ or test/ or any managed directory area of a repository.
- Directory names and file names should never contain blanks or non-printing characters.
- Generated source or binary files should NEVER be added to the repository (that includes javah output). There are some exceptions to this rule, in particular with some of the generated configure scripts.
- Files not needed for typical building or testing of the repository should not be added to the repository.
The very first step in building the OpenJDK is making sure the system itself has everything it needs to do OpenJDK builds. Once a system is setup, it generally doesn't need to be done again.
Building the OpenJDK is now done with running a
configurescript which will try and find and verify you have everything you need, followed by running
make, e.g.Where possible the
configurescript will attempt to located the various components in the default locations or via component specific variable settings. When the normal defaults fail or components cannot be found, additional
configureoptions may be necessary to help
configurefind the necessary tools for the build, or you may need to re-visit the setup of your system due to missing software packages.
configurescript file does not have execute permissions and will need to be explicitly run with
bash, see the source guidelines.
System SetupBefore even attempting to use a system to build the OpenJDK there are some very basic system setups needed. For all systems:
And for specific systems:
- Be sure the GNU make utility is version 3.81 or newer, e.g. run "
- Install a Bootstrap JDK
All OpenJDK builds require access to a previously released JDK, this is often called a bootstrap JDK. Currently, for this JDK release we require JDK 7 Update 7 or newer. The JDK 7 binaries can be downloaded from Oracle's JDK 7 download site. For build performance reasons is very important that this bootstrap JDK be made available on the local disk of the machine doing the build. You should add its
bindirectory to the
PATHenvironment variable. If
configurehas any issues finding this JDK, you may need to use the
- Insure that GNU make, the Bootstrap JDK, and the compilers are all in your PATH environment variable
Linux Solaris Windows Mac OS X Install all the software development packages needed including alsa, freetype, cups, and xrender.
See specific system packages.
Install all the software development packages needed including Studio Compilers, freetype, cups, and xrender.
See specific system packages.
Install XCode 4.5.2 and also install the "Command line tools" found under the preferences pane "Downloads"
LinuxWith Linux, try and favor the system packages over building your own or getting packages from other areas. Most Linux builds should be possible with the system's available packages.
Note that some Linux systems have a habit of pre-populating your environment variables for you, for example
JAVA_HOMEmight get pre-defined for you to refer to the JDK installed on your Linux system. You will need to unset
JAVA_HOME. It's a good idea to run
envand verify the environment variables you are getting from the default system settings make sense for building the OpenJDK.
Studio CompilersAt a minimum, the Studio 12 Update 1 Compilers (containing version 5.10 of the C and C++ compilers) is required, including specific patches.
The Solaris SPARC patch list is:
- 118683-05: SunOS 5.10: Patch for profiling libraries and assembler
- 119963-21: SunOS 5.10: Shared library patch for C++
- 120753-08: SunOS 5.10: Microtasking libraries (libmtsk) patch
- 128228-09: Sun Studio 12 Update 1: Patch for Sun C++ Compiler
- 141860-03: Sun Studio 12 Update 1: Patch for Compiler Common patch for Sun C C++ F77 F95
- 141861-05: Sun Studio 12 Update 1: Patch for Sun C Compiler
- 142371-01: Sun Studio 12.1 Update 1: Patch for dbx
- 143384-02: Sun Studio 12 Update 1: Patch for debuginfo handling
- 143385-02: Sun Studio 12 Update 1: Patch for Compiler Common patch for Sun C C++ F77 F95
- 142369-01: Sun Studio 12.1: Patch for Performance Analyzer Tools
The Solaris X86 patch list is:
- 119961-07: SunOS 5.10_x86, x64, Patch for profiling libraries and assembler
- 119964-21: SunOS 5.10_x86: Shared library patch for C++_x86
- 120754-08: SunOS 5.10_x86: Microtasking libraries (libmtsk) patch
- 141858-06: Sun Studio 12 Update 1_x86: Sun Compiler Common patch for x86 backend
- 128229-09: Sun Studio 12 Update 1_x86: Patch for C++ Compiler
- 142363-05: Sun Studio 12 Update 1_x86: Patch for C Compiler
- 142368-01: Sun Studio 12.1_x86: Patch for Performance Analyzer Tools
The Oracle Solaris Studio Express compilers at: Oracle Solaris Studio Express Download site are also an option, although these compilers have not been extensively used yet.
Windows Unix ToolkitBuilding on Windows requires a Unix-like environment, notably a Unix-like shell. There are several such environments available of which Cygwin and MinGW/MSYS are currently supported for the OpenJDK build. One of the differences of these systems from standard Windows tools is the way they handle Windows path names, particularly path names which contain spaces, backslashes as path separators and possibly drive letters. Depending on the use case and the specifics of each environment these path problems can be solved by a combination of quoting whole paths, translating backslashes to forward slashes, escaping backslashes with additional backslashes and translating the path names to their "8.3" version.
CYGWINCYGWIN is an open source, Linux-like environment which tries to emulate a complete POSIX layer on Windows. It tries to be smart about path names and can usually handle all kinds of paths if they are correctly quoted or escaped although internally it maps drive letters
<drive>:to a virtual directory
You can always use the
cygpathutility to map pathnames with spaces or the backslash character into the
C:/style of pathname (called 'mixed'), e.g.
cygpath -s -m "path".
Note that the use of CYGWIN creates a unique problem with regards to setting
PATH. Normally on Windows the
PATHvariable contains directories separated with the ";" character (Solaris and Linux use ":"). With CYGWIN, it uses ":", but that means that paths like "C:/path" cannot be placed in the CYGWIN version of
PATHand instead CYGWIN uses something like
/cygdrive/c/pathwhich CYGWIN understands, but only CYGWIN understands.
The OpenJDK build requires CYGWIN version 1.7.16 or newer. Information about CYGWIN can be obtained from the CYGWIN website at www.cygwin.com.
By default CYGWIN doesn't install all the tools required for building the OpenJDK. Along with the default installation, you need to install the following tools.Note that the CYGWIN software can conflict with other non-CYGWIN software on your Windows system. CYGWIN provides a FAQ for known issues and problems, of particular interest is the section on BLODA (applications that interfere with CYGWIN).
Binary Name Category Package Description ar.exe Devel binutils The GNU assembler, linker and binary utilities make.exe Devel make The GNU version of the 'make' utility built for CYGWIN m4.exe Interpreters m4 GNU implementation of the traditional Unix macro processor cpio.exe Utils cpio A program to manage archives of files gawk.exe Utils awk Pattern-directed scanning and processing language file.exe Utils file Determines file type using 'magic' numbers zip.exe Archive zip Package and compress (archive) files unzip.exe Archive unzip Extract compressed files in a ZIP archive free.exe System procps Display amount of free and used memory in the system
MinGW/MSYSMinGW ("Minimalist GNU for Windows") is a collection of free Windows specific header files and import libraries combined with GNU toolsets that allow one to produce native Windows programs that do not rely on any 3rd-party C runtime DLLs. MSYS is a supplement to MinGW which allows building applications and programs which rely on traditional UNIX tools to be present. Among others this includes tools like
make. See MinGW/MSYS for more information.
Like Cygwin, MinGW/MSYS can handle different types of path formats. They are internally converted to paths with forward slashes and drive letters
<drive>:replaced by a virtual directory
/<drive>. Additionally, MSYS automatically detects binaries compiled for the MSYS environment and feeds them with the internal, Unix-style path names. If native Windows applications are called from within MSYS programs their path arguments are automatically converted back to Windows style path names with drive letters and backslashes as path separators. This may cause problems for Windows applications which use forward slashes as parameter separator (e.g.
cl /nologo /I) because MSYS may wrongly replace such parameters by drive letters.
