The Perl tests of the OpenEmbedded Genesis image ran its shell scripts
and C programs through system(), qx, pipes and fork, wrote fake
commands through copied read_file and write_file helpers, and matched
regexes against shell and C source. Many rejections checked only a
non-zero exit, so an earlier, unrelated failure also passed. The
quality gate judged 91% of their 762 checks bad.
The shell and C checks move to bats in xCAT-test/bats. The C programs
are compiled there and run: the plain console, test drivers for the
shared status view and the shell launcher, and genesis-udp-send against
a loopback listener. Each rejection now matches the message of its own
guard. The .t files keep the artifact checks (kas, recipes, kernel
fragments, units, README), read with XCAT::Test::File. Test files left
with no checks are deleted.
Claim map (file -> where its claims went):
- go_xcat_uninstall_installed_only.t -> go_xcat_uninstall_installed_only.bats
- genesis_readme_paths.t, genesis_base_sources_staged.t -> kept in the
.t (README read by slurp_repo_file, tarball read by Archive::Tar)
- genesis_openembedded_{export,hardware,runtime}.t -> the bats file of
the same name
- genesis_openembedded_{bmcsetup,metrics,extensions,qeth,actions,
discovery}.t -> scripts to the bats file of the same name, recipes
and units kept in the .t
- genesis_openembedded_console.t -> built and run checks, state.c,
plain_ui.c and shell.c to genesis_openembedded_console.bats; Newt
source checks kept in the .t
- genesis_openembedded.t -> oe/build to genesis_openembedded_build.bats;
shell matches to the runtime, hardware and extensions bats; console
matches to the console bats or the console .t; artifacts kept
Red proof: every moved or new test was run against a mutated copy of
the code it covers and went red, for example: rpm -q not checked,
tar without --mtime, VEX report not exported, support dir after PATH,
digest check off, layer2 forced, wildcard device allowed, arguments
eval'd, tcp probe skipped, SOURCE_PORT 302, AF_INET only, stage timer
compared, launch error ignored, kas run twice, status written in place.
One mutation stayed green: tar without --owner=0 when the test runs as
root.
Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
Deleting xCAT-genesis-builder/README.md removed 99 lines that describe the
build path this branch keeps: the kas virtualenv, the architectures `oe/build`
accepts, the console keys, `oe/export`, the signed-extension workflow with
`oe/export-extension`, and the `oe/report` baseline comparison. Only the legacy
RPM builder half was dead. The reader is now told nothing about how to build a
Genesis image.
xCAT-genesis-base/README.md carries that half with its paths on the new
directory. The legacy RPM builder chapter stays deleted.
go-xcat keeps xCAT-genesis-builder in GO_XCAT_UNINSTALL_LIST. No build
produces that rpm any more, but a management node installed before 2.19 still
has it, and the list is what `go-xcat uninstall` hands to the package manager.
genesis_readme_paths.t asserts every in-tree path the README names exists, and
that its link to the architecture plan resolves. It reads the file because the
file is the artifact: the text is the instruction. It matches any xCAT-*
directory rather than the new name alone, so a path left on the old name fails
instead of going uncollected, and it bails below a fixed floor so a pattern
that stops matching cannot pass on an empty plan.
Signed-off-by: Daniel Hilst <392820+dhilst@users.noreply.github.com>
Daniel Hilst