fix(sonic): emit up/down for BGP_NEIGHBOR_AF - #2618

Problem

The SONiC config generator wrote {"admin_status": "true"} for every
BGP_NEIGHBOR_AF row. That field is typed stypes:admin_status in
the AF grouping, and the typedef is a strict enumeration:

typedef admin_status {
    description "Administrative status (up/down)";
    type enumeration { enum up; enum down; }
}

— sonic-types.yang.j2#L88-L94

So "true" is outside the type, not a lenient spelling of it. Upstream's own
YANG test config for this table uses "up", with a real boolean
(send_default_route) immediately beside it, so the model distinguishes
deliberately: tests_config/bgp.json#L387-L395

Every other table this generator writes — PORT, PORTCHANNEL, LOOPBACK,
VLAN, VLAN_INTERFACE and the MGMT_ tables — already emits up/down.
BGP_NEIGHBOR_AF was the lone outlier.

Why it is not cosmetic

config reload runs YANG validation and aborts on failure, via
config_file_yang_validation. On sonic-utilities master the
no-filename form validates /etc/sonic/config_db.json:

if filename is None and file_format == 'config_db':
    if not force:
        ... if not config_file_yang_validation(cfg_file): raise click.Abort()

— config/main.py#L2226-L2258

That is exactly how this project applies a generated config (config load
followed by config reload -y), so "true" becomes a hard failure in a
future release.

On 202505 the validator exists but is reached only under
if filename is not None, the explicit-filename path:
config/main.py#L1961-L1967. So an explicit-filename reload
already aborts today, while the no-filename form does not. The block quoted
above is what master adds.

Every link is pinned to a commit SHA so the line ranges stay valid:
sonic-buildimage 8aaea5e3649a, sonic-utilities
e3768cbdc0f4 and 7511bf2f8ea3 (202505).

Impact

Behaviour on the switch is unchanged: frrcfgd's hdl_admin_status converts
up/down to true/false internally, so the consumer already expects this
spelling: frrcfgd.py#L1456-L1466

Switches running a previously generated config hold "true" and will show the
new value as a diff the next time their config is regenerated. Harmless, for
the same reason.

The bundled ConfigDB validator rejected every generated AF row before this
change (Input should be 'up' or 'down') and accepts them now. That removes the
largest source of noise from osism sonic validate for these tables, but the
command is not yet usable as a gate: PORT.adv_speeds is modelled as a list
while ConfigDB carries a string, and BGP_NEIGHBOR.local_addr is flagged
against its union's leafref arms while a plain inet:ip-address arm makes the
value legal. Both are separate.

Tests

Twelve assertions in the BGP test module are updated to match. 712 passed in
tests/unit/tasks/conductor/sonic/, 3139 passed in tests/unit (4
pre-existing xfails), flake8 and black clean.

Note for reviewers

No committed golden pins the old value: the SONiC E2E goldens are not in
main, they exist only on the draft stack rooted at:

• Add synthetic base fixtures and the SONiC E2E job #2566

The matching golden update belongs to that stack and lands with it, so this
change breaks nothing currently in main.

🤖 Generated with Claude Code