In addition to the tools which will be installed by default, you have to manually install the
msys-unzippackages. This can be easily done with the MinGW command line installer:
mingw-get.exe install msys-zip
mingw-get.exe install msys-unzip
Visual Studio 2010 Compilers
The 32-bit and 64-bit OpenJDK Windows build requires Microsoft Visual Studio C++ 2010 (VS2010) Professional Edition or Express compiler. The compiler and other tools are expected to reside in the location defined by the variable
VS100COMNTOOLSwhich is set by the Microsoft Visual Studio installer.
Only the C++ part of VS2010 is needed. Try to let the installation go to the default install directory. Always reboot your system after installing VS2010. The system environment variable VS100COMNTOOLS should be set in your environment.
Make sure that TMP and TEMP are also set in the environment and refer to Windows paths that exist, like
/cygdrive/c/temp, and not
C:\tempis just an example, it is assumed that this area is private to the user, so by default after installs you should see a unique user path in these variables.
Mac OS XMake sure you get the right XCode version.
ConfigureThe basic invocation of the
configurescript looks like:This will create an output directory containing the "configuration" and setup an area for the build result. This directory typically looks like:
bash ./configure [options]
configurewill try to figure out what system you are running on and where all necessary build components are. If you have all prerequisites for building installed, it should find everything. If it fails to detect any component automatically, it will exit and inform you about the problem. When this happens, read more below in the
Description Configure Command Line Windows 32bit build with freetype specified
bash ./configure --with-freetype=/cygdrive/c/freetype-i586 --with-target-bits=32
Debug 64bit Build
bash ./configure --enable-debug --with-target-bits=64
Configure OptionsComplete details on all the OpenJDK
configureoptions can be seen with:Use
bash ./configure --help=short
-helpto see all the
configureoptions available. You can generate any number of different configurations, e.g. debug, release, 32, 64, etc. Some of the more commonly used
OpenJDK Configure Option Description
set the debug level to fastdebug (this is a shorthand for
select the location of the Advanced Linux Sound Architecture (ALSA)
Version 0.9.1 or newer of the ALSA files are required for building the OpenJDK on Linux. These Linux files are usually available from an "alsa" of "libasound" development package, and it's highly recommended that you try and use the package provided by the particular version of Linux that you are using.
select the Bootstrap JDK
provide the JVM options to be used to run the Bootstrap JDK
select the path to the cacerts file.
See http://en.wikipedia.org/wiki/Certificate_Authority for a better understanding of the Certificate Authority (CA). A certificates file named "cacerts" represents a system-wide keystore with CA certificates. In JDK and JRE binary bundles, the "cacerts" file contains root CA certificates from several public CAs (e.g., VeriSign, Thawte, and Baltimore). The source contain a cacerts file without CA root certificates. Formal JDK builders will need to secure permission from each public CA and include the certificates into their own custom cacerts file. Failure to provide a populated cacerts file will result in verification errors of a certificate chain during runtime. By default an empty cacerts file is provided and that should be fine for most JDK developers.
select the CUPS install location
The Common UNIX Printing System (CUPS) Headers are required for building the OpenJDK on Solaris and Linux. The Solaris header files can be obtained by installing the package SFWcups from the Solaris Software Companion CD/DVD, these often will be installed into the directory
The CUPS header files can always be downloaded from www.cups.org.
select the CUPS include directory location
select the debug information level of release, fastdebug, or slowdebug
select location of the compiler install or developer install location
select location of the Windows Direct X SDK install
The Microsoft DirectX 9.0 SDK header files and libraries from the Summer 2004 edition are required for building OpenJDK. This SDK can be downloaded from Microsoft DirectX 9.0 SDK (Summer 2004). If the link above becomes obsolete, the SDK can be found from the Microsoft Download Site (search with "DirectX 9.0 SDK Update Summer 2004"). Installation usually will set the environment variable
DXSDK_DIRto it's install location.
select the freetype files to use.
Expecting the freetype libraries under
lib/and the headers under
Version 2.3 or newer of FreeType is required. On Unix systems required files can be available as part of your distribution (while you still may need to upgrade them). Note that you need development version of package that includes both the FreeType library and header files.
