wolny / ilastik-build-conda

Conda recipes for building ilastik dependencies.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

ilastik-build-conda

ilastik depends on 60+ packages. Most of those packages are already provided for us by the Anaconda Python distribution. For 20+ the packages that aren't provided by Anaconda, we use the recipes in this repo.

These recipes are built using the conda-build tool. The resulting binaries are uploaded to the ilastik anaconda channel, and can be installed using the conda package manager.

Contents

Installing ilastik for development

Preamble: Depending on what you are trying to do, you may not need to follow any of these steps. The ilastik binary is shipped with a complete conda environment, .py files, etc. For many purposes, simply downloading the binary and editing the (python) source code by hand may suffice. However, these instructions can give you a more complete developer setup, suitable e.g. for C++ development.

Here's how to install everything you need to develop ilastik.

0. Prerequisite: Install Miniconda

# Install miniconda to the prefix of your choice, e.g. /my/miniconda

# LINUX:
wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

# MAC:
wget https://repo.continuum.io/miniconda/Miniconda3-latest-MacOSX-x86_64.sh
bash Miniconda3-latest-MacOSX-x86_64.sh

# Activate conda
CONDA_ROOT=`conda info --root`
source ${CONDA_ROOT}/bin/activate root

NOTE:

When using conda, make sure you are not using any of python's site-specific or user-specific customization features. In particular, make sure the following environment variables are not defined in your terminal:

  • PYTHONPATH
  • PYTHONUSERSITE
  • PYTHONUSERBASE

Also, make sure there are no python-related directories in ~/.local/.

1. Create a fresh environment, and install ilastik

Some ilastik workflows require commercial solvers, for which one must purchase or obtain an academic license. If you don't have CPLEX and Gurobi on your machine, you can install everything else with this command:

conda create -n ilastik-devel ilastik-dependencies-no-solvers -c ilastik-forge -c conda-forge

If you have both CPLEX and Gurobi on your machine, you can install the full ilastik development setup, including full support for tracking and multicut.

First, define these environment variables:

export CPLEX_ROOT_DIR=/path/to/ibm/ILOG/CPLEX_Studio1251
export GUROBI_ROOT_DIR=/path/to/gurobi650/linux64

Now you can install the ilastik-dependencies package:

conda create -n ilastik-devel ilastik-dependencies -c ilastik-forge -c conda-forge

Note: To be really sure that you're getting the right version of ilastik-dependencies, you can require a specific version and build of the package with PKGNAME=VERSION=BUILD syntax:

conda create -n ilastik-devel ilastik-dependencies=1.2.0=6 -c ilastik-forge -c conda-forge

If you only have one of CPLEX or Gurobi, and you're seeking to develop for a workflow that requires it, you must install some dependencies of that workflow individually. For example, to install tracking with CPLEX, but not Gurobi:

conda create  -n ilastik-devel install ilastik-dependencies-no-solvers -c ilastik-forge -c conda-forge
conda install -n ilastik-devel multi-hypotheses-tracking-with-cplex -c ilastik-forge -c conda-forge

For example, to install multicut with Gurobi support:

conda create  -n ilastik-devel install ilastik-dependencies-no-solvers -c ilastik-forge -c conda-forge
conda install -n ilastik-devel nifty-with-gurobi -c ilastik-forge -c conda-forge

2. Run ilastik

${CONDA_ROOT}/envs/ilastik-devel/run_ilastik.sh --debug

3. (Optional) Clone ilastik git repo

So far, our environment contains the ilastik source, but not the git repos. If you need to edit the ilastik python code, replace the ilastik-meta directory with the full git repo.

Note: This will remove both ilastik-meta and ilastik-dependencies, but all of the other dependencies in your environment will remain.

CONDA_ROOT=`conda info --root`
DEV_PREFIX=${CONDA_ROOT}/envs/ilastik-devel
conda remove -n ilastik-devel ilastik-meta

# Re-install ilastik-meta.pth
cat > ${DEV_PREFIX}/lib/python2.7/site-packages/ilastik-meta.pth << EOF
../../../ilastik-meta/lazyflow
../../../ilastik-meta/volumina
../../../ilastik-meta/ilastik
EOF

# Option 1: clone a fresh copy of ilastik-meta
git clone http://github.com/ilastik/ilastik-meta ${DEV_PREFIX}/ilastik-meta
cd ${DEV_PREFIX}/ilastik-meta
git submodule update --init --recursive
git submodule foreach "git checkout master"

# Option 2: Symlink to a pre-existing working copy, if you have one.
cd ${DEV_PREFIX} && ln -s /path/to/ilastik-meta

Generating a release binary

  1. (Prerequisite) Update the version number.

