Extend the module E2E to the declared surface and to durability (#18 §E5) - #29
Merged
YellowSnnowmann merged 13 commits intoAug 18, 2026
Conversation
`rust_out` and `tmp` are 3.6 MB Linux x86-64 ELF executables with debug info, committed into a repository developed on macOS. Nothing references either name in source, config, CI, or docs; both are the output of an ad-hoc build that was staged by accident, and every clone has carried 7.3 MB of them since. Ignored with anchored paths so a legitimate `tmp` directory inside a crate is unaffected.
The workspace declared three different answers: 1.96 at the root and for the module crate, 1.85 for `core` and both adapters, and nothing at all for `api`. The low ones were not true — the root facade requires 1.96 and depends on all of them, so nothing in this workspace builds on 1.85, and no CI job checked the claim. Aligning on 1.96 immediately surfaced three lints that the understated MSRV had been suppressing: clippy only suggests `is_multiple_of` when the declared minimum supports it. Those are fixed here rather than allowed, since they are a consequence of this change and not separable from it. `api/` gaining an explicit MSRV is worth noting for downstream: it is the dependency-light crate an embedding host binds directly, so it now states a floor rather than leaving consumers to infer one.
The layout section omitted `tinymemory-core` entirely — the largest crate in the repository, and the one a real host actually depends on. It also omitted `crates/tinymemory-module`. Adds both, and states plainly that `core` is not dependency-light the way `api` is, since that asymmetry is the thing a reader most needs to know before choosing which to depend on. Also documents `git submodule update --init --recursive`. Nothing builds without it, and because `core` names its engines by path through `vendor/`, an uninitialized checkout fails at manifest resolution with an error that does not mention submodules.
`audit_provider` checks that a driver's advertised capabilities match its reachable accessors — a structural check, proving the shape is honest, not the behaviour. Nothing checked that two drivers answer the same question the same way, which is the claim "swap the engine without the host learning anything new" actually rests on. `tinymemory-conformance` is that check. `assert_provider` takes any bound driver and drives the contract: the mandatory three families, upsert on `(namespace, key)`, namespace isolation, provenance preservation, recall limits and scoping, export pagination, import round-tripping, and unicode / empty / 64 KiB / control-character content. It depends on `tinymemory-api` and nothing else of substance. A suite that pulled in an engine could not prove interchangeability, having already chosen one; reaching `tinymemory-core` would drag in a bundled SQLite besides. Ships an in-memory reference driver, for two reasons. A suite that only ever ran against real engines cannot distinguish "the engine is wrong" from "the assertion is wrong", and a driver whose behaviour is obvious by inspection separates those. It also documents the contract by example — the upsert, the verbatim taint, the cursor that terminates on `None` rather than on an empty page. Writing it immediately found a distinction worth naming. The contract permits a driver that accepts writes and discards them; `NullMemoryProvider` is exactly that, and the storage assertions are vacuous for it. Rather than weakening each assertion to tolerate an empty read — which would let a driver that *intends* to retain and silently does not slip through — the suite probes retention once and reports which half it ran. The contract-shape assertions are never skipped. Provenance is the sharp assertion. A driver reading back `Internal` for content stored as `ExternalSync` has laundered external content into internal-trust content, and every downstream gate keyed on taint is then silently wrong. Refs tinyhumansai#18 (§E1)
Issue tinyhumansai#18 §E3. `AGENTS.md` mandates a `tests/` directory exercising only the public API; the repository had none at the root, and `core/tests/` holds fixtures with no test files. Four targets, matching the four §E3 names: `driver_selection.rs` pins admission: reserved ids resolve to fixed classes, an external driver with no entry is refused fail-closed, an untrusted external driver is refused even with one, a reserved id's class cannot be overridden by config, and a class typo is echoed back so the operator can find the line. `capability_negotiation.rs` covers both directions of the bind-time negotiation, including a deliberately lying provider that advertises a summary tree it has no accessor for — the failure `audit_provider` exists