Re: #786: Automatically detect bundled sources in plugin dependency

[karollewandowski]

Fixes: #786.

In the case when multiple source artifacts exist for a plugin, only the first one will be attached because of the IDEA issue: IDEA-292168.

This may be the case when bundled plugin contains sources - IDE sources are attached for bundled plugins so we will have plugin and IDE sources attached (attaching IDE sources was implemented in the scope of #153). I intentionally ordered plugin sources first to avoid IDEA-292168.

I was wondering whether we should configure sources when intellij.downloadSources is false and plugin is bundled, but I think it's better to add them always whenever we have them without downloading. What may look strange is that Ivy file will contain plugin's source entry in "regular" and -withSources versions. Let me know if I should change it.

Additional small cleanups:

• added default predicate to collectJars() for cleaner client code
• removed PluginDependency.sourcesDirectory as it is legacy and not used at all
• added clarification about including gradle plugin project in test IntelliJ plugin project for debugging purposes

[karollewandowski]

Windows runners tests failed due to: java.io.IOException: There is not enough space on the disk. Is it known/addressed?

[karollewandowski]

Windows runners tests failed due to: java.io.IOException: There is not enough space on the disk. Is it known/addressed?

Windows runners have 14 GB of space on disk D which is used for executing actions (see: actions/runner-images#1341). We've used this limit after my changes because I added tests requiring the GoLand, and IDEA 2021.2.4 (additional 1.46 GB + 2.3 GB = 3.76 GB).
2021.2.4 is the lowest version compatible with the oldest Go plugin having /lib/src artifacts. I've checked PHP and Python plugins which do have sources as well for a potential alternative compatible with currently used 2020.1, but they don't have them in versions older than 2021.2.4.

As the solution, I will update our default IDEA version used in tests to 2021.2.4 and will adjust other tests where needed.

This issue may happen in the future again if we download more files/IDEs, but the alternative solution is not trivial (see actions/runner-images#1341) and maybe we won't need it for a longer time.

[karollewandowski]

Windows runners tests failed due to: java.io.IOException: There is not enough space on the disk. Is it known/addressed?

Windows runners have 14 GB of space on disk D which is used for executing actions (see: actions/virtual-environments#1341). We've used this limit after my changes because I added tests requiring the GoLand, and IDEA 2021.2.4 (additional 1.46 GB + 2.3 GB = 3.76 GB). 2021.2.4 is the lowest version compatible with the oldest Go plugin having /lib/src artifacts. I've checked PHP and Python plugins which do have sources as well for a potential alternative compatible with currently used 2020.1, but they don't have them in versions older than 2021.2.4.

As the solution, I will update our default IDEA version used in tests to 2021.2.4 and will adjust other tests where needed.

This issue may happen in the future again if we download more files/IDEs, but the alternative solution is not trivial (see actions/virtual-environments#1341) and maybe we won't need it for a longer time.

I updated the default tested platform version to 2021.2.4 and we have ~2 GB margin now.
We can decrease it by 0.5 GB by getting rid of IC-14.1.4 usage in some tests, but I'm leaving it for now.

Edit:
Tests on Windows failed again.
The next week I will verify if it's not a GitHub Actions cache issue (AFAIK it may have old IDEs cached before my changes and my optimizations don't take effect until it is cleaned).

[karollewandowski]

[Applies to entire file]
I cleaned the tests:

• reduced code duplication
• reduces code noise in testing methods, so they include actual tested context and assertions
• added messages to assertions - before there was a lot of "expected true, but got false" messages and now there will be more precise

All existing tests and logic should be preserved.

[karollewandowski]

This downloads and unpacks IDEs next to the test Gradle home directory.
I created custom logic for downloading IDEs to not depend on the plugin logic (if reusing is even possible). It can be also used in other tests requiring local installations.

[karollewandowski]

Agreed with Yann to skip a feature flag for this as it would introduce unnecessary complexity.
Attaching sources depends on generating Ivy files and if we wanted to isolate the old logic of attaching IDE sources and control only attaching bundled sources, we would have to generate unique Ivy files for each context combination (no sources, IDE sources only, bundled sources only, IDE+bundled sources) and it would grow if we added new ways of attaching sources. An alternative is to control attaching sources as a whole, but we decided to skip this as it is a pretty safe feature and doesn't affect building the project itself.

I'm delaying the merge until the new version is released. I'll rebase the branch and fix failed checks after the release.