Fix IllegalArgumentException when drawing unified diff code minings

[tobiasmelcher]

UnifiedDiffLineHeaderCodeMining.draw() caches Font references in its foregrounds list and in cachedFont. These fonts can be disposed under the cache when:

1. The mining is disposed while a paint event for it is still in flight (e.g. "hide diff" recomputes the mining list and a pending paintControl still references the old annotations).
2. The widget font is replaced by an instance with equal FontData. The previous invalidation used Font.equals(), which is a value compare and does not detect a disposed handle.

Once an exception was thrown, the stale cache was never refreshed, so every subsequent scroll repaint produced another error.

Fix:

• dispose() nulls the cached state so a stray paint after disposal cannot enter the fast path.
• Cache invalidation also checks cachedFont.isDisposed().
• The fast-path loop validates f.font.isDisposed() before use; if stale, the cache is dropped and the slow path rebuilds it in the same paint.

Fixes #2772

[tobiasmelcher]

@iloveeclipse Unfortunately, I was unable to reproduce the issue on a Linux machine.
Could you please provide some additional information?

Were there any exceptions in the error log besides the java.lang.IllegalArgumentException?
Did you change the text editor font settings while the unified diff was being rendered?

From the current behavior, it seems that cached fonts may have been disposed unexpectedly, but I’m not sure under which circumstances this could happen.
Is it possible that LineHeaderCodeMining#draw is being called after the instance has already been disposed via LineHeaderCodeMining#dispose?

[Copilot]

Pull request overview

This PR fixes repeated IllegalArgumentException errors thrown while painting unified diff line-header code minings when cached Font instances become disposed (e.g., after “hide diff” triggers recomputation while a paint is still in flight). It also hardens unified diff navigation runnables against empty diff lists.

Changes:

• Clear cached mining paint state on dispose() and invalidate caches when the cached widget font is disposed.
• Detect disposed cached foreground fonts during the fast-path draw loop and fall back to rebuilding caches in the same paint.
• Prevent NextRunnable / PreviousRunnable from indexing into empty unified diff lists.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
team/bundles/org.eclipse.compare/compare/org/eclipse/compare/unifieddiff/internal/UnifiedDiffCodeMiningProvider.java	Improves cache invalidation and draw fast-path robustness when fonts are disposed; clears cached state on dispose.
team/bundles/org.eclipse.compare/compare/org/eclipse/compare/unifieddiff/internal/PreviousRunnable.java	Adds a guard to return early when there are no diffs to navigate.
team/bundles/org.eclipse.compare/compare/org/eclipse/compare/unifieddiff/internal/NextRunnable.java	Adds a guard to return early when there are no diffs to navigate.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[iloveeclipse]

Not sure if it is related, I always use Eclipse with disabled themes. Beside this, RHEL 9.6 with X11, KDE5, nothing special. I believe all errors were same, but I will check tomorrow, they were literally flooding my error log.

[github-actions]

Test Results

    54 files  ±0      54 suites  ±0   56m 41s ⏱️ + 2m 40s
 4 677 tests ±0   4 655 ✅ ±0   22 💤 ±0  0 ❌ ±0 
11 925 runs  ±0  11 772 ✅ ±0  153 💤 ±0  0 ❌ ±0 

Results for commit 31a855b. ± Comparison against base commit 82c7987.

♻️ This comment has been updated with latest results.

[iloveeclipse]

Just tested the patch. Looks OK now, I don't see exceptions anymore. Regarding exceptions: they were all identical, exact same stack.

[iloveeclipse]

I've pushed one commit on top, could you please check?

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

[iloveeclipse]

LGTM. Tested yet again, still works :-)