to catch, and which was previously asserted only in unit tests inside the contract crate. `taint_end_to_end.rs` drives provenance through store, get, list, recall, and the export/import round trip, at every driver this workspace ships. It also pins the fail-closed reading of an unknown persisted value, which is the one direction that cannot be undone. `null_provider.rs` asserts the compiled-out configuration is genuinely usable: every mandatory method answers rather than panicking, it reports Ready rather than a fault, and no optional family is either advertised or reachable. Two of the four are deliberately narrower than §E3 describes, and both say so in their module docs. `driver_selection.rs` cannot yet assert that a bound provider's `driver_id()` matches configuration, because nothing selects an engine from config — that is §A5. `taint_end_to_end.rs` cannot drive the sync path, because sync is welded to the engine until §B. Both are written against current behaviour, per the sequencing note in the issue, and each names where its missing leg joins. Refs tinyhumansai#18 (§E3)
Issue tinyhumansai#18 §C3. The adapter advertised Core, Recall and Portability and nothing else, which is the reason anything wanting a summary tree, entities, or a diff ledger reached past the contract to the engine directly — the families existed, but not through `MemoryProvider`. `crates/tinymemory-module` had grown all of them, because it needed them and nowhere else had them. They were never module-specific. Every one delegates to `tinymemory-core` on a blocking thread, and the two types they hold — `MemoryClient` and the host config — are core and contract types respectively. So the whole `ModuleMemoryProvider` moves down to `tinymemory-tinycortex` as `engine::TinycortexProvider`, and the module crate keeps only the thing that genuinely is its own: turning a `ModuleConfig` into the engine's runtime configuration. Its `provider.rs` goes from 2189 lines to 38. Nineteen family implementations move: documents, ingest, graph, goals, tool-memory, tree, entities, diff, sources, maintenance, people, chunks, retrieval, profile, and episodic, alongside the mandatory three. The diff family gets a `memory-git` feature rather than riding along unconditionally. It is what drags `git2` / `libgit2-sys` / `libz-sys` — a native build — into the graph, and this adapter had no such dependency before today; making it unconditional would hand every consumer a libgit2 build they never asked for. `cargo tree --no-default-features` confirms none is linked. The gate reaches `capabilities()`, not just the accessor. A build without `memory-git` neither advertises nor reaches `Diff`, so `audit_provider` still passes — which is the whole reason that audit exists. That rule is extracted as `engine::advertised_capabilities` so it can be tested directly: constructing a provider needs a `MemoryClient`, which needs the host's process-global seams installed, and a test that installs a process global is order-dependent. The new module is `engine`, not `provider`: the crate already has a `provider` function returning the mandatory-only driver, and both are worth keeping — a host with no workspace, config, or client still has the lighter one. Refs tinyhumansai#18 (§C3)
Issue tinyhumansai#18 §A5. The driver registry could answer "is this driver id real, and is it allowed to answer for memory", and nothing asked it: `DriverRegistry::admit` had no caller outside its own tests, `MemoryHostConfig::memory_provider()` had no reader, and the memory client factory constructed TinyCortex unconditionally. Configuration could not choose an engine. `DriverRegistry::select` closes that: it reads the engine from the host's configuration and puts it through `admit`, so both surfaces are now live. Two corrections to the issue, both load-bearing. §A5 names `memory_provider()` as the selector. That method is a `provider:model` routing string for the memory *workload* — which language model does summarisation and entity extraction — not the store the memory lives in. Reading it would have let a model change repoint a company's storage. Selection reads a new `memory_driver()` instead, defaulted to `None` so it breaks no existing implementation, and a test pins that the two fields stay independent. §A5 also asks that `create_memory_*` return a bound `Arc<dyn MemoryProvider>`. It cannot, and the reason is structural rather than unfinished: since §C3 `adapters/tinycortex` depends on `tinymemory-core`, so a core factory returning a constructed adapter provider is a dependency cycle. Selection therefore resolves the decision and the host constructs — which is what `src/registry`'s module docs have said all along: "It resolves the class, not the instance." A configuration naming no engine gets the reserved embedded default, so adding selection does not turn "I configured nothing" into a host that fails to start. Going through `select` does not loosen admission either: an external engine named in config is still refused without endpoint, credential and trust. Refs tinyhumansai#18 (§A5)
The `memory-git` feature added alongside the lifted diff family gates that family, and the test asserting it is *withheld* without the feature is `#[cfg(not(feature = "memory-git"))]`. Both existing jobs — `--all-features` and default — compile that test out, so it was checked by nothing: a feature-gated test whose default fate is to be built by one job and executed by none. Adds the configuration as its own step, and asserts the property the feature exists for: `cargo tree --no-default-features` must link no `git2` or `libgit2-sys`. A guard rather than a comment, because "this feature keeps the native build out of the graph" is a claim that silently stops being true the first time a dependency picks it up transitively. Verified: `diff_is_withheld_when_the_snapshot_store_is_compiled_out` appears in `--no-default-features -- --list` and is absent from the `--all-features` listing, which is what "running nowhere" looked like. Refs tinyhumansai#18 (§C3, §E2)
Issue tinyhumansai#18 §D4, with §D3 alongside it. `api/Cargo.toml` spells out a `cargo tree` command in a comment and asks that the contract crate never link a storage engine, a native library, an HTTP client, or an async runtime. It was left as a comment, so nothing ran it. A forbidden dependency does not arrive by someone typing it into the manifest; it arrives transitively, through a feature enabled two crates away, which is exactly the way nobody notices. The forward form is the one that works, and the manifest already explains why: `cargo tree -i <crate> -p tinymemory-api` discards the `-p` scope, prints the whole-workspace inverse tree, and exits 0 looking clean even when this crate is the one at fault. This runs what the comment says to run. Verified in both directions. The rule holds today — no match against `rusqlite|libsqlite|git2|reqwest|regex|tokio`. And injecting `regex = "1"` into the manifest makes the guard fire on `regex`, `regex-automata` and `regex-syntax`, then reverting makes it pass again. A guard nobody has watched fail is not yet a guard. §D3 asks that `--no-default-features` still compile and bind `NullMemoryProvider`. It does, so this pins it rather than changing anything: the minimal configuration builds, and tinyhumansai#21's `null_provider` integration test runs against it — so "usable" is asserted, not just "compiles". That test is the one that would catch a null driver which panics instead of answering, or reports a fault instead of `Ready`. Refs tinyhumansai#18 (§D3, §D4)
Issue tinyhumansai#18 §E6: "backend error, timeout, and partial-page responses on every remote adapter — currently zero coverage". The three existing per-adapter test files all drive a backend that answers correctly, which is the half that was never in doubt. Six tests, each running against all three adapters over a real TCP socket, using the axum-double harness the happy-path tests already use: - a 500 on write is reported rather than swallowed; - a 500 on read is not laundered into `Ok(None)`; - a 401 is not presented as an empty store, on either `list` or `recall`; - a `200 OK` carrying unparseable JSON is an error and not a panic; - an unreachable backend is reported rather than hanging; - a paginated export terminates instead of looping. The read case is the one that matters. `Ok(None)` after a 500 says "this memory does not exist" when the truth is "I could not ask", and a caller cannot tell those apart: it writes the memory again, or tells a user their memory is gone, or a sync job treats the empty read as authoritative and prunes. Nothing surfaces until much later. All three adapters already behave correctly — this pins behaviour rather than fixing it. That is worth stating plainly, because six tests passing on the first run is exactly what a test that never exercises its subject also looks like. So the read assertion additionally requires the backend's status to survive into the error message; without that it would pass just as happily if the adapter had failed on URL construction and never reached the network. Confirmed against the live errors: `memory API v3/container-tags/list returned HTTP 500`, `memory API memories?top_k=1000 returned HTTP 500`, and `memory API api/v1/datasets returned HTTP 500`. The assertions are deliberately about whether a failure comes back at all, not about which error it is. The adapters' HTTP layer `bail!`s into `anyhow`, so every one of these arrives as `MemoryError::Other` and "unsupported" is not yet distinguishable from "failed" — that is §A4, and it is not what this change is. The unreachable-backend test binds a port, reads its number, and drops the listener, so the address is reliably closed rather than merely unlikely to be in use. The export test is bounded by a timeout so a non-terminating implementation fails the test instead of hanging the suite. Refs tinyhumansai#18 (§E6)
