wdevs/ResolveSpec
5
0
Fork 0
mirror of https://github.com/bitechdev/ResolveSpec.git synced 2026-10-02 03:22:09 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

fix: fire resolvespec AfterRead/AfterCreate/AfterDelete, key restheadspec total cache by record id

Browse Source
This commit is contained in:
warkanum
2026-09-30 23:40:32 +02:00
parent 5933637a88
commit a65ca5f5ce
8 changed files with 448 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files
audit/single_tran.md
+3 -2
View File
@@ -97,8 +97,8 @@
- DONE P7: `pkg/security/txsettings.go`: `SecurityList.SetTxSettings(fn)`, `StampTxSettings`, `ApplyTxSettings` (configurable map, decided by user; `set_config(name, value, true)`, value hex-encoded, name validated, Postgres only, fail closed). Every spec's `RegisterSecurityHooks` registers it on `OnTxBegin`. Tests: `pkg/security/txsettings_test.go`, `pkg/resolvespec/tx_settings_test.go`. Docs: `pkg/common/TRANSACTIONS.md`.
- DONE real-Postgres check (resolvespec, testserver via compose): create `tx=1 pooled=0`, read `tx=1 pooled=0`, update `tx=2 pooled=0` (was `pooled=1`), single delete `tx=1 pooled=0`, batch create/delete `tx=1 pooled=0`. Compose now uses host networking (bridge fails here): testserver on 8123, Postgres on 8124 (was 8080/5434); integration test DSNs updated. Smoke script covers read and update. websocketspec/mqttspec/resolvemcp/restheadspec/funcspec not measured on real Postgres.
- DONE regression tests: per-spec read/create/update/delete hook-on-tx, failure-rollback (Before*/After*/`OnTxBegin`) and second-tx tests in all six specs (`ops_tx_test.go`, `tx_test.go`, `read_tx_test.go`); stamping tests for resolvespec, resolvemcp, funcspec; pgsql adapter preload tests (same connection, error returned); source guard `pkg/common/tx_guard_test.go` (no direct `RunInTransaction`/`BeginTx`, no `Tx = h.db` beyond the allowlisted BeforeHandle placeholders, no pool statements in spec handlers).
- NOTE: resolvespec never fires `AfterRead`/`AfterCreate` (hook types exist, no call site); not tested, pre-existing.
- NOTE: restheadspec total-count cache is process-wide and ignores the record id; read tests call `resetTotalCache`.
- FIXED: resolvespec now fires `AfterRead` (in the read tx; `Result` = scanned slice for single and list) `AfterCreate` (in the create tx, per record, all four create paths) and `AfterDelete` (in the delete tx, once per request; a failure rolls the delete back). Before, neither fired, so `AfterRead` column-level security masking (registered by `RegisterSecurityHooks`) was silently skipped on resolvespec reads. Failing `AfterRead`/`AfterCreate` fails the request and rolls back. Tests: `pkg/resolvespec/ops_tx_test.go` (incl. end-to-end column hiding).
- FIXED: restheadspec total-count cache key now includes the record id; a read by id (total 1) no longer poisons the list total for the 2-minute TTL. Tests: `pkg/restheadspec/cache_key_test.go`. resolvespec is unaffected (its count runs before the id filter, so the total is the list total by design). The cache is still process-wide, so read tests call `resetTotalCache`.
- NOTE: `pkg/security` `TestDatabaseAuthenticator` fails with `-count=2` (also on `cd96404`, before this work); use `-count=1`.
## Tests
@@ -118,3 +118,4 @@
- `dbtrace` shows `pooled=0` for every handler op on a hooked model.
- RLS GUC set in `OnTxBegin` is visible to read, create, update, delete queries and hooks.
- No `Tx: h.db` / `hookCtx.Tx = h.db` left in spec handlers.
- OPEN: websocketspec `BeforeDisconnect`/`AfterDisconnect` are defined but never executed (connection lifecycle, not DB). Allowlisted in `TestEveryDefinedHookHasACallSite`; wire them to remove the entry.