v1.0.0

1.0.0 - 2026-10-17

Added

• Initial release.
• spark_analyze tool: overall performance report for a spark export — platform/Minecraft/spark versions, capture window, TPS/MSPT/CPU/memory/GC/players/world snapshot, time-series metrics, plugin/mod list, thread groups, hotspots (self + inclusive) with tick-window filtering, worst tick windows, plugin/mod attribution, and heap summaries.
• spark_query tool: per-view queries over the same file — methods, threads, tree, windows, sources, metrics, plugins, world, system, and heap.
• Support for spark data version 2 (flat childrenRefs trees) and legacy nested children/time exports.
• gzip-wrapped exports, spark health raw data (degrades to a health report when no sampler stacks are present), and execution/allocation/lock sampler modes.
• Platforms: Paper / Spigot / Fabric / Forge / NeoForge, proxies, and spark-nukkit (Nukkit exports parse identically; tick statistics are reported as unavailable).
• Automatic Mojang mappings: profiles whose sampled frames look ProGuard-obfuscated (flat class names such as agh.a, typically vanilla servers) are remapped to official names in every view; frames that already use readable names (Paper, Spigot, Forge/NeoForge SRG, Fabric intermediary, proxies) are detected and skipped with an explanatory note. Resolution order: an explicit mappings_path tool parameter, a validated sibling file next to the export (<export>.mappings, <export>.map, server.txt, client.txt), the on-disk version cache (~/.dsh/spark-insights-mappings/<version>.txt, also user-populated), and a download from Mojang's piston-meta service that is validated before being cached atomically. Downloads share one 25-second budget and a transient failure is retried after two minutes instead of being disabled for the process (definitive answers such as a version without mappings are not retried); cache changes are detected via a stat footprint so edited or newly added mapping files refresh cached profiles. Descriptor-aware overload resolution (ambiguous overloads keep the obfuscated name), lambda$… frame rewriting, source-attribution key rewriting, and a bounded parsed-mapping cache are included. SPARK_INSIGHTS_NO_NETWORK=1 disables downloads; SPARK_INSIGHTS_MAPPINGS_DIR relocates the cache.
• Schema-generated protobuf decoder from official spark protos with a hand-written wire-format fallback.
• Parse-result cache keyed by file version (most recent 2 files).
• Safety limits for untrusted input: 256 MiB per file, 512 MiB decompressed, 4 million nodes, 4096-frame depth, 64 MiB validated mapping files, and a bounded parsed-mapping cache.
• English and Chinese locales; no telemetry, and the only network requests are the optional Mojang mappings downloads described above.