Issue tinyhumansai#18 §E7. `AGENTS.md` documents an `examples/` directory and tells the reader to run `cargo run --example basic`. Neither existed: no `examples/` directory at all, so the documented command has been failing for as long as it has been documented. (The issue attributes the reference to `README.md`; it is `AGENTS.md` lines 30 and 74.) The example binds the null driver, which needs no engine, no workspace and no network, so it runs anywhere. It walks the order a host actually follows — admit an id, then construct, then check the negotiated capabilities, then use the mandatory families — because that order is the part worth demonstrating. Admission is engine-neutral and answers "is this id real and may it answer for memory"; construction needs everything an engine needs. Swapping `NullMemoryProvider` for an adapter's provider changes nothing else in the file. Writing it surfaced an API gap, which is the argument for having a compiled example at all: `FallbackReason` implemented `Display` but not `std::error::Error`, so the obvious `registry.admit(..)?` in a function returning `Box<dyn Error>` or `anyhow::Error` did not compile. Adding the impl is purely additive and changes nothing about the type; it makes the message usable where refusals actually travel. CI runs the example rather than only building it. `cargo build --all-targets` compiles examples, so a broken one still passes — and a compiled example can panic on its first line. Running it is what makes the documented command a promise rather than a comment. Not attempted here: §E7's "one example per engine". The three hosted engines need a live endpoint and a credential, which is why `adapters/remote/examples/conformance.rs` is a manual CLI rather than a CI target, and the embedded engine needs the host's process-global seams installed. Neither belongs in a `cargo run --example` a contributor is told to run. Refs tinyhumansai#18 (§E7)
Issue tinyhumansai#18 §D5, §E8, and the second half of §E2. Three CI additions that all answer the same kind of question — what is this build actually costing, and is it still true — so they land together. **Feature powerset (§E2).** `cargo hack --feature-powerset --depth 2`, check only. Cargo features are additive: enabling one for a crate enables it for every consumer in the graph, so a combination nobody builds deliberately can still be built by somebody else's dependency. `--depth 2` covers every pair without the blow-up of the full set. §E2 also lists per-feature lanes — `--features tinycortex`, `mem0`, `supermemory`, `cognee`, `sync-composio`. Those are not here because those features do not exist yet: they are §D1, and §D1 needs OpenHuman to opt into features it currently gets unconditionally. The powerset covers whatever features exist, so it starts useful and stays useful as §D1 adds them. Verified by hand across all eight combinations the workspace can express today (core: none, contacts, memory-git, test-support and their pairs; the tinycortex adapter with and without memory-git). Every one compiles, so this pins a property that currently holds rather than papering over a break. **Dependency budget (§D5).** `scripts/ci/dependency-budget.sh` prints the crate count for every configuration and fails when the minimal one grows past a ceiling. Today: minimal 40, api 39, the tinycortex adapter 168 — and 172 with `memory-git`, which is the native git stack the feature exists to keep out, so the +4 is the gate from §C3 working. The ceiling is 50 against a current 40, deliberately. A limit set at today's exact count fails on the first legitimate addition, gets raised without thought, and teaches everyone to ignore it. Only the minimal configuration is gated; the richer numbers are reported, because a number nobody chose is not a budget. **Coverage (§E8).** `AGENTS.md` asks for 80% of meaningful library behaviour and nothing measured it. The workspace is at **76.22% lines / 76.44% regions / 63.81% functions** — below the number it asks for, which is worth seeing rather than assuming. Some of the gap is stark: `core/src/tree/score/store.rs` and `core/src/tree/score/extract/mod.rs` are at 0.00%. Reported, not enforced, to begin with. A threshold picked before anyone has seen the number is a guess, and a gate that fails on day one gets disabled rather than fixed. The powerset and coverage run as their own job: both are slower than the main lane and independent of it, so a failure in one should not mask the other or delay the fast feedback the main job gives. Refs tinyhumansai#18 (§D5, §E2, §E8)
