Add foo-root=bar and rd.foo.root=bar to the legacy migration fixture
and assert both survive intact. Against the word-boundary strip these
assertions fail with the tokens truncated to foo- and rd.foo.; the
existing second-run check keeps covering them through both executions.
Start the RHEL 7 run from a sysconfig carrying the shared-root value the
previous script wrote, as on an already-deployed node, plus options from
the stock RHEL 7 file. Assert that exactly one root= remains, that it
points at the node subdirectory, that rd.neednet=1 and rootflags=nofail
survive, and that a second run leaves KDUMP_COMMANDLINE_APPEND
unchanged.
Against the previous prepend-only sed the count and idempotency
assertions fail: the legacy root= survives as the last token, which is
the one dracut honors.
Runs the RHEL NFS path of the postscript against a scratch tree that
stands in for the mounted dump export. Checks that kdump.conf points at
the node's own subdirectory, that only that subdirectory is created on
the target, and that nothing lands at the export root. RHEL 7 keeps the
dracut workaround but under the node subdirectory.
Five of the eight assertions fail against the previous behavior.
Vinícius Ferrão