Bug Fixes
-
lager updateno longer leaves a file on the box after it is deleted upstream. The step that flattens the repository layout on the box copied additively: it overwrote changed files but never removed ones the new tree no longer contained, and it then deleted the tracked source, so the result was beyond the reach ofgit checkout -forgit reset --hard. A file removed upstream therefore stayed on every box indefinitely and was copied into the runtime image by the next build — boxes were found carrying a module deleted thirteen minor versions earlier. Each top-level entry is now removed and then moved into place, so a subtree is rebuilt rather than merged into and deletions take effect on the first update. Files installed at the repository root by any other route are untouched. No manual cleanup is needed. -
A source-only change now invalidates the cached Docker image. The build
hash covered only the Dockerfile and
requirements.txt, so a pure-Python change relied entirely on the layer cache invalidating correctly. Every file under the box’s source tree now feeds the hash;__pycache__and.pycare excluded so regenerated artifacts do not force needless rebuilds. Because the stored hash on an existing box was computed with the old formula, the first update after this release rebuilds the image in full, once per box. -
A J-Link script no longer leaks onto debug operations that never asked for
one. The box kept a single script file and handed it to any operation that
did not supply its own, so
reset,erase,memrdand the gdbserver path silently inherited whichever script was written last — by a different net, by an earlier session, or by a test suite that had since finished. Scripts are now written per net, an operation with no net gets no script rather than an ambient one, and a net’s script is cleared when its debug session ends. A net with a genuinely required customInitTargetstill gets it on every operation, and an older box’s shared file is removed when the debug service next starts. -
A J-Link script no longer leaves a just-erased target unattachable. A
user
.JLinkScriptreplaces J-Link’s built-in per-deviceInitTarget(), and on an nRF5340 that built-in is what brings the DAP up on a blank part. When it was displaced, the attach following an erase failed withCould not read CPUID register— and becauseflasherases by default, one scripted flash could leave the part blank and the net failing every later attach. The attach path now retries without the script rather than wedging the net. -
lager nets statenow says why a net has no state.state: nullmeant three unrelated things with no way to tell them apart: the instrument had not answered before the request deadline, its probe failed, or the role has no probe at all. Null entries now carry areason—"deadline","no probe for role", or"unreadable: <detail>". The command prints the unexpected ones in a footnote after the table, grouped by reason, and--jsoncarries them aslive_state_reason. A USB hub that cannot be opened now names itself and its cause instead of failing silently. A CLI newer than its box simply sees noreasonand renders as before.