- 1 始める前の注意
- 2 LibreOfficeビルドに必要な依存関係の解決
- 3 ソースコードのクローンとLibreOfficeのビルド
- 4 ビルドしたバイナリの実行
- 5 ビデオチュートリアル
- 6 ヒント
- 6.1 デバッガーでの実行
- 6.2 警告を有効にする
- 6.3 パフォーマンス
- 6.4 makeのメッセージを冗長にする
- 6.5 複数の作業ディレクトリで作業する
- 6.6 rpmが見つかりません/antが見つかりません/gnome-vfs-2.0パッケージがありません
- 6.7 DEBまたはRPMパッケージのビルド
- 6.8 改行コードとGitのautcrlfの処理
- 6.9 Pootleの.poファイルを使って翻訳されたUIをビルドする
Since this turns out to be the common problem with newbies, please note: do not build as root! This will break your build, and as the result, you'll need to cleanup everything (including ccache) and start over. The steps that need elevation use sudo explicitly in the following explanation.
Do also read our generic building hints.
- Additional information may be found on the page Linux Build Dependencies.
In general, the easiest way to build LibreOffice is on Linux.
First you need to install the build dependencies:
- On Debian/Ubuntu:
sudo apt-get build-dep libreoffice
- On OpenSUSE 11.4+:
sudo zypper si -d libreoffice
- On Fedora 22+ & derivatives:
sudo dnf builddep libreoffice
- On RHEL 7+ & derivatives:
sudo yum-builddep libreoffice
- On DragonFly
cd /usr/dports/editors/libreoffice && make patch
- On FreeBSD
cd /usr/ports/editors/libreoffice && make patch
- On NetBSD
cd /usr/pkgsrc/misc/libreoffice && make depends
Note that these commands might not install everything that is necessary, because they use the distro's package as a source of dependencies. That means they reflect the version of LibreOffice that is packaged by the distro and the options that are used for building it.
For more details or information for other distros such as Arch, see the page on build dependencies on Linux.
clone the repository and start to build:
$ git clone https://gerrit.libreoffice.org/core libreoffice $ cd libreoffice $ ./autogen.sh
At this point, you might want to consider additional options to pass to autogen with
$ ./autogen.sh --help
For example consider adding a Debugging option. Changing some of these options will require to do a full rebuild: e.g., adding
--enable-dbgutil. Since the following command
make takes a vast quantity of time to run for the first time, you probably would like to save time by specifying the settings you'd need depending on the ultimate goal of your build. In case of doubt, possibly it's good to ask for advise on #libreoffice-dev.
git clone takes a while too.
This will create you a good local installation, which you can then start with:
$ instdir/program/soffice --writer $ instdir/program/soffice --calc
While this introduction was done on an Ubuntu system, it should work with little modification on any Linux system (see below for details, if needed). We would recommend to watch the video along with performing the steps – it not only walks you through the setup, it also has an excellent soundtrack! ;)
For starting your build directly under a debugger:
$ make debugrun
Then you can open Writer with the command:
(gdb) run --writer
Note that soffice.bin will be terminated the first time it is run directly (or via make debugrun), this is normal, you simply need to run it again to make use of it. Running soffice will deal with this for you.
Various versions of GCC are well-known to emit more/unhelpful/bogus warnings at higher optimization levels. Don't even try to combine
build with GCC.
Building LibreOffice takes time, a lot of time. Exactly how much depends on how powerful your machine is. But there are tools you can use to speed-up things.
ccache is a tool that caches the results of C/C++ compilation to try to re-use them later. It will not speed-up your initial build (on the contrary it will slow it down a bit), but it can dramatically speed up later re-building. If you plan to change lots of header files in LibreOffice, it is a worthy tool to have.
By default, ccache will be enabled automatically if it is found on your system. For best results, you may want to increase the cache size or enable cache compression; `man ccache` is your friend.
Note that the default cache size is just 5GB which is far too small to be useful for a LibreOffice build; for a build without debug symbols, you should have at least 8GB cache, and for a build with debug symbols for everything probably 32GB is a useful starting point. You can set the size like this (note that this setting is stored inside the cache directory, if you change the location by setting
$CCACHE_DIR you have to re-do it):
ccache --max-size 32G
ccache also supports compression, which is very likely a good idea to enable (does anybody have hard numbers?), especially if you are using a small or slow storage medium, like a SSD or laptop hard disk. You can enable compression by adding this to your
.bashrc or equivalent:
Icecream / distcc
If you are in an environment where you have access to multiple machines with spare cpu cycles, you can take a look at icecream or distcc tools for distributed build. Icecream is recommended and easier to setup.
