Skip to content

Fix shared viewer refresh coalescing and stale batch timers - #30

Merged
hsoliwal merged 1 commit into
m3/jface-all-viewers-core-repair-20261021from
aix/jface-queue-batch-identity-20261021
Oct 22, 2026
Merged

hsoliwal merged 1 commit into
m3/jface-all-viewers-core-repair-20261021from
aix/jface-queue-batch-identity-20261021

Conversation

@hsoliwal

@hsoliwal hsoliwal commented Oct 21, 2026 •

Copy link
Copy Markdown
Owner

Problem

Adjacent refresh posts currently accumulate redundant work. A scheduled callback from a flushed batch can also drain a newer batch, breaking the later batch's debounce window.

Change

Repair the existing M3Viewers.Queue:

  • Coalesce only an incoming single REFRESH whose target is identical to the pending tail REFRESH target.
  • Preserve operation barriers, intervening updates, and explicit direct apply operations.
  • Give each new pending batch a primitive identity and require scheduled callbacks to match it before draining.
  • Add deterministic regressions and shared-viewer coverage without changing public/protected API or adding another queue.

This stacks on #27 at 67cd89ec3ea55e6bf13811666ffe90fe2c6adc08. It preserves that PR's prior repairs and import-order correction. The existing broader all-viewers programme remains applicable.

The production and test files are actual outputs of the unchanged canonical Synexia SDK/receiver. Recipe and complete execution evidence: hsoliwal/com.synexia#12797.

Executed validation

  • Baseline: five actual SWT/JFace regression/control tests ran; three expected failures reproduced and two ordering/direct-apply controls passed.
  • After: 18 tests passed, zero failures/errors/skips: 14 in M3ViewersTest, four unchanged in M3ViewerDeltaTest.
  • Nine variants exercised: table, tree, list, native Combo, CCombo, checkbox table/tree, virtual table/tree with non-lazy providers. Selection and applicable checked/grayed state were preserved.
  • Both owners' public/protected javap output is byte-identical.
  • Strict JDK21 compile (--release 21 -Xlint:all -Werror) passed.
  • Canonical SDK: eight tests passed. Receiver: check 2 → apply 2 → check 0.
  • A second, clean relocated preparation/SDK/receiver run passed offline and emitted identical Java bytes.
  • Exact payloads, provider hashes, commands and coverage dispositions accompany this change under docs/m3/views/jface-queue-20261021/; full receipts live in the linked recipe PR.

Acceptance boundary

The real UI run used published JFace 3.40.0, GTK SWT 3.135.0 and its genuine JNI, Jupiter 5.14.4, Temurin21 and official Xvfb. This qualifies the changed M3 owners on that focused closure. Full current source/Tycho/PDE acceptance on Java25 still requires the owning reactor's mvn verify gate.

Lazy providers, Problems marker-pipeline end-to-end behavior, other source-specific view adapters, Nebula and the native-platform matrix remain separately tracked. This PR does not claim the entire all-viewers programme is complete. Existing dual Apache-2.0/EPL-2.0 notices are retained.

Coalesce only adjacent single refresh posts for the identical target. Bind timers to their pending batch so older callbacks cannot drain newer work. Preserve ordered updates and explicit direct apply operations, with focused SWT regression and nine-variant coverage.

Copy link
Copy Markdown
Owner Author

Stacked incremental correction: UI #34 is based on this exact #30 head and retains its batch epoch, adjacent refresh coalescing and UI #27 Display-thread contract. It only pins the first post's absolute monotonic deadline against later longer delays and adds the SWT owning regression. Canonical exact-source recipes are in Synexia #12885; no public ABI or JNI symbol changes. Focused/native tests of this new increment remain pending.

@hsoliwal
hsoliwal merged commit 1d1fdc5 into m3/jface-all-viewers-core-repair-20261021 Oct 22, 2026
hsoliwal added a commit that referenced this pull request Oct 22, 2026
Stacked on qualified JFace queue #30, preserving queue identity, existing timer behavior, worker-safe APIs and source ancestry. Retain full platform validation as a separate gate; no squash, rebase, branch deletion.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant