class-of-service: add show class-of-service rewrite-rule (#6848)

[psaab]

Closes #6848. Advances #4228 (Gap 7 residual).

The gap

show class-of-service had interface, classifier, scheduler-map and
forwarding-class. rewrite-rule — the fifth Junos CoS show command — was
never wired.

It matters more now than when #4228 was written. At that point rewrite-rules
held only dscp, so the missing command hid one table. Since then the config
models three more families — ieee-802.1 (#4228 Gap 4), inet-precedence and
exp (#4316) — and all three are accepted-but-inert: they commit clean and
have no runtime effect, because the dataplane rewrites DSCP on egress only.

So an operator could configure four kinds of rewrite rule, three of which do
nothing, with no command to display any of them. The only signal was a
commit-time advisory that scrolls past once.

Enforcement is a column, not a footnote

Rewrite rule: rw-dscp, Code point type: dscp, Enforced: yes
  Forwarding class  Loss priority  Code point
  best-effort       low            000000
  premium           low            101110

Rewrite rule: rw-pcp, Code point type: ieee-802.1, Enforced: no (accepted for
Junos compatibility; the dataplane rewrites dscp only)
  Forwarding class  Loss priority  Code point
  premium           high           101

Rendering an inert rule identically to an enforced one would reproduce, inside
the command built to expose the problem, exactly the silence that made it worth
fixing. The marker names what the operator loses rather than only that a flag is
off.

The four families do not carry equal data

dscp and ieee-802.1 compile to full entry lists. inet-precedence and exp
record only rule names — the compiler builds no runtime structure because
nothing consumes one. Those render Code points not modeled rather than an
empty table, which would imply a fidelity the config does not have.

TestShowTextCoSRewriteRuleNameOnlyFamiliesAreProducible6848 authors such a
rule with a code point in real set syntax and asserts the value does not
surface — so if the compiler ever starts modeling those entries, the test fails
and says to render them, rather than leaving a view that quietly under-reports
real config.

Both surfaces

The local CLI and the gRPC server are separate dispatchers. A command registered
in only one works interactively and silently falls through to help text on the
remote cli binary — the surface most operators use. Both are wired, sharing
one topic encoder rather than two that can drift.

Drive-by fix: completion promised a filter the parser did not implement

Both commands' cmdtree nodes offer rule names directly under the command, so
tab-completing a name and pressing enter submits a bare positional — which
the keyword-only parser ignored, dumping every rule. The shared parser now
accepts it, so classifier and rewrite-rule both do what their completion
advertises.

This deliberately touches the pre-existing classifier command. Leaving it
would mean two sibling commands with identical advertised grammar behaving
differently, which is worse than the gap. No test pinned the old
dump-everything behavior.

Drift guard

format.CoSRewriteRuleTypes is the SSOT for the type filter; the cmdtree
completion node is pinned against it. Without that, completion could offer a
value the renderer ignores (operator filters, gets everything) or a family could
render but be uncompletable (operator never finds it) — both quiet.

Validation

Full go test ./... green, scored from a real exit code (REAL_EXIT=0),
not a pipe.

The #6532 fabric secret-render audit caught the new topic as reachable from
the peer chassis over the cluster-fabric allowlist and required it be added to
showTextAuditTopics. A genuine cross-cutting catch, not a formality — ShowText
is peer-reachable, so every topic it renders is in scope for that audit.

Fail-on-revert by mutation, build+vet clean at each step so every RED is an
assertion, not a build break:

#	Mutation	RED
N1	every family reported enforced	inert pin (format + gRPC)
N2	negative control — dscp reported inert	enforced pin, both surfaces
N3	name-only families dropped from the renderer	4 format tests + 3 gRPC tests
N4	gRPC dispatch arm removed (remote unwired)	all 4 gRPC tests, unknown topic
N5	bare-positional name revert	local grammar pin
N6	exp dropped from cmdtree type	SSOT coherence pin

N2 is the control that makes N1 mean anything: without it, "some marker is
present" would satisfy the inert assertions.

Docs

docs/cos-validation-notes.md gains a section on the inert-rule problem, the
unequal-data split, and the rule that a family which starts being enforced must
flip cosRewriteRuleEnforced in the same change that drops its commit
advisory — otherwise this command reports a working rewrite as inert.

Not included

clear class-of-service statistics (#4228 Gap 7b) remains deferred: there is no
CoS-queue stat-reset RPC — counters reset only on config change — so it needs a
helper-side reset path.

🤖 Generated with Claude Code

https://claude.ai/code/session_015oARShYtiJJ2H4UB4nXGqi