Improved overflow support: Indexing rescanner

[sungshik]

This PR implements the generation of synthetic CREATED/MODIFIED/DELETED events for DIRTY files to auto-handle overflow events. There are two new classes: IndexingRescanner and BaseFileVisitor. To conveniently reuse some of the functionality of MemorylessRescanner, there are refactoring-kind-of changes in that class, too.

[codecov]

Codecov Report

Attention: Patch coverage is 76.82927% with 19 lines in your changes missing coverage. Please review.

Project coverage is 80.1%. Comparing base (2aa7db2) to head (493165d).
Report is 19 commits behind head on improved-overflow-support-main.

Files with missing lines	Patch %	Lines
...g/swat/watch/impl/overflows/IndexingRescanner.java	79.1%	6 Missing and 4 partials ⚠️
...ing/swat/watch/impl/overflows/BaseFileVisitor.java	58.8%	5 Missing and 2 partials ⚠️
...swat/watch/impl/overflows/MemorylessRescanner.java	87.5%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@ Coverage Diff @@## improved-overflow-support-main #26 +/- ##
==================================================================
+ Coverage 80.0% 80.1% +0.1% - Complexity 110 120 +10 
==================================================================
Files 14 16 +2 Lines 486 539 +53 Branches 47 53 +6 ==================================================================
+ Hits 389 432 +43 - Misses 69 75 +6 - Partials 28 32 +4 

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[sungshik]

Clarifying comments

[DavyLandman]

On first glance it looks good. I'm curious about the perf impacts of this.

I'm curious what the effect is of replacing the existing one this one?

As in, this code is not yet run in the torture test.

[sungshik]

Thanks for the review! The low-hanging fruit is done. I'm leaving one conversation open and will first experiment a bit with performance.

[sungshik]

Regarding performance...

There happen to be two torture tests that use a non-recursive directory watch, so I ran them with/without indexing (repeated 20x). Here are the preliminary/anecdotal results, just to get a rough idea:

• Test 1: manyRegistrationsForSamePath
  • MemorylessRescanner: 4838 ms
  • IndexingRescanner: 7982 ms -- 3 seconds slower (+65%)
• Test 2: manyRegisterAndUnregisterSameTime
  • MemorylessRescanner: 321288 ms
  • IndexingRescanner: 326445 ms -- 5 seconds slower (+1.5%)

A working hypothesis to explain the difference is that Test 1 has 2000 threads that concurrently try to access the file system to index the test directory, while Test 2 has only 4 such threads.

In the spirt of "many small PRs", I'll go ahead and merge this one, and I'll add a TODO to #20 to separately conduct a more serious comparison (see also this comment) to understand the performance tradeoff, and optimize if needed/possible.