The wrapper zip files are redundant, and not required after extraction.
Gradle 7.5+ will delete these automatically, but we delete them for older
versions to avoid caching the wrapper distributions twice.
Sinc the `gradle-home-cache-excludes` parameter does not support
wildcards, we remove them explicitly.
To save space, future versions of Gradle are likely to delete the downloaded distribution
after extracting it. See gradle/gradle#3605 and gradle/gradle#19495.
To cater for this we will now save/restore the extracted distribution rather than the
downloaded zip file.
MacOS runners are initialized with a Gradle User Home directory including
the `~/.gradle/notifications` directory. This was causing the action to skip
restoring the Gradle User Home on MacOS.
This fix limits the pre-existing GUH check to the `~/.gradle/caches` directory
which isn't pre-initialized in the runner.
Fixes#155
There may be cases where it a "fresh" cache entry would be beneficial,
for example if the Gradle User Home cache entry grows over time.
This setting would run the build as if no prior cache entry exists.
By default, the action will attempt to restore a Gradle User Home
cache entry from a different set of matrix inputs (or a different Job entirely)
if an existing entry is not found for the current Job (including matrix inputs).
By specifying the experimental `gradle-home-cache-strict-match` parameter, a user
can avoid this fuzzy matching and ensure that a job execution starts with an
empty Gradle User Home if no entry from a prior execution is found.
Instead of using a fallback strategy to locate a configuration-cache entry
based on the current job and git SHA, these entries are now keyed based on their
file content with the keys persisted in the primary Gradle User Home entry.
This removes the chance of having a configuration-cache entry restored that is
incompatible with the restored Gradle User Home state, and makes the logic easier
to understand.
This change involved a fairly major refactor, with the CacheEntryExtractor being
split out from the primary cache implementation, and adding a separate extractor
implementation for configuration-cache.
- Remove configuration parameter for cache-entry-definitions. This allows richer modelling of extracted entries.
- Include both *.zip and *.tar.gz in entry definition for java-toolchains.
If the user supplies no 'arguments' parameter, the action will function as a
'setup-gradle' action, adding Gradle to the PATH and enabling other features
without actually running a Gradle build.
Any subsequent Gradle invocations in the workflow will benefit from:
- Save/restore of Gradle User Home
- Save/restore of configuration-cache data
- Capture of build-scan URLs
These features are enabled via Gradle User Home, so any Gradle invocation that
uses the same Gradle User Home will be included.
Previously, the action was restoring/saving the configuration-cache data for each
step that applied the action. In order to support Gradle invocations that are _not_
managed by the action, the configuration-cache restore is now performed in the initial
action step, and save is performed in the final post-action step.
The build root directories are recorded for each invocation via an init script.
Instead of relying on the separate cache implementations to check for the
existence of cached products, we now explicitly track whether or not the execution
is the first time the action has been invoked for a job.
Using a single cache entry for all files of a type is necessary to avoid
overloading the cache service. However, this mechanism is not very efficient
for certain artifacts like wrapper zips and generated-gradle-jars, where the
same individual files are often shared between different jobs.
With this change, any configured file patterns that do not end in '*' will
be cached as individual files. At this time this includes downloaded wrapper
zips and generated-gradle-jars.
Fixes#78
This is a pure refactor, moving from a separate .cache file per bundle to a single cache-metadata.json file describing all bundles. Instead of storing cache metadata in a separate .cache file per artifact bundle, all of the metadata is now stored in a single `.json` file.
This will make it easier to implement more flexible artifact-caching strategies, such as caching each wrapper zip separately.
* Always include cache protocol version in cache key
* Store all cache metadata in a single JSON file
* Rename cache-metadata file and bump protocol version
* Polish and documentation
Instead of writing the URL to a file on disk, reading it later and
using the Actions API to record the output parameter and write the notice,
these things are now done directly via Actions commands emitted directly
from the init script.
Instead of writing this file to a temp directory and referencing it
on the Gradle command line, the init script is now written to Gradle
User Home so that it is picked up automatically.
Instead of passing `--no-daemon` on the command line, the same
functionality is now acheived by writing a gradle.properties file
when initializing Gradle User Home.
Failures to store cache entries should not fail the action or the Job.
This fix attempts to catch and log any unexpected errors that occur when
saving cache entries.
Fixes: #119Fixes: #120
- Warn and continue on failure to restore a Gradle distribution from cache
- Warn and continue on failure to save a Gradle distribution to cache
- Extract common functionality for consistent handling of cache failures
Fixes#116
Instead of tracking a single 'fully-restored' flag, track the restore status of each
cache entry restore. If any of these are requested but not restored, then the overall
Gradle User Home cache is not fully restored.
Added special handling for the case when zero artifact bundles are set: this is used
in tests to simulate a not-fully-restored state.
Previously, only .jar files were bundled, with other files (modules, POMs, zips, etc)
being left in Gradle User Home. All downloaded files are now included in the bundle.
Fixes#100
The `gradle-home-cache-includes` and `gradle-home-cache-excludes` parameters were initially implemented
as JSON string inputs. This makes these inputs non-idiomatic and easier to get wrong.
This change converts them to multi-line input parameters.
Fixes#106