Support for Icecream is built into LibreOffice, it is enough to add
./autogen.sh, and the configure process will pre-set number of jobs to use, and will try to find the icecream gcc wrappers in
/usr/lib/icecc/bin or in
/opt/icecream/bin. Should you have them somewhere else, please use
--with-gcc-home=/path/to/your/icecream/bin switch to override that.
To make use of distcc, first install and start
distccd on all the build machines and make sure all the compiler versions are compatible (this is important for C++ code). Now on the driving machine install and configure
distcc to use the other build hosts. Then pass
--with-parallelism=N to autogen.sh, where N is the total number of CPUs available in your cluster and run
distcc-pump make CC=distcc CXX=distcc to start the distributed build.
--enable-pch to autogen.sh enables use of precompiled headers (PCH), which reduce build time at the expense of more disk space and more source files possibly rebuilt when a header changes. The option has several levels (system, base, normal, full), the higher level the higher possible gain but also disk space usage and chance for larger rebuilds on changes.
The build process can be told to run multiple tasks in parallel. The parallelism is controlled by the autogen.sh parameter --with-parallelism. If you have enough memory, I found that using --with-parallelism=n where n is your number of cores, or 2 for --with-parallelism if you have only 1 core, give me the fastest build time.
--with-parallelism already defaults to the the number of cores/cpus on your system, unless you use --enable-icecream - then to 10.
Building with existing libraries may speed up your build, but you have to fiddle with the dependencies.
The first build or a clean build takes approximately
8/number_cores hours on a recent processor. With ccache, subsequent daily build time varies between 2 and 10 min (depending on the number of modified or added files and their complexity).
Make will hide the exact command invokations. To see what make does exactly for a single run one can invoke it as
$ make GMAKE_OPTIONS='VERBOSE=1'
If you want to work on both the master branch, and the current release branch(es) at the same time, you can share git repos, and external source tarballs (note the savings can be significant, especially if you also use the quite huge l10n repo). Note that switching between different release branches within the same working dir is not recommended; even if there are no dependency problems, the amount of stuff you'd have to rebuild is practically equal to a clean rebuild. If you can afford the diskspace, maintaining separate trees is the recommended way.
There are two methods to share git repos across working directories: git clone --reference and git-new-workdir. A referenced clone can avoid re-downloading the repo, but the new workdir will still have its own .git/config, etc. This is supported by upstream git developers. git-new-workdir, however, is not supported and can cause problems if you try to check out the same branch in two working directories, etc.
Prepare the first build on your disk just like explained above. For the second, and all other builds, do this to clone the initial core repo:
$ git clone --reference /path/to/master --branch name-of-branch ssh://logerrit/core /dir/to/be/created
You'll notice a much quicker clone operation. After that, setup the new work dir with these two extra configure (or autogen.sh) options:
$ ./autogen.sh ... --with-referenced-git=/path/to/master --with-external-tar=/path/to/master/external/tarballs
Again, cloning will be much faster. You have to use an absolute path for the --with-external-tar option. After that, you proceed with building just like for the single-workdir case.
The --with-referenced-git option is only needed if you enabled some submodules (translations, dictionaries or help) or you're building the libreoffice-4-0 branch (or one that is even older). On newer branches submodules are disabled by default.
If you want to cherry-pick between the working directories, you can setup your local copy of git to act like a remote, like this:
$ cd /path/to/stable-workingcopy $ git remote add unstable /path/to/master/.git $ cd /path/to/master $ git remote add stable /path/to/stable-workingcopy/.git
You then just cherry-pick commits like you would to a "true" remote on another server.
autogen.sh on master with the specific parameters you need, then run
make fetch to fetch the submodules that may be needed depending on the parameters you passed to autogen.sh
then switch branch with
$ ./g checkout -b libreoffice-4-0 origin/libreoffice-4-0
--disable-epm This will fix "rpm not found", as autogen.sh expects to find either dpkg or rpm on the system.
--without-java This will fix "ant not found"
As of the master towards LibreOffice 3.6, to build the UNO SDK you need to have Doxygen installed so that HTML documentation for the C++ UNO interface can be generated. One option is to install Doxygen from its website and make sure the
doxygen executable is found on the
PATH, or specify its exact location via
--with-doxygen=pathname configure switch. Another option is to configure
--without-doxygen, in which case the HTML documentation for the C++ UNO interface will be missing from the UNO SDK being built. A third option is to configure
--disable-odk, in which case the UNO SDK is not built at all.