Issue tinyhumansai#18 §E5. The loader E2E drove three of the eighteen families the module advertises, and nothing checked that a write reached the workspace it was given. **Every declared method is routed.** The module advertises `Capabilities::all()` and declares 88 methods; the existing manifest test compares two lists, so it passes for a method that is declared, listed, and answers "unknown member" — the bus-level version of a capability set that overstates its accessors. This calls each declared method and distinguishes the kind of refusal: a wired method rejects the empty argument list with `ai.tinyhumans.tinybus.Error.BadArguments`, while an unwired one is `ai.tinyhumans.tinybus.Error.UnknownMethod`. The two names are what make the test discriminate rather than pass vacuously, and the unrouted shape is taken from the build under test rather than hard-coded. Eighteen bespoke round trips would assert more, and would also need eighteen sets of valid arguments and eighteen engine preconditions. This asserts the one thing that is true of all of them and is cheap to keep true. **A write lands in the host's workspace.** Every other test here stores and reads back inside one admission, so a module that kept its store in a temporary directory of its own, or in memory, passes all of them — and the difference shows on the user's next launch. §E5 asks for a shutdown/restart cycle, and it is not here because it cannot be: TinyBus never unloads a library, so a second admission in the same process is refused with `ModuleRefused { reason: "module initialization failed" }`. That is the same constraint that already forces every test in this file to be the only one in its process. A real restart needs a second process against a shared workspace, which is a change to the CI loop rather than a test, so the directory assertion covers what restart would have been checking and the limitation is written down rather than left as a gap someone rediscovers. Verified the way CI runs them — every ignored test in its own process, all ten green. Refs tinyhumansai#18 (§E5)
|
Important Review available on request
Reviews should be triggered manually for repositories with fewer than 10 stars. Select Trigger review above or comment ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![tinysweeper[bot]](https://avatars.githubusercontent.com/in/4526585?s=60&v=4){kind=small}
| submodules: recursive | ||
| persist-credentials: false | ||
|
|
||
| - uses: dtolnay/rust-toolchain@stable |
There was a problem hiding this comment.
dtolnay/rust-toolchain is pinned to stable, which is mutable
A tag or branch can be repointed by whoever owns dtolnay, and the new code runs with this workflow's secrets. Pin to a full commit SHA and let Dependabot bump it.
[RULE] unpinned-action ·
| with: | ||
| components: llvm-tools-preview | ||
|
|
||
| - uses: Swatinem/rust-cache@v2 |
There was a problem hiding this comment.
Swatinem/rust-cache is pinned to v2, which is mutable
A tag or branch can be repointed by whoever owns Swatinem, and the new code runs with this workflow's secrets. Pin to a full commit SHA and let Dependabot bump it.
[RULE] unpinned-action ·
|
|
||
| - uses: Swatinem/rust-cache@v2 | ||
|
|
||
| - uses: taiki-e/install-action@v2 |
There was a problem hiding this comment.
taiki-e/install-action is pinned to v2, which is mutable
A tag or branch can be repointed by whoever owns taiki-e, and the new code runs with this workflow's secrets. Pin to a full commit SHA and let Dependabot bump it.
[RULE] unpinned-action ·
How this change flows0 changed behaviours across 1 relationship. 2 surrounding behaviours are shown (60 graph nodes walked). 37 further behaviours left out to keep the diagram readable. flowchart LR
n0["admit"]:::impacted
n1["select"]:::impacted
n1 -->|calls| n0
classDef changed fill:#0d4429,stroke:#238636,color:#e6edf3
classDef impacted fill:#161b22,stroke:#6e7681,color:#c9d1d9
classDef flagged fill:#5a1e02,stroke:#d93f0b,color:#ffffff
classDef blocking fill:#67060c,stroke:#f85149,color:#ffffff
Green: changed behaviour. Grey: surrounding behaviour. Arrows name the call, use, implementation, or test relationship. Orange: has findings. Red: has a finding that blocks the merge. |
tinysweeper
Bot
added
the
priority: p2
This was referenced Aug 17, 2026
M3gA-Mind
reviewed
Aug 17, 2026
M3gA-Mind
left a comment
M3gA-Mind
left a comment
There was a problem hiding this comment.
Reviewed this PR's own commit (3c083a8). Clean.
Two tests added to module_e2e.rs: what_is_written_lands_in_the_workspace_it_was_given (durability) and every_declared_method_is_actually_routed (the declared surface). Both match what the title claims and what §E5 asks for.
Both are #[ignore]d, which is normally where I would expect to find a test that runs nowhere — an ignored test is invisible to a plain cargo test. It is handled correctly here. The existing lane enumerates with --ignored --list, runs each --exact in its own process under timeout 300, and — the part that matters —
if [ -z "$tests" ]; then
echo "No ignored E2E tests were found — the list step is broken." >&2
exit 1
fifails when the enumeration comes back empty. That guard is exactly what makes #[ignore] safe to rely on: without it, a rename or a --test target typo would silently reduce the lane to zero tests and still exit 0. Good that it was already there and that this PR's additions slot into it without special-casing.
One thing I could not confirm from the diff: §E5 also asks for a shutdown/restart cycle. what_is_written_lands_in_the_workspace_it_was_given covers durability of a write, but whether it exercises a restart — i.e. tears the module down and reloads it against the same workspace — is UNVERIFIED from reading the patch alone. If it does not, worth noting on #18 that the restart leg is still open; if it does, worth a word in the test's doc comment, since "durability" and "survives a reload" are different properties and only the second one catches state held in the module process.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Issue #18 §E5. The loader E2E drove three of the eighteen families the module advertises, and nothing checked that a write reached the workspace the host gave it.
Every declared method is routed
The module advertises
Capabilities::all()and declares 88 methods. The existingthe_manifest_declares_every_method_the_module_servescompares two lists — so it passes for a method that is declared, listed, and answers "unknown member". That is the bus-level version of a capability set overstating its accessors, which is exactly whataudit_providerexists to catch one layer up.This calls each declared method and discriminates on the kind of refusal:
ai.tinyhumans.tinybus.Error.BadArgumentsai.tinyhumans.tinybus.Error.UnknownMethodThose being two distinct names is what makes the test discriminate rather than pass vacuously — I checked before shipping it, since a test that passes on its first run is also what a vacuous one looks like. The unrouted shape is derived from a live call in the build under test, not hard-coded, so it cannot drift from tinybus's spelling.
Eighteen bespoke round trips would assert more — and would need eighteen sets of valid arguments and eighteen engine preconditions. This asserts the one property true of all of them, at a cost that keeps it true.
A write lands in the host's workspace
Every other test in the file stores and reads back inside a single admission. A module that kept its store in a temporary directory of its own, or in memory, passes all of them. The difference shows on the user's next launch.
So: count the workspace directory before admission, store over the bus, count again.
The restart cycle is absent, and cannot be written here
§E5 asks for one. I wrote it, ran it, and it fails structurally:
TinyBus never unloads a library, so a second admission in the same process is refused. That is the same constraint that already forces every test in this file to be
#[ignore]d and run one-per-process — the file's own module docs describe it.A genuine restart needs a second process against a shared workspace, which is a change to the CI loop rather than a test. The workspace-directory assertion covers what restart would have been checking (the write is durable and in the right place), and the limitation is written into the test's comment rather than left as a gap for someone to rediscover.
Public API / behavior changes
None. Tests only.
Validation
Run the way CI runs them — every ignored test in its own process:
cargo fmt --manifest-path crates/tinymemory-module/Cargo.toml --all -- --checkcargo clippy --manifest-path crates/tinymemory-module/Cargo.toml --all-targets -- -D warningscargo test --manifest-path crates/tinymemory-module/Cargo.toml --libcargo build --locked --manifest-path crates/tinymemory-module/Cargo.toml --releasecargo test --all-featuresRelated
Part of #18 (§E5).