  2. Edit ilastik.__version_info__ (in ilastik/__init__.py) and commit your change.

  3. Commit to ilastik-meta and add a matching git tag:

     cd ${DEV_PREFIX}/ilastik-meta
     git commit -m "Alpha Release 1.2.3a4" ilastik lazyflow volumina
     git push origin master
     git tag -m "Alpha Release" -a 1.2.3a4
     git push --tags origin
    
  4. Double-check your conda configuration (.condarc). You should allow access to the ilastik channel and defaults, but nothing else:

     $ cat ~/.condarc
     channels:
     - ilastik
     - defaults
    
  5. Build ilastik-meta and ilastik-dependencies packages, and upload to the ilastik anaconda channel.

     WITH_SOLVERS=1 conda build ilastik-meta ilastik-dependencies
     anaconda upload -u ilastik ${CONDA_ROOT}/conda-bld/linux-64/ilastik-meta*.tar.bz2
     anaconda upload -u ilastik ${CONDA_ROOT}/conda-bld/linux-64/ilastik-dependencies*.tar.bz2
    

Troubleshooting Tip: If the ilastik-meta tag has been relocated since you last built the ilastik-meta package, you should probably clear conda's git cache for that repo, to ensure you have the new tags: rm -rf $(conda info --root)/conda-bld/git_cache/github.com/ilastik/ilastik-meta

  1. (Optional) Install to a local environment and test

     conda create -n test-env ilastik-dependencies=1.2.3a4 -c ilastik-forge -c conda-forge
     cd ${CONDA_ROOT}/envs/test-env
     ./run_ilastik.sh
    
  2. Create tarball/app

    Linux:

         $ grep Usage ./create-tarball.sh
         ## Usage: create-tarball.sh [--skip-tar] [--git-latest] [--no-tracking] [... extra install-args, e.g. --use-local or -c ilastik ...]
         
         $ ./create-tarball.sh -c ilastik-forge -c conda-forge
    

    Mac:

         $ grep Usage ./osx-packages/create-osx-app.sh
         ## Usage: create-osx-app.sh [--compress] [--git-latest] [--no-tracking] [... extra install-args, e.g. --use-local or -c ilastik ...]
         
         $ ./osx-packages/create-osx-app.sh --compress -c ilastik-forge -c conda-forge
    

    If any options are used in the Linux or Mac binary creation scripts above, they must be passed in this order:

    • --skip-tar: (Linux only) Create the ilastik-release environment, but don't compress it into a .tar.bz2 file.
    • --compress: (Mac only) After creating the .app bundle, compress it into a .tar.bz2 file.
    • --git-latest: Use the latest master branch of ilastik, lazyflow, and volumina instead of the most recent tag. (Don't use for official releases.)
    • --no-tracking: Omit tracking-specific dependencies
    • --use-local: Tells conda to use your custom builds of each package, if available.
    • -c ilastik-forge: Tells conda to use packages from the ilastik-forge channel (in case it's missing from ~/.condarc).

    Windows:

         ## create new environment for packaging and activate it
         $ conda create -n ilastik-release ilastik-dependencies -c ilastik-forge -c conda-forge
         $ activate ilastik-release
    
         ## install exe and preconfigured installer generation script
         $ conda install ilastik-exe ilastik-package -c ilastik-forge -c conda-forge
         
         ## build the installer using Inno setup
    
         ## ACTION REQUIRED: open the file ${YOUR_CONDA_ENV_PATH}/package/ilastik.iss in Inno Setup 
         ## and build the installer.
    
         ## delete environment
         $ activate root
         $ conda env remove -n ilastik-release
    

How to build these packages yourself

Note: see https://github.com/ilastik/ilastik-publish-packages for an automated way of building all packages required by ilastik

Warning: the description below is outdated

All of the recipes in this repo should already be uploaded to the ilastik anaconda channel. The linux packages were built on CentOS 5.11, so they should be compatible with most modern distros. The Mac packages were built with MACOSX_DEPLOYMENT_TARGET=10.7, so they should theoretically support OSX 10.7+.

But if, for some reason, you need to build your own binary packages from these recipes, it should be easy to do so:

# Prerequisite: Install conda-build
source activate root
conda install conda-build

# Clone the ilastik build recipes
git clone http://github.com/ilastik/ilastik-build-conda
cd ilastik-build-conda

# Build a recipe, e.g:
conda build --numpy=1.11 vigra

# Now install your newly built package, directly from your local build directory:
conda install --use-local -n ilastik-devel vigra

Now run ilastik from with your ilastik-meta repo:

cd /path/to/ilastik-meta

# Run ilastik
PYTHONPATH="ilastik:lazyflow:volumina" python ilastik/ilastik.py

As mentioned above, some packages require CPLEX and Gurobi. To build those packages, you must define some environment variables first:

# Configure environment for building with solvers active
export WITH_SOLVERS=1
export CPLEX_ROOT_DIR=/path/to/ibm/ILOG/CPLEX_Studio1251
export GUROBI_ROOT_DIR=/path/to/gurobi650/linux64

# Build some recipes that depend on solvers
conda build ilastik-dependencies

Appendix: Writing a new recipe

The conda documentation explains in detail how to create a new package, but here's a quick summary:

0. Prerequisite: Install conda-build

source activate root
conda install conda-build

1. Create recipe files

Add a directory to this repo:

cd ilastik-build-conda
mkdir somepackage
cd somepackage

A complete recipe has at least 3 files:

  • meta.yaml
  • build.sh (used for both Mac and Linux)
  • bld.bat (used for Windows)

...additional files (such as patches) may be needed for some recipes.

