nvidia-holoscan / holoscan-cli / 41440469516

83%

push

github

fix(install): validate and safely replace development registrations (#277)

* fix(install): validate and safely replace development registrations

Signed-off-by: Tom Birdsong <tbirdsong@nvidia.com>

* fix(install): address development registration review feedback

Signed-off-by: Tom Birdsong <tbirdsong@nvidia.com>

* fix(install): detect providers shadowing development builds

Signed-off-by: Tom Birdsong <tbirdsong@nvidia.com>

* fix(install): preserve dev hooks on failed replacement

Development registration could remove a working legacy .pth/helper pair if
writing either replacement failed. The metadata path could also report an
installation failure after its new .pth was active when removing the obsolete
helper failed. Both outcomes obscure which registration will be used by the
next Python process and can leave a previously working module unavailable.

Stage legacy hook files in a private directory under site-packages, back up
existing registrations, and replace the helper before activating the .pth.
Restore any replaced files if a later replacement fails. Treat obsolete
helper removal as a warning after metadata registration succeeds, and
document both failure behaviors. Add regression tests for each legacy write
failure and for metadata cleanup failure.

Validation: focused install tests, pre-commit, and the applicable local unit
suite passed. The runner-policy failure encountered during CI validation is
tracked separately in #286.

Signed-off-by: Tom Birdsong <tbirdsong@nvidia.com>

* fix(install): skip legacy hooks for invalid metadata builds

Implicit discovery previously warned about malformed provider metadata but could still register a stale legacy hook from the same build. Skip that entire build when metadata is invalid, while leaving legacy-only builds and valid sibling builds discoverable.

Cover the invalid-metadata plus stale-hook case and document the working-directory restriction when replacing an active registration.

Signed-off-by... (continued)

150 of 159 new or added lines in 1 file covered. (94.34%)

4856 of 5870 relevant lines covered (82.73%)

0.83 hits per line