You can always download latest FreeType version from the FreeType website.
Building the freetype 2 libraries from scratch is also possible, however on Windows refer to the Windows FreeType DLL build instructions.
Note that by default FreeType is built with byte code hinting support disabled due to licensing restrictions. In this case, text appearance and metrics are expected to differ from Sun's official JDK build. See the SourceForge FreeType2 Home Page for more information.
select the location to find hotspot binaries from a previous build to avoid building hotspot
select 32 or 64 bit build
select the JVM variants to build from, comma separated list that can include: server, client, kernel, zero and zeroshark
select the RAM size that GNU make will think this system has
msvcr100.dllfile to include in the Windows builds (C/C++ runtime library for Visual Studio).
This is usually picked up automatically from the redist directories of Visual Studio 2010.
select the number of cores to use (processor count or CPU count)
select the location of the X11 and xrender files.
The XRender Extension Headers are required for building the OpenJDK on Solaris and Linux.
The Linux header files are usually available from a "Xrender" development package, it's recommended that you try and use the package provided by the particular distribution of Linux that you are using.
The Solaris XRender header files is included with the other X11 header files in the package SFWxwinc on new enough versions of Solaris and will be installed in
MakeThe basic invocation of the
makeutility looks like:This will start the build to the output directory containing the "configuration" that was created by the
make helpfor more information on the available targets.
There are some of the make targets that are of general interest:
Make Target Description empty build everything but no images
build everything including images
build all configurations
create complete j2sdk and j2re images
install the generated images locally, typically in
remove all files generated by make, but not those generated by
remove all files generated by both and
configure(basically killing the configuration)
give some help on using
make, including some interesting make targets
When the build is completed, you should see the generated binaries and associated files in the
j2sdk-imagedirectory in the output directory. In particular, the
build/*/images/j2sdk-image/bindirectory should contain executables for the OpenJDK tools and utilities for that configuration. The testing tool
jtregwill be needed and can be found at: the jtreg site. The provided regression tests in the repositories can be run with the command:
cd test && make PRODUCT_HOME=`pwd`/../build/*/images/j2sdk-image all
configurefile looks horrible! How are you going to edit it?
configurefile is generated (think "compiled") by the autoconf tools. The source code is in
configure.acvarious .m4 files in common/autoconf, which are much more readable.
Q: Why is the
configurefile checked in, if it is generated?
A: If it was not generated, every user would need to have the autoconf tools installed, and re-generate the
configurefile as the first step. Our goal is to minimize the work needed to be done by the user to start building OpenJDK, and to minimize the number of external dependencies required.
Q: Do you require a specific version of autoconf for regenerating
A: Currently, no, but this will likely be the case when things have settled down a bit more. (The reason for this is to avoid large spurious changes in
configurein commits that made small changes to
Q: What are the files in
common/makefiles/support/*for? They look like gibberish.
A: They are a somewhat ugly hack to compensate for command line length limitations on certain platforms (Windows, Solaris). Due to a combination of limitations in make and the shell, command lines containing too many files will not work properly. These helper files are part of an elaborate hack that will compress the command line in the makefile and then uncompress it safely. We're not proud of it, but it does fix the problem. If you have any better suggestions, we're all ears! :-)
Q: I want to see the output of the commands that make runs, like in the old build. How do I do that?
A: You specify the
LOGvariable to make. There are several log levels:
warn— Default and very quiet.
info— Shows more progress information than warn.
debug— Echos all command lines and prints all macro calls for compilation definitions.
trace— Echos all $(shell) command lines as well.
Q: When do I have to re-run
A: Normally you will run
configureonly once for creating a configuration. You need to re-run configuration only if you want to change any configuration options, or if you pull down changes to the
Q: I have added a new source file. Do I need to modify the makefiles?
A: Normally, no. If you want to create e.g. a new native library, you will need to modify the makefiles. But for normal file additions or removals, no changes are needed. There are certan exceptions for some native libraries where the source files are spread over many directories which also contain courses for other libraries. In these cases it was simply easier to create include lists rather thane excludes.
