1.2.4

[deanxbox]

1.2.4

Task Manager 1.2.4 focuses on conflict analysis, clearer attribution, and more reliable session capture.

Added

• Pairwise lock attribution that records both the waiting thread and the lock owner thread when the JVM exposes that information
• Repeated conflict tracking across the rolling capture window so the mod can highlight recurring waiter mod -> owner mod contention instead of isolated one-off waits
• Broader thread role classification for worker pools, IO pools, chunk generation, chunk meshing, chunk upload, network, GC, render, and main logic threads
• Alternate owner candidates in thread drilldown so low-confidence ownership can be reviewed instead of being flattened into a single guess
• Dedicated conflict findings in the UI and HTML export, split into confirmed contention, repeated conflict candidates, weak heuristics, and unrelated slowdown causes

Changed

• Scheduling-conflict detection no longer depends only on Worker-Main-* thread names and now uses thread names plus stack ancestry
• Conflict wording is more conservative and now distinguishes Measured, Inferred, Pairwise inferred, Weak heuristic, and reserved support for Known incompatibility
• Lock summaries prefer pairwise conflict phrasing when both sides of a wait can be identified
• Export summaries and diagnosis text now include top conflict candidates instead of only generic lock warnings
• Thread detail views now expose role labels, role-source labels, and alternate owner candidates

Fixed

• MANUAL_DEEP recording now continues to capture session samples after recording starts, even if the Task Manager screen is closed
• Session exports no longer fall back to No session samples were captured for valid MANUAL_DEEP recordings started with F11
• Conflict and slowdown reporting no longer overstates certainty by calling inferred waits confirmed mod conflicts

Notes

• Known incompatibility is currently a confidence tier and UI label, not a populated incompatibility database. In 1.2.4, conflict findings are still based on measured contention plus inferred ownership rather than a bundled registry of hardcoded bad mod pairs.