Summary
Three live memory RPC handlers have no test naming them anywhere in this crate. They are not leftovers from a removed domain — all three are registered controllers the desktop app calls today, one of them on a 5-second poll. They lost their coverage as collateral in #6161, which removed the memory engine from the test build.
| Handler |
RPC method |
Reached from |
memory::sources::rpc::add_rpc |
openhuman.memory_sources_add |
app/src/services/memorySourcesService.ts:93 |
memory::sync::sync_status::rpc::status_list_rpc |
openhuman.memory_sync_status_list |
the sync status table |
memory::sync::composio::providers::slack::rpc::sync_status_rpc |
Slack sync status |
the sync status surface |
Why the coverage went
All three were exercised only by tests/raw_coverage/memory_threads_raw_coverage_e2e.rs and memory_sync_tree_round21_raw_coverage_e2e.rs — 4,800-line targets that drove ~40 subjects at once through an in-process engine. #6161 deleted those targets with the engine, on the correct ground that most of what they asserted was the engine's behaviour and is covered upstream in tinymemory.
What went with them, and should not have, is the part that was never the engine's: this host's own policy in these handlers. No upstream work covers it, because upstream does not have these handlers.
A sweep of every deleted test file for host RPC handlers left with no surviving reference found 13. Seven are 5–17 line passthroughs to the driver and are legitimately upstream's to cover. Three were restored in #6170 (top_entities_rpc, chunks_for_entity_rpc, delete_chunk_rpc). These are the remaining three.
What is untested, per handler
status_list_rpc — degrades every failure path to an empty list: a binding error, a driver that does not serve SourceSync, and a driver call that fails. The handler's comment states this is inherited behaviour kept deliberately, because "this surface renders a status table, and every caller of it today treats 'no rows' as 'nothing syncing'". A later tightening of any of the three into an RPC error would turn a 5-second poll into a visible failure on the Memory Sources screen, and nothing would catch it.
sync_status_rpc — filters to ACTIVE slack connections only, and answers a specific degraded zero-value shape when per-connection detail is no longer readable: per_channel_cursors == "{}", synced_ids_count == 0, requests_used_today == 0, daily_request_limit == 0, plus a log line explaining the degraded read. The deleted test asserted exactly this and named it slack_sync_status_rpc_reports_the_degraded_zero_value_shape.
add_rpc — generates the source id (src_<uuid>), maps the request into a MemorySourceEntry, and calls apply_kind_defaults so an add that leaves caps unset gets conservative per-kind defaults rather than zeros. apply_kind_defaults itself lives in tinymemory-sources and is upstream's to test; what belongs here is that this handler invokes it, and that the id and field mapping are what the registry receives.
Scope
Sibling *_tests.rs files already exist at all three sites, so this is adding test functions, not new modules or new production code:
src/openhuman/memory/sources/rpc_tests.rssrc/openhuman/memory/sync/sync_status/rpc_tests.rssrc/openhuman/memory/sync/composio/providers/slack/rpc_tests.rs
Follow the pattern memory/read_rpc/admin_tests.rs established and memory/read_rpc/entities_tests.rs (added in #6170) extends: bind a driver with binding::install_for_test, then assert how this host answers, not what a store contains. NullMemoryProvider covers the family-absent paths; a small local double covers the ones that need a driver to answer a specific way.
Not in scope
- The seven thin passthroughs (
get_rpc, update_rpc, remove_rpc, chunk_score_rpc, entity_index_for_rpc, list_items_rpc, read_item_rpc). Each is 5–17 lines of binding → member → map, and the contract behaviour underneath is covered by tinymemory's conformance suite.
- Anything in
tinymemory. The engine side has no gap here — every engine module the deleted tests exercised has upstream coverage.
Acceptance criteria
Context
Correction (while implementing)
Two same-named status_list_rpc handlers exist, and the table above originally named the wrong one:
memory/sources/rpc_part_02.rs::status_list_rpc() serves memory_sources_status_list — the 5-second UI poll — and is a 5-line passthrough to status::status_list. Not worth host coverage; it belongs with the seven thin ones listed as out of scope.memory/sync/sync_status/rpc.rs::status_list_rpc(config) serves memory_sync_status_list and carries the documented three-path degradation. That is the one covered.
Also: this issue reported sync_status_rpc as having one test file naming it. That was a false positive — the only hit is a comment at slack/rpc_tests.rs:22. Zero real coverage, as claimed.
Summary
Three live memory RPC handlers have no test naming them anywhere in this crate. They are not leftovers from a removed domain — all three are registered controllers the desktop app calls today, one of them on a 5-second poll. They lost their coverage as collateral in #6161, which removed the memory engine from the test build.
memory::sources::rpc::add_rpcopenhuman.memory_sources_addapp/src/services/memorySourcesService.ts:93memory::sync::sync_status::rpc::status_list_rpcopenhuman.memory_sync_status_listmemory::sync::composio::providers::slack::rpc::sync_status_rpcWhy the coverage went
All three were exercised only by
tests/raw_coverage/memory_threads_raw_coverage_e2e.rsandmemory_sync_tree_round21_raw_coverage_e2e.rs— 4,800-line targets that drove ~40 subjects at once through an in-process engine. #6161 deleted those targets with the engine, on the correct ground that most of what they asserted was the engine's behaviour and is covered upstream intinymemory.What went with them, and should not have, is the part that was never the engine's: this host's own policy in these handlers. No upstream work covers it, because upstream does not have these handlers.
A sweep of every deleted test file for host RPC handlers left with no surviving reference found 13. Seven are 5–17 line passthroughs to the driver and are legitimately upstream's to cover. Three were restored in #6170 (
top_entities_rpc,chunks_for_entity_rpc,delete_chunk_rpc). These are the remaining three.What is untested, per handler
status_list_rpc— degrades every failure path to an empty list: a binding error, a driver that does not serveSourceSync, and a driver call that fails. The handler's comment states this is inherited behaviour kept deliberately, because "this surface renders a status table, and every caller of it today treats 'no rows' as 'nothing syncing'". A later tightening of any of the three into an RPC error would turn a 5-second poll into a visible failure on the Memory Sources screen, and nothing would catch it.sync_status_rpc— filters toACTIVEslack connections only, and answers a specific degraded zero-value shape when per-connection detail is no longer readable:per_channel_cursors == "{}",synced_ids_count == 0,requests_used_today == 0,daily_request_limit == 0, plus a log line explaining the degraded read. The deleted test asserted exactly this and named itslack_sync_status_rpc_reports_the_degraded_zero_value_shape.add_rpc— generates the source id (src_<uuid>), maps the request into aMemorySourceEntry, and callsapply_kind_defaultsso an add that leaves caps unset gets conservative per-kind defaults rather than zeros.apply_kind_defaultsitself lives intinymemory-sourcesand is upstream's to test; what belongs here is that this handler invokes it, and that the id and field mapping are what the registry receives.Scope
Sibling
*_tests.rsfiles already exist at all three sites, so this is adding test functions, not new modules or new production code:src/openhuman/memory/sources/rpc_tests.rssrc/openhuman/memory/sync/sync_status/rpc_tests.rssrc/openhuman/memory/sync/composio/providers/slack/rpc_tests.rsFollow the pattern
memory/read_rpc/admin_tests.rsestablished andmemory/read_rpc/entities_tests.rs(added in #6170) extends: bind a driver withbinding::install_for_test, then assert how this host answers, not what a store contains.NullMemoryProvidercovers the family-absent paths; a small local double covers the ones that need a driver to answer a specific way.Not in scope
get_rpc,update_rpc,remove_rpc,chunk_score_rpc,entity_index_for_rpc,list_items_rpc,read_item_rpc). Each is 5–17 lines ofbinding → member → map, and the contract behaviour underneath is covered bytinymemory's conformance suite.tinymemory. The engine side has no gap here — every engine module the deleted tests exercised has upstream coverage.Acceptance criteria
status_list_rpc: all three degradation paths assertOkwith an empty list, not an error.sync_status_rpc:ACTIVE-only filtering and the four zero-value fields are asserted.add_rpc: an add with caps unset reaches the registry with defaults applied and a generated id.cargo clippy -p openhuman --features "$(bash scripts/ci/product-features.sh)" -- -D warningsclean.Context
tinyhumansai/tinymemory#149— unrelated and upstream: sync families driven over the bus, which is new coverage rather than restored coverageCorrection (while implementing)
Two same-named
status_list_rpchandlers exist, and the table above originally named the wrong one:memory/sources/rpc_part_02.rs::status_list_rpc()servesmemory_sources_status_list— the 5-second UI poll — and is a 5-line passthrough tostatus::status_list. Not worth host coverage; it belongs with the seven thin ones listed as out of scope.memory/sync/sync_status/rpc.rs::status_list_rpc(config)servesmemory_sync_status_listand carries the documented three-path degradation. That is the one covered.Also: this issue reported
sync_status_rpcas having one test file naming it. That was a false positive — the only hit is a comment atslack/rpc_tests.rs:22. Zero real coverage, as claimed.