Write meta.yaml:

$ cat > meta.yaml
package:
  name: somepackage
  version: 1.2.3

source:
  fn: somepackage-1.2.3.tar.bz2
  url: http://www.randompackages.org/somepackage/somepackage-1.2.3.tar.bz2
  md5: b060bb137d6bd8accf8f0c4c59d2746d

requirements:
  build:
    - zlib
    - python
  run:
    - zlib
    - python

about:
  home: http://www.somepackage.com
  license: WYSIWYG v3

Write build.sh:

$ cat > build.sh
# configure, make, and install
configure --prefix=$PREFIX --with-zlib=$PREFIX
make -j${CPU_COUNT}
make install

Write bld.bat:

$ cat > bld.bat
mkdir build
cd build

REM Configure step
cmake -DCMAKE_PREFIX_PATH=%LIBRARY_PREFIX% -DCMAKE_INSTALL_PREFIX=%LIBRARY_PREFIX% -DCMAKE_BUILD_TYPE=Release %SRC_DIR%

REM Build step
devenv SomePackage.sln /Build "%RELEASE_TARGET%"
if errorlevel 1 exit 1

REM Install step
devenv SomePackage.sln /Build "%RELEASE_TARGET%" /Project INSTALL
if errorlevel 1 exit 1

2. Build the package

# Switch back to the `ilastik-build-conda` directory
$ cd ../

# Build the package
$ conda build somepackage

3. Upload the package to your anaconda channel.

conda install anaconda-client

# Upload to your personal channel:
anaconda upload /my/miniconda/conda-bld/osx-64/somepackage-1.2.3-0.tar.bz2

# Or to ilastik's anaconda channel:
anaconda upload -u ilastik /my/miniconda/conda-bld/osx-64/somepackage-1.2.3-0.tar.bz2

Appendix: Compiler details

When writing your own recipes, use gcc provided by conda.

Instead of using your system compiler, all of our C++ packages use the gcc package provided by conda itself (or our own variation of it). On Mac, we use LLVM's clang instead to get C++11 features. On Linux, using conda's gcc-4.8 is an easy way to get C++11 support on old OSes, such as our CentOS 5.11 build VM.

To use the gcc package, add these requirements to your meta.yaml file:

requirements:
  build:
    - gcc 4.8.5 # [linux]
  run:
    - libgcc # [linux]

And in build.sh, make sure you use the right gcc executable. For example:

export CC=${PREFIX}/bin/gcc
export CXX=${PREFIX}/bin/g++

# conda provides default values of these on Mac OS X,
# but we don't want them when building with gcc
export CFLAGS=""
export CXXFLAGS=""
export LDFLAGS=""

./configure --prefix=${PREFIX} ...etc...
make
make install

# Or, for cmake-based packages:
mkdir build
cd build
cmake .. \
    -DCMAKE_C_COMPILER=${PREFIX}/bin/gcc \
    -DCMAKE_CXX_COMPILER=${PREFIX}/bin/g++ \
    -DCMAKE_INSTALL_PREFIX=${PREFIX} \
    ...etc...

make
make install

Appendix: Linux VM Details

The Anaconda distribution is built on a CentOS 5.11 VM. To build the ilastik stack on that OS, you'll need to install the following:

Appendix: TODO/TBD

  • General

  • In cases where we provide an alternative build of a package that Continuum already provides, we need to make sure our special channel takes priority over the defaults channel used by conda. (Edit: Ideally, we could just use a custom "build string", but due to conda/conda#918, that doesn't work. Instead, we just use a deliberately strange version number in our custom packages, e.g. version: 5.10.1.99.)

  • It would be nice if we built "debug" versions of important packages (e.g. Python, vigra, Qt) and attached them to the [debug] conda-build "feature".

  • The final binaries produced via create-Linux-tarball.sh and create-osx-app.sh are quite large. They could be reduced by excluding unecessary dylibs and stripping the remaining dylibs. Also, directories like include, etc. should be excluded.

  • Mac

  • For unknown reasons, the py2app module does not work "out-of-the-box" for this conda build. (The resulting app crashes frequently.) It probably has something to do with our new dependency on gcc-4.8 and libgcc. The current version of create-osx-app.sh uses a hacky workaround for this issue. It would be nice if we could figure out what the real issue is.

  • Windows

  • So far, this repo includes no package build scripts for Windows.

  • Generate a final binary package from the built dependencies

  • Should we attempt to track different versions of the MSVC++ std library via a conda "feature"?

About

Conda recipes for building ilastik dependencies.


Languages

Language:Shell 42.6%Language:Python 42.0%Language:Batchfile 11.3%Language:C++ 2.8%Language:C 1.0%Language:CMake 0.3%