Filed by the domain:engine execution seat while implementing batch i of #16159 (PR #16308). ⛔ Observation, not a runtime defect — no assignee, no pm:queue; triage grades it.
Blocked-by: #16308 — the constants this is about do not exist on main until that PR lands.
The census
#16159 converts engine error codes from inline literals into importable *_CODE constants. The card's argument is about EXTERNAL consumers, but the same cost is already being paid inside this repo: three first-party call sites match these refusals by code and therefore carry a second spelling of the wire string, kept equal to the engine's by nothing but a grep.
| site |
code re-spelled |
what it does |
packages/rest/src/error-response.ts:1024 and :1029 |
ERR_DATASOURCE_UNAVAILABLE |
matches the refusal, then re-authors the same spelling into the response envelope it builds — two spellings in one file |
packages/rest/src/import-runner.ts:537 |
ERR_SUMMARY_RECOMPUTE |
the documented "the records WERE written, so treat a failed roll-up as a warning" recovery |
packages/metadata-protocol/src/seed-loader.ts:2081 |
ERR_SUMMARY_RECOMPUTE |
the same recovery, independently spelled |
Measured on main at 2756e07d1 plus PR #16308's branch. Both codes are now exported: DATASOURCE_UNAVAILABLE_CODE and SUMMARY_RECOMPUTE_CODE from @objectstack/objectql.
⚠️ The three sites are NOT one job — the dependency direction splits them
packages/rest already declares "@objectstack/objectql": "workspace:*", so its two sites can import the constants today. That half is mechanical.packages/metadata-protocol does not depend on @objectstack/objectql at all. Converting its site would add a package dependency from the protocol layer onto the query engine, which may well be the wrong direction on purpose. ⛔ Do not treat it as the same edit. Either it stays a deliberate wire-string match (in which case a comment saying so is the deliverable, not an import), or the layering question is answered first.
Why it is worth a card rather than a shrug
check:error-code-provenance counts each of these as a stamp site in the consumer's own package — they are listed and legal today, and they stay legal after conversion; nothing is red either way. What changes is drift exposure: a spelling that lives in three packages can diverge from what the engine throws with no compile error to say so, and that is the exact failure #16159 exists to remove. Converting the producers without ever converting the first-party consumers leaves the repo demonstrating the pattern it tells customers not to use.
⛔ Deliberately excluded from PR #16308: that PR is a producer-side sweep, this is a consumer-side change in two other packages, and no gate asks for it. Keeping them apart is what keeps each PR reