This package contains a binary and supporting libraries for doing extraction
on repos using Kythe. It is intended as the last step where extraction is
actually performed by invoking the builds, along with any immediate
preprocessing required.
runextractor
is an executible that is used as the inner-most entrypoint for
Kythe Extraction. This binary is intended to encapsulate any logic required for
extracting that is common to the build-system. So for example any configuration
that is on a per-repo basis should be handled upstream, not in this binary.
runextractor
is expected to be run from the root of a repository, so that
any access to config files (gradle.build
, pom.xml
, etc) is sensible, and
also so that execution of build/compile commands works.
Use:
./runner \
--builder=MAVEN \
--mvn_pom_preprocessor=/opt/kythe/extractors/javac_extractor.jar \
--javac_wrapper=/opt/kythe/extractors/javac-wrapper.sh
Environment Variables
Because runextractor
is invoked in a manner not conducive to cleanly passing
commandline flags, there is non-trivial setup done with environment variables.
When calling runextractor
, here are the required environment variables:
- KYTHE_ROOT_DIRECTORY: The absolute path for file input to be extracted.
- KYTHE_OUTPUT_DIRECTORY: The absolute path for storing output.
- KYTHE_CORPUS: The corpus label for extracted files.
For Java there are other required env vars:
- JAVAC_EXTRACTOR_JAR: A absolute path to a jar file containing the java
extractor.
- REAL_JAVAC: A path to a "normal" javac binary (not a wrapped binary).
And an optional env var:
- KYTHE_OUTPUT_FILE: If you specify an output file ending in
.kzip
, any
compilation units will be output into that file. Note that for now,
KYTHE_OUTPUT_DIRECTORY is still required for some incidental things in
the Java extractor, so that must still be specified.
TODO(#156): If we can get rid of docker in docker, working dir relative paths
might be easier to work with.
We support Kythe extraction on a few different build systems.
Gradle
In order to run Kythe extraction on a gradle repo, we must first modify the
gradle.build
file to hook into a separate javac wrapper binary.
gradle_build_modifier.go
takes an input gradle.build
file and appends the
bits necessary for replacing javac calls with Kythe's javac-wrapper.sh
.
allprojects {
gradle.projectsEvaluated {
tasks.withType(JavaCompile) {
options.fork = true
options.forkOptions.executable = '/opt/kythe/extractors/javac-wrapper.sh'
}
}
}
If the input file already contains reference to
options.forkOptions.executable
, then gradle_build_modifier.go
does nothing.
Future work
The current implementation uses simple string-based matching, without actually
understanding the structure. If that becomes necessary in the future, it might
be better to use the existing Java libraries for org.codehaus.groovy.ast
to
properly parse the build.gradle file and have more precise picture. In
particular org.codehaus.groovy.ast.CodeVisitorSupport
might be sufficient.
Maven
Maven's build config is handled in Java by mavencmd/pom_xml_modifier.go
. It
utilizes the etree xml library to parse and modify the mvn pom.xml
config file
in a similar way as described above for gradle. One notable difference between
gradle and maven here is that gradle actually embeds the refrence to the javac
wrapper directly into the build file, while the modifications to the maven pom
xml file merely allow future configuration at runtime.
CMake
CMake repositories are extracted from compile_commands.json
after building
the repository with -DCMAKE_EXPORT_COMPILE_COMMANDS=ON
. It then invokes the
cxx_extractor
binary as if it were a compiler for each of those commands.
Bazel
Actually we have no custom work here. We extract compilation records from Bazel
using the extra action mechanism. The extractrepo tool therefore doesn't handle
Bazel directly, but repositories using Bazel for languages we already support
should work without extra effort.