| Commit message (Collapse) | Author | Age |
|---|
| |
|
|
|
|
|
|
|
| |
Noticed this while messing with remote runfiles.
RELNOTES: local_repository can take a relative path.
--
MOS_MIGRATED_REVID=118780775
|
| |
|
|
|
|
|
|
| |
This is problematic for coming up with a less stateful BlazeModule API, which
- in turn - is a requirement for running multiple commands in the same server.
--
MOS_MIGRATED_REVID=118696892
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The SkylarkRepositoryModule declare the `repository_rule` function
to Skylark to define new remote repository types (http://goo.gl/OZV3o0).
The work is delagated to the `SkylarkRepositoryFunction` by the
`RepositoryDelegatorFunction`. `SkylarkRepositoryContext` defines the
`ctx` object passed to the `repository_rule` implementation function.
This change also introduce a `SkylarkPath` and the necessary methods
in `SkylarkRepositoryContext` to showcase the creation of a
`local_repository` like repository.
Issue #893: step 3 of the roadmap http://goo.gl/OZV3o0.
--
MOS_MIGRATED_REVID=114895003
|
| |
|
|
|
|
|
| |
This makes it possible to compile //third_party/ijar with a bootstrapped Bazel on Windows in dslomov's tree.
--
MOS_MIGRATED_REVID=114428109
|
| |
|
|
|
|
|
|
|
| |
Moved RepositoryValue to RepositoryDirectoryValue so that it could be cached
(and not re-downloaded) even if the WorkspaceAST caused a Skyframe restart
(as mentioned in https://github.com/bazelbuild/bazel/blob/master/src/main/java/com/google/devtools/build/lib/rules/repository/RepositoryDelegatorFunction.java#L130-L133).
--
MOS_MIGRATED_REVID=113358489
|
| |
|
|
|
|
|
| |
our plans for symlink support on Windows.
--
MOS_MIGRATED_REVID=113043269
|
| |
|
|
|
|
|
|
|
|
| |
Fixes #733.
RELNOTES: Relative paths can now be used for 'path' with new_local_repository
and local_repository.
--
MOS_MIGRATED_REVID=111620894
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
directory that I should populate". The directory itself is not created because local_repository actually puts a symlink in its place.
As a side effect, make HTTP downloading, git cloning and archive decompressing not be SkyFunctions. This is necessary because it needs to be the RepositoryFunction and not a
dependent SkyFunction that populates the output directory, because it that's the case what happens is:
1. RepositoryDelegatorFunction cleans up the directory and prepares it for RepositoryFunction
2. RepositoryFunction calls env.getValue(<function that populates the directory>)
3. That value hasn't been computed yet, thus RepositoryDelegatorFunction returns early
4. The function that populates the directory is called
5. RepositoryDelegatorFunction is restarted
6. RepositoryDelegatorFunction cleans up the directory
7. RepositoryFunction calls env.getValue(), and nothing is done because the value has already been computed
8. RepositoryDelegatorFunction proudly returns, even though the directory is actually empty
Another way to solve this problem would be to make RepositoryFunction not clean the directory up on Skyframe restarts, but that means that we'd need to keep state somewhere, which doesn't strike me as a particularly great idea because let's keep state outside of Skyframe only when absolutely necessary (e.g. the marker files for cross-server instance persistence of downloaded repositories). That "somewhere" could either be a member variable of RepositoryDelegatorFunction or the file system.
Note that this change causes external communication to be re-done in a few more cases than before (see that changes to the test cases), but I'd rather we be correct and simple than fast. We can optimize things later if needed and there is enough complexity going around, thank you very much.
--
MOS_MIGRATED_REVID=110134397
|
| |
|
|
|
|
|
| |
"overlaid BUILD files" hack in RepositoryValue.
--
MOS_MIGRATED_REVID=109877252
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In particular:
- Separate the implementation of maven_server into a RepositoryFunction and one that creates the MavenServerValue (ideally, maven_server wouldn't exist but we'll have to make to for the time being)
- Refactor the logic of determining whether an external repository needs to be re-fetched to RepositoryDelegatorFunction
- Make RepositoryFunctions not be SkyFunctions anymore (they are called from RepositoryDelegatorFunction, though, who *is* a SkyFunction)
- Add a Skyframe dirtiness checker that makes --nofetch RepositoryValues not be cached
- Add a bunch of test cases and javadoc
There is only one wart that I know of that remains: changes to BUILD files of new_* repository rules that weren't refetched when their RepositoryValue was initiall created on server restart won't take effect. This is because we don't add those BUILD files to the created RepositoryValue. This will fix itself once the ExternalFilesHelper refactoring is submitted.
--
MOS_MIGRATED_REVID=109768345
|
| |
|
|
|
|
|
| |
RepositoryValue and not depending it from repository functions.
--
MOS_MIGRATED_REVID=108685867
|
|
|
build.lib.
--
MOS_MIGRATED_REVID=106689603
|
Kristina Chodorow
Ulf Adams
Damien Martin-Guillerez
Lukacs Berki