Record: the installer registers the handler

Outcome

make install no longer stops at installing the programs: it registers the handler as well, so a downloaded pass opens as soon as the install is done.

The work is in commit 21f7df5.

• scripts/program-install.sh ends by running the freshly installed pkpass-integrate handler install, after every program and the icon are in place.
• PKPASS_INTEGRATION_NO_HANDLER=1 installs the programs and registers nothing.
• A registration that cannot happen leaves the installed programs in place, prints why, and exits nonzero.
• tests/91-install-handler.sh checks the plain install, the opt-out, and the failure path, all against a throwaway HOME and prefix.
• Feature 05-pkpass-handler.md gains PKPASS-HANDLER-R021; task 06-install-registers-handler-implement.md claims it.

make test is now 23 tests, all passing.

Decisions

• The owner’s instruction overrides the earlier choice. The project had made registration a separate, deliberate step on the reasoning that changing what opens a file type should be visible and reversible. The owner asked for the handler to be part of make install, and the reversibility is what carries the safety, not the separation: the record, the uninstall, and the opt-out are unchanged.
• The registration runs the installed program. The installer invokes $PREFIX/bin/pkpass-integrate, so the entry’s Exec points into the installation prefix. Had the build tree’s copy been run, the entry would have named a temporary directory that make clean removes.
• The choice comes from the environment and the failure is closed. The installer is unattended (prompts/03-conventions.md section 7.1), so the opt-out is PKPASS_INTEGRATION_NO_HANDLER=1 rather than a question, and an install that cannot register exits nonzero after reporting that the programs are installed. Reporting success with the association missing is the failure this avoids.
• scripts/install.sh, the release installer, is deliberately not changed. The owner’s instruction named make install. The release path takes a tarball and prints the integration step; making it register is a separate decision, and it is recorded in TODO.md as an open question rather than assumed here.

Verification

• make test — 23 tests, all pass, on 2026-10-10. 91-install-handler is the new one.
• The installer was run by hand against a throwaway HOME and prefix, on 2026-10-10, and left both types resolving to pkpass-handler.desktop with the definition, three icons, and the record in place; the opt-out left the programs installed and the desktop untouched.
• The test sandboxes HOME, XDG_DATA_HOME, XDG_CONFIG_HOME, and XDG_DATA_DIRS, so the machine running the tests never has its own desktop changed. The real $HOME/.local was not written to during this work.

Commits

• 21f7df5 — feat: register the handler from make install. The installer, the test, the Makefile comment, the README, TODO.md, the feature requirement, and the new task.
• The commit that adds this record is separate, per records/README.md, and cites the work commit above rather than itself.