Gradle

Clone Tools
  • last updated a few seconds ago
Constraints
Constraints: committers
 
Constraints: files
Constraints: dates
Track less state in MerkleTreeSnapshotBuilder

Merge branch 'release'

Improve performance for getRegularFileSnapshot (#6199)

Only intern on demand and revert "Rebaseline build on nativeMonolithicOverlapping".

Merge pull request #6179 from gradle/ew/docs/release-notes-4.10

Gradle 4.10 release notes/features

Rename 'toolingBuildersNative' project to 'toolingNative' to work around limitations in CI infrastructure.

  1. … 25 more files in changeset.
Include only necessary JS

Including the guides.js duplicates a bunch of stuff already done

in build-tool.js. The new multi-language sample handling JS has

been migrated to the CDN.

    • -1
    • +43
    /subprojects/docs/src/main/resources/head.html
Optimize docs text-rendering

    • -2
    • +1
    /subprojects/docs/src/docs/css/manual.css
Reorder some sections of release notes by importance

    • -24
    • +24
    /subprojects/docs/src/docs/release/notes.md
Add more details to Kotlin DSL and snapshot plugin versions parts of release notes

    • -2
    • +10
    /subprojects/docs/src/docs/release/notes.md
Remove unused import.

Re-enable performance test.

    • -1
    • +1
    /subprojects/performance/templates.gradle
Temporarily ignore Gradle build performance test.

Remove the `sourceSets` property from the Java base plugin extension. The source sets are accessible through the project `sourceSets` extension instead.

Merge branch 'release'

Bump wrapper to latest release snapshot

    • -1
    • +1
    /gradle/wrapper/gradle-wrapper.properties
Publish 4.10-20180806080315+0000

Fix DefaultCacheAccessTest

Merge pull request #6150 from gradle/marc/fixes/1412-contention-handling-improvements

Improve file lock contention handling

Handle missing type correctly

When communicating with an old version, the length of the received

packet will be shorter than the byte array. Now, `FileLockPacketPayload`

only reads the type when it has been received and returns UNKNOWN if it

has not. Previously, it would have worked by accident because the last

byte was zero which corresponds to the ordinal number of UNKNOWN.

Rename and update the Javadoc for a tapi model method to make its purpose clearer.

Make the tapi client side type mapping a little more efficient.

Expose the variant name and header directories for each C++ binary in the C++ tooling model.

Rework construction of the native binary models so that the binary names are not formatted and parsed a bunch of times during construction and so do not lose information.

  1. … 22 more files in changeset.
Expose the output object file for each C++ source file in the C++ tooling model.

  1. … 3 more files in changeset.
Expose details about the compiler executable and compile working dir for each binary in the C++ tooling model.

  1. … 3 more files in changeset.
Expose details about the task to run to produce the object files and the binaries files for each binary in the C++ tooling model.

Make file lock packet type explicit

`FileLockCommunicator` now includes an additional byte that represents

the `FileLockPacketType` when sending packets to other processes.

`DefaultFileLockContentionHandler` now checks for the

`LOCK_RELEASE_CONFIRMATION` type instead of interpreting any packet for

a lock for which it already has received a packet as a lock release

confirmation.

Adding the additional byte will be ignored by old versions which only

read the first 9 bytes. New versions can still read packets from old

versions that do not include the byte. They use the `UNKNOWN` type in

this case. Thus, this protocol change is backwards compatible.

Publish 4.10-20180805031819+0000

Fix `ListProperty` and `SetProperty` implementations to coerce `GString` values to `String` when the property element type is `String`, and remove work around from C++ model builder.

Add some convenience methods to add elements to a `ListProperty` using an array or `Iterable`.

Rebaseline build on nativeMonolithicOverlapping

It looks like the tests with 0 workers got slower (around 4% for

`nativeMonolithicOverlapping` and 1% for `nativeMonolithic`).

The builds with 12 parallel workers seem to be as fast as the

baseline.

Rebaselining so I can investigate the regression.