Q: When I run
configure --help, I see many strange options, like
--dvidir. What is this?
A: Configure provides a slew of options by default, to all projects that use autoconf. Most of them are not used in OpenJDK, so you can safely ignore them. To list only OpenJDK specific features, use
configureprovides OpenJDK-specific features such as
--with-builddeps-serverthat are not described in this document. What about those?
A: Try them out if you like! But be aware that most of these are experimental features. Many of them don't do anything at all at the moment; the option is just a placeholder. Other depends on pieces of code or infrastructure that is currently not ready for prime time.
Q: How will you make sure you don't break anything?
A: We have a script that compares the result of the new build system with the result of the old. For most part, we aim for (and achieve) byte-by-byte identical output. There are however technical issues with e.g. native binaries, which might differ in a byte-by-byte comparison, even when building twice with the old build system. For these, we compare relevant aspects (e.g. the symbol table and file size). Note that we still don't have 100% equivalence, but we're close.
Q: I noticed this thing X in the build that looks very broken by design. Why don't you fix it?
A: Our goal is to produce a build output that is as close as technically possible to the old build output. If things were weird in the old build, they will be weird in the new build. Often, things were weird before due to obscurity, but in the new build system the weird stuff comes up to the surface. The plan is to attack these things at a later stage, after the new build system is established.
Q: The code in the new build system is not that well-structured. Will you fix this?
A: Yes! The new build system has grown bit by bit as we converted the old system. When all of the old build system is converted, we can take a step back and clean up the structure of the new build system. Some of this we plan to do before replacing the old build system and some will need to wait until after.
Q: What is @GenerateNativeHeaders?
A: To speed up compilation, we added a flag to javac which makes it do the job of javah as well, as a by-product; that is, generating native .h header files. These files are only generated if a class contains native methods. However, sometimes a class contains no native method, but still contains constants that native code needs to use. The new GenerateNativeHeaders annotation tells javac to force generation of a header file in these cases. (We don't want to generate native headers for all classes that contains constants but no native methods, since that would slow down the compilation process needlessly.)
Q: Is anything able to use the results of the new build's default make target?
A: Yes, this is the minimal (or roughly minimal) set of compiled output needed for a developer to actually execute the newly built JDK. The idea is that in an incremental development fashion, when doing a normal make, you should only spend time recompiling what's changed (making it purely incremental) and only do the work that's needed to actually run and test your code. The packaging stuff that is part of the
imagestarget is not needed for a normal developer who wants to test his new code. Even if it's quite fast, it's still unnecessary. We're targeting sub-second incremental rebuilds! ;-) (Or, well, at least single-digit seconds...)
Q: I usually set a specific environment variable when building, but I can't find the equivalent in the new build. What should I do?
A: It might very well be that we have missed to add support for an option that was actually used from outside the build system. Email us and we will add support for it!
Build Performance Tips
Building OpenJDK requires a lot of horsepower. Some of the build tools can be adjusted to utilize more or less of resources such as parallel threads and memory. The
configurescript analyzes your system and selects reasonable values for such options based on your hardware. If you encounter resource problems, such as out of memory conditions, you can modify the detected values with:
--with-num-cores— number of cores in the build system, e.g.
--with-memory-size— memory (in MB) available in the build system, e.g.
It might also be necessary to specify the JVM arguments passed to the Bootstrap JDK, using e.g.
--with-boot-jdk-jvmargs="-Xmx8G -enableassertions". Doing this will override the default JVM arguments passed to the Bootstrap JDK.
One of the top goals of the new build system is to improve the build performance and decrease the time needed to build. This will soon also apply to the java compilation when the Smart Javac wrapper is making its way into jdk8. It can be tried in the build-infra repository already. You are likely to find that the new build system is faster than the old one even without this feature.
At the end of a successful execution of
configure, you will get a performance summary, indicating how well the build will perform. Here you will also get performance hints. If you want to build fast, pay attention to those!
Building with ccache
A simple way to radically speed up compilation of native code (typically hotspot and native libraries in JDK) is to install ccache. This will cache and reuse prior compilation results, if the source code is unchanged. However, ccache versions prior to 3.1.4 does not work correctly with the precompiled headers used in OpenJDK. So if your platform supports ccache at 3.1.4 or later, we highly recommend installing it. This is currently only supported on linux.
Building on local disk
If you are using network shares, e.g. via NFS, for your source code, make sure the build directory is situated on local disk. The performance penalty is extremely high for building on a network share, close to unusable.
Building only one JVM
The old build builds multiple JVMs on 32-bit systems (client and server; and on Windows kernel as well). In the new build we have changed this default to only build server when it's available. This improves build times for those not interested in multiple JVMs. To mimic the old behavior on platforms that support it, use
Selecting the number of cores to build on
configurewill analyze your machine and run the make process in parallel with as many threads as you have cores. This behavior can be overridden, either "permanently" (on a
--with-num-cores=Nor for a single build only (on a make basis), using
If you want to make a slower build just this time, to save some CPU power for other processes, you can run e.g.
make JOBS=2. This will force the makefiles to only run 2 parallel processes, or even
make JOBS=1which will disable parallelism.
If you want to have it the other way round, namely having slow builds default and override with fast if you're impatient, you should call
--with-num-cores=2, making 2 the default. If you want to run with more cores, run
Solving build problemsIf the build fails (and it's not due to a compilation error in a source file you've changed), the first thing you should do is to re-run the build with more verbosity. Do this by adding
LOG=debugto your make command line.
The build log (with both stdout and stderr intermingled, basically the same as you see on your console) can be found as
build.login your build directory.
You can ask for help on build problems with the new build system on either the build-dev or the build-infra-dev mailing lists. Please include the relevant parts of the build log.
A build can fail for any number of reasons. Most failures are a result of trying to build in an environment in which all the pre-build requirements have not been met. The first step in troubleshooting a build failure is to recheck that you have satisfied all the pre-build requirements for your platform. Scanning the
configurelog is a good first step, making sure that what it found makes sense for your system. Look for strange error messages or any difficulties that
configurehad in finding things.
Some of the more common problems with builds are briefly described below, with suggestions for remedies.
- Corrupted Bundles on Windows:Some virus scanning software has been known to corrupt the downloading of zip bundles. It may be necessary to disable the 'on access' or 'real time' virus scanning features to prevent this corruption. This type of "real time" virus scanning can also slow down the build process significantly. Temporarily disabling the feature, or excluding the build output directory may be necessary to get correct and faster builds.
- Slow Builds:If your build machine seems to be overloaded from too many simultaneous C++ compiles, try setting the
makecommand line. Then try increasing the count slowly to an acceptable level for your system. Also:Creating the javadocs can be very slow, if you are running javadoc, consider skipping that step.
Faster CPUs, more RAM, and a faster DISK usually helps. The VM build tends to be CPU intensive (many C++ compiles), and the rest of the JDK will often be disk intensive.
Faster compiles are possible using a tool called ccache.
- File time issues:If you see warnings that refer to file time stamps, e.g.Warning message:These warnings can occur when the clock on the build machine is out of sync with the timestamps on the source files. Other errors, apparently unrelated but in fact caused by the clock skew, can occur along with the clock skew warnings. These secondary errors may tend to obscure the fact that the true root cause of the problem is an out-of-sync clock.
File `xxx' has modification time in the future.
Clock skew detected. Your build may be incomplete.
If you see these warnings, reset the clock on the build machine, run "
gmake clobber" or delete the directory containing the build output, and restart the build from the beginning.
- Error message:
Trouble writing out table to diskIncrease the amount of swap space on your build machine. This could be caused by overloading the system and it may be necessary to use:to reduce the load on the system.
- Error Message:
libstdc++ not found:This is caused by a missing libstdc++.a library. This is installed as part of a specific package (e.g. libstdc++.so.devel.386). By default some 64-bit Linux versions (e.g. Fedora) only install the 64-bit version of the libstdc++ package. Various parts of the JDK build require a static link of the C++ runtime libraries to allow for maximum portability of the built images.