Note: Installing 'libgnomevfs2-dev' package on Ubuntu manually will fix "No package 'gnome-vfs-2.0' found" on older release branches. It should not be needed anymore on master by now. Use synaptic package manager or run in terminal
$ sudo apt-get install libgnomevfs2-dev
If you are wanting to build DEB or RPM packages, this should do the trick...
- RPM and DEB
$ ./autogen.sh --with-distro=LibreOfficeLinux --with-package-format="deb rpm" --enable-epm $ make
$ ./autogen.sh --with-distro=LibreOfficeLinux --with-package-format=rpm --enable-epm $ make
$ ./autogen.sh --with-distro=LibreOfficeLinux --with-package-format=deb --enable-epm $ make
Look for the files in the ./workdir/installation/ directory. There is a folder for each of the different sets of packages that includes subfolder[s] for the type or types of packages, either DEB, RPM, or both. Inside the subfolder[s] is another folder called install, and it includes two subfolders. One subfolder contains an archive of all of the packages zipped together (the ..._download folder) and the other contains a DEBS/RPMS folder with each individual package. The rest should be self-explanitory!
If autogen.sh produces output like the following:
./autogen.sh: line 1: $':\r': command not found ' '--srcdir=/home/user/libreoffice' '--enable-option-checking=fatal' configure: WARNING: you should use --build, --host, --target configure: WARNING: invalid host type: ******************************************************************** * * Running LibreOffice build configuration. * ******************************************************************** ' not recognizedystem type... Invalid configuration ` failedre: error: /bin/bash ./config.sub Error running configure at ./autogen.sh line 241.
... then you probably have line ending problems. This might occur if you mistakenly set Git's
To fix the issue, run these commands (WARNING: uncommitted changes will be lost!):
git config --unset core.autocrlf git rm --cached -r . git reset --hard git clean -x -f ./autogen.sh
LibreOffice's user interface is translated using .po files hosted on Pootle server and translations are update and packaged as language packs with every new release for every branch respectively. That often means language pack is not reflecting all the improvements made in-between releases and nightly builds exists only for a handful of languages. Translators might benefit from testing their work before new release been published and but early adopters could test translations before official release.
- Firstly you need to download .po files from Pootle. You might log-in to Pootle to download .po files.
- rebuild the source with language pack option on using command
./autogen.sh --with-lang="<langCode>"and don't forget to replace <langCode> with actual language code. If using an autogen.input file, no quotes are used, so this becomes: --with-lang=de ja ar
- extract downloaded .po files and copy content to relevant directories in source tree using command (on GNU/Linux)
cp -R <langCode>/libo_ui/* libreoffice/translations/source/<langCode>/and don't forget to replace
<langCode>(two occurrences) with actual language code
- this step might be tricky. Seams that some extracted .po files are named 'messages.po' and some are named differently. What you need is every .po file converted to .mo file and
- for files that are **not named** 'messages.po', conversion is done by running
- for files that **are** named 'messages.po' you need to convert to .mo files manually running
msgfmt -o instdir/program/resource/<langCode>/LC_MESSAGES/<moFile> <dir>/messages.pofor every such file. Here three things must be changed: <langCode> to actuall language code, <moFile> with data from left column of the table bellow and <dir> with data from right side of table bellow. Just make sure you prefix path '<dir>/messages.po' with full path from current working to destination.
- for files that are **not named** 'messages.po', conversion is done by running
As mentioned in step 4, some .po files are named 'messages.po' and some are named differently. Table bellow lists directories which might hold 'messages.po' files and on which step 4.2 should be applied.
Last step in process is to run
sudo make install to install with newly added translations.
If you would like to build LibreOffice-master and check the translation, you may need "make translations" to merge .po file with the .pot files.
Help files are also translated on Pootle so for translated Help file this can be done
- by downloading .po files of translated Help
./autogen.sh --with-help --with-lang="<langCode>"replacing
<langCode>with actual language code. If you want to build Help as HTML files, do
--with-help=HTML. If using an autogen.input file, no quotes are used, so this becomes: --with-lang=de ja ar
cp -fR <langCode>/libo_help/* libreoffice/translations/source/<langCode>/helpcontent2/source/text/, but remember to replace two instances of <langCode> with actual language code and make sure you navigate to this folder correctly from current working directory
mv libreoffice/translations/source/<langCode>/helpcontent2/source/text/auxiliary.po libreoffice/translations/source/<langCode>/helpcontent2/source/. Replace two instances of <langCode> with actual language code and make sure you navigate to this folder correctly from current working directory
sudo make install
(Thanks to Kiyotaka and Jihui)