feat(release): withhold partial backports and park those that build on code their line lacks - #5269

…inal

A release note describes what the original does on main. A pick git had
to fit to a line that already had part of the change, a resolution, or
a unit whose first commit came out empty no longer does the same thing,
yet its note went into the batch's release-note block with the rest,
and the changelog took it from there.

A unit now counts as changed when a surviving pick's patch-id differs
from its source or the unit's net patch-id differs from the original
merge's; the latter catches an empty first commit, which leaves no pick
to compare. A changed original's note moves under "Check these notes",
fenced and neutralised as before, with a range-diff per differing pick
and the two diffs to compare. The range-diff uses a creation factor of
999 because the default leaves a pick that lost most of its change,
#5088's test-only pick for one, unpaired. The record and the report
list such originals as changed.

Assisted-By: LLM
Signed-off-by: Myasnikov Daniil <[email protected]>

On 1.6, seven of the twelve originals rule 6 holds back build on code
the line never got: #5071 changes files #3940 added on main, #4974
reads a variable only #4386 defines. Each conflicts on every run, and
nothing tells a maintainer which missing pull request the conflict
comes from, which every human backport so far had to work out by hand.

For each original the policy takes and nothing carries, run now reads
what it builds on from git alone. A modified or deleted file the line
lacks is named by the first-parent commit on main that added it (V1).
A removed or changed line is blamed at M^1, mapped to the first-parent
commit that brought the blamed commit, and kept only while its text,
edges trimmed, is absent from the line's file (V2c), so a hand backport
under another SHA clears it. Generated files and the keys
update-crd.sh writes into an ApplicationDefinition are not blamed.

Paths go to git literally, blame gets a validated full SHA and no
--end-of-options, which git refuses there, and every call has a
two-minute deadline. Caps, deadlines, binary files, symlinks,
submodules, unmapped commits and skipped rename detection become
diagnostics shown with the original on every outcome and in the
report; malformed git output stops the run. This commit only
annotates; nothing is parked yet.

Assisted-By: LLM
Signed-off-by: Myasnikov Daniil <[email protected]>

…ne, and hold one whose dependency is pending

On 1.6, the fix in #5071 conflicts on every run: it modifies files
that the feat in #3940 added on main, and that feat never reaches the
line. The fix in #5082 builds on #5071 and on code that never comes
either. Trying them on each run says nothing a maintainer can act on,
and a conflict reads as work to do.

Edges come from the dependencies the previous commit reads. One whose
prerequisite the line or the open batch carries, whatever its rule or
label, that this run picked or found empty, or whose evidence is
present at the branch head once earlier picks moved it, is satisfied
and dropped first. A direct push, an original rules 2-4, the start, a
skip label or a drop keep out, or one itself parked never comes and
parks the original. A prerequisite still to come (a conflict, a
deferral, a workflow wait, rule 6) holds it as deferred{needs}, the
strongest kind deciding: missing, workflow, deferred, rule 6; a chain
inherits.

run, check and the metadata operations share one order: partial unit
on the batch, a direct reason, an unpicked partial, the workflow
exception, a dependency park, a wait, then the original's own state.
In the scheduled mode a workflow original whose remaining
never-dependencies are only tests or docs waits for a maintainer
instead of parking, which is the case of the release fix in #5142.
check takes the mode the open record saved, and the scheduled one
with no batch open, where a closed record's workflow waits wait on
nothing any more and the original is judged as the scheduled run
judges it. check and the record-only operations keep a partial park
the open record holds while the batch head still shows it, since only
the run's picker sees it after the picks before it. Records refuse
waits and parks that form cycles, point at later merges or contradict
each other. notify lists a wait after what decides it, from the
record alone. backport/include lifts neither park nor wait.

Assisted-By: LLM
Signed-off-by: Myasnikov Daniil <[email protected]>

A park is a decision a maintainer has to act on: port what applies as
a pull request of its own, or drop the original. Listing every park in
every notify comment would make that list noise within a week, and a
park nobody hears about is a fix that silently never ships.

notify now names each park the open batch's record holds that is not
yet marked announced, and once its comment is written marks those parks
in the record with gh pr edit --body-file. It writes only while the
pull request is open and its body still carries the record notify read,
so a run that rewrote the body in between loses nothing; a mark that
fails exits 2 and the next notify names the park once more. run carries
the mark into the next record, a merged or closed carrier's too, only
while the park is the same event: merge, kind, prerequisites, commits
and files. A park that changed is news and is named again. With no
batch open nothing is named (M2); the next batch's first notify does
it.

Assisted-By: LLM
Signed-off-by: Myasnikov Daniil <[email protected]>