- Linux Error Message:
cannot restore segment prot after relocThis is probably an issue with SELinux (See http://en.wikipedia.org/wiki/SELinux). Parts of the VM is built without the
-fPICfor performance reasons.
To completely disable SELinux:
$ su root
In the window that appears, select the SELinux tab
Alternatively, instead of completely disabling it you could disable just this one check.
- Select System->Administration->SELinux Management
- In the SELinux Management Tool which appears, select "Boolean" from the menu on the left
- Expand the "Memory Protection" group
- Check the first item, labeled "Allow all unconfined executables to use libraries requiring text relocation ..."
- Windows Error Messages:
*** fatal error - couldn't allocate heap, ...
rm fails with "Directory not empty"
unzip fails with "cannot create ... Permission denied"
unzip fails with "cannot create ... Error 50"
The CYGWIN software can conflict with other non-CYGWIN software. See the CYGWIN FAQ section on BLODA (applications that interfere with CYGWIN).
- Windows Error Message:
spawn failedTry rebooting the system, or there could be some kind of issue with the disk or disk partition being used. Sometimes it comes with a "Permission Denied" message.
The Makefiles in the OpenJDK are only valid when used with the GNU version of the utility command
gmakeon Solaris). A few notes about using GNU make:
- You need GNU make version 3.81 or newer. If the GNU make utility on your systems is not 3.81 or newer, see "Building GNU make".
- Place the location of the GNU make binary in the
- Solaris: Do NOT use
/usr/bin/makeon Solaris. If your Solaris system has the software from the Solaris Developer Companion CD installed, you should try and use
gmakewhich will be located in either the
- Windows: Make sure you start your build inside a bash shell.
- Mac OS X: The XCode "command line tools" must be installed on your Mac.
Information on GNU make, and access to ftp download sites, are available on the GNU make web site . The latest source to GNU make is available at ftp.gnu.org/pub/gnu/make/.
Building GNU makeFirst step is to get the GNU make 3.81 or newer source from ftp.gnu.org/pub/gnu/make/. Building is a little different depending on the OS but is basically done with:
Minimum Build EnvironmentsThis file often describes specific requirements for what we call the "minimum build environments" (MBE) for this specific release of the JDK. What is listed below is what the Oracle Release Engineering Team will use to build the Oracle JDK product. Building with the MBE will hopefully generate the most compatible bits that install on, and run correctly on, the most variations of the same base OS and hardware architecture. In some cases, these represent what is often called the least common denominator, but each Operating System has different aspects to it.
In all cases, the Bootstrap JDK version minimum is critical, we cannot guarantee builds will work with older Bootstrap JDK's. Also in all cases, more RAM and more processors is better, the minimums listed below are simply recommendations.
With Solaris and Mac OS X, the version listed below is the oldest release we can guarantee builds and works, and the specific version of the compilers used could be critical.
With Windows the critical aspect is the Visual Studio compiler used, which due to it's runtime, generally dictates what Windows systems can do the builds and where the resulting bits can be used.
NOTE: We expect a change here off these older Windows OS releases and to a 'less older' one, probably Windows 2008R2 X64.
With Linux, it was just a matter of picking a stable distribution that is a good representative for Linux in general.
NOTE: We expect a change here from Fedora 9 to something else, but it has not been completely determined yet, possibly Ubuntu 12.04 X64, unbiased community feedback would be welcome on what a good choice would be here.
It is understood that most developers will NOT be using these specific versions, and in fact creating these specific versions may be difficult due to the age of some of this software. It is expected that developers are more often using the more recent releases and distributions of these operating systems.
Compilation problems with newer or different C/C++ compilers is a common problem. Similarly, compilation problems related to changes to the
/usr/includeor system header files is also a common problem with older, newer, or unreleased OS versions. Please report these types of problems as bugs so that they can be dealt with accordingly.
Base OS and Architecture OS C/C++ Compiler Bootstrap JDK Processors RAM Minimum DISK Needs Linux X86 (32-bit) and X64 (64-bit) Fedora 9 gcc 4.3 JDK 7u7 2 or more 1 GB 6 GB Solaris SPARC (32-bit) and SPARCV9 (64-bit) Solaris 10 Update 6 Studio 12 Update 1 + patches JDK 7u7 4 or more 4 GB 8 GB Solaris X86 (32-bit) and X64 (64-bit) Solaris 10 Update 6 Studio 12 Update 1 + patches JDK 7u7 4 or more 4 GB 8 GB Windows X86 (32-bit) Windows XP Microsoft Visual Studio C++ 2010 Professional Edition JDK 7u7 2 or more 2 GB 6 GB Windows X64 (64-bit) Windows Server 2003 - Enterprise x64 Edition Microsoft Visual Studio C++ 2010 Professional Edition JDK 7u7 2 or more 2 GB 6 GB Mac OS X X64 (64-bit) Mac OS X 10.7 "Lion" XCode 4.5.2 or newer JDK 7u7 2 or more 4 GB 6 GB
Specific Developer Build EnvironmentsWe won't be listing all the possible environments, but we will try to provide what information we have available to us.
NOTE: The community can help out by updating this part of the document.
FedoraAfter installing the latest Fedora you need to install several build dependencies. The simplest way to do it is to execute the following commands as user
yum install gcc gcc-c++
In addition, it's necessary to set a few environment variables for the build:
CentOS 5.5After installing CentOS 5.5 you need to make sure you have the following Development bundles installed:
- Development Libraries
- Development Tools
- Java Development
- X Software Development (Including XFree86-devel)
Plus the following packages:
- cups devel: Cups Development Package
- alsa devel: Alsa Development Package
- Xi devel: libXi.so Development Package
The freetype 2.3 packages don't seem to be available, but the freetype 2.3 sources can be downloaded, built, and installed easily enough from the freetype site. Build and install with something like:
sudo -u root make install
Mercurial packages could not be found easily, but a Google search should find ones, and they usually include Python if it's needed.
Debian 5.0 (Lenny)After installing Debian 5 you need to install several build dependencies. The simplest way to install the build dependencies is to execute the following commands as user
aptitude build-dep openjdk-7
aptitude install openjdk-7-jdk libmotif-dev
In addition, it's necessary to set a few environment variables for the build:
Ubuntu 12.04After installing Ubuntu 12.04 you need to install several build dependencies. The simplest way to do it is to execute the following commands:
sudo aptitude build-dep openjdk-7
sudo aptitude install openjdk-7-jdk
In addition, it's necessary to set a few environment variables for the build:
OpenSUSE 11.1After installing OpenSUSE 11.1 you need to install several build dependencies. The simplest way to install the build dependencies is to execute the following commands:
sudo zypper source-install -d java-1_7_0-openjdk
sudo zypper install make
In addition, it is necessary to set a few environment variables for the build:
Finally, you need to unset the
export -n JAVA_HOME
Mandriva Linux One 2009 SpringAfter installing Mandriva Linux One 2009 Spring you need to install several build dependencies. The simplest way to install the build dependencies is to execute the following commands as user
urpmi java-1.7.0-openjdk-devel make gcc gcc-c++ freetype-devel zip unzip libcups2-devel libxrender1-devel libalsa2-devel libstc++-static-devel libxtst6-devel libxi-devel
In addition, it is necessary to set a few environment variables for the build:
OpenSolaris 2009.06After installing OpenSolaris 2009.06 you need to install several build dependencies. The simplest way to install the build dependencies is to execute the following commands:
pfexec pkg install SUNWgmake SUNWj7dev sunstudioexpress SUNWcups SUNWzip SUNWunzip SUNWxwhl SUNWxorg-headers SUNWaudh SUNWfreetype2
In addition, it is necessary to set a few environment variables for the build:
End of OpenJDK README-builds.html document.
Please come again!