v0.6.1

Released: 2026-05-13

Republish of v0.6.0 with a smaller VSIX. v0.6.0 was auto-published by the GitHub Actions release workflow on v* tag push, before a .vscodeignore fix excluded a stray local indexing database (.cocoindex_code/) that had been picked up by the packager. v0.6.1 ships the same code with a clean 679 KB VSIX instead of v0.6.0’s 6.4 MB. Prefer v0.6.1. No functional or behavioral differences between the two releases.

Architecture

• Direct-memory RTT path replaces libjlinkarm’s high-level RTT API. LogScope now reads the SEGGER RTT control block directly via target memory reads (the approach used by probe-rs, OpenOCD, and Nordic’s nrfutil) instead of routing through JLINK_RTTERMINAL_*. This bypasses libjlinkarm’s host-side tracked_RdOff cache.
• The legacy path remains available via LOGSCOPE_RTT_LEGACY=1 for one-flip rollback if the new path causes regressions.

Fixed

• Mid-session board resets are now detected and recovered automatically. Pressing the reset button while LogScope is connected no longer leaves the panel stuck on the pre-reset buffer; the helper detects the failed SWD reads, re-attaches the J-Link probe, and surfaces the new boot banner.
• The probe-rs-style last_written_rd_off detection pattern is implemented: after the helper writes RdOff to the target, it remembers the value and compares it against the target’s RdOff on the next poll. If they diverge, a target reset has happened and the helper re-attaches without halting the CPU. On hardware where SWD recovers on its own after reset, this is a single-digit-ms path.

Recovery time on Nordic chips (known limitation)

• Hardware testing on the nRF54L15 shows a mid-session reset recovery time of approximately 2.5 to 3 seconds end to end. Empirically the J-Link DLL jams its DAP state during the reset window on this chip and jlink.connect() alone (without close() first) fails in a few ms with “Unspecified error”. The full sequence close() + sleep + open() + connect() + re-attach is required to clear the jam, which sets the floor. The cheap re-init path is still attempted first (free when it fails fast) and on other vendors’ hardware (STM32, SiLabs, TI; untested at v0.6.0) it may succeed, in which case recovery is sub-second. See TODO.md for the optimization follow-ups planned for v0.6.1.

Added

• Channel names from the SEGGER_RTT control block display in the sidebar (for example, “Terminal” instead of “Channel 0”). Names come from the sName pointer in each up-buffer descriptor and are re-emitted after every successful re-attach so post-reset firmware updates with different channel names show correctly.
• pylink error classification: JLinkException codes are now classified as transient (retry via polling), probe-fatal (immediate full reconnect), or target-state (surface; no auto-recovery). Time-based grace (300 ms) before escalating to recovery prevents spurious reconnects on isolated bad reads.
• Python test suite at test/python/ covering CB scanning, parsing, ring-buffer math, error classification, and the session state machine. Run via npm run test:python (41 tests).
• CI step runs the Python tests on every PR.

Rollback

• If you hit a regression, set the LogScope: Rtt: Legacy Mode VS Code setting to true (or "logscope.rtt.legacyMode": true in settings.json) to force the old JLINK_RTTERMINAL_* path. Reload the window for the change to take effect. The environment variable LOGSCOPE_RTT_LEGACY=1 also works for command-line use. Please file an issue at https://github.com/NovelBits/logscope/issues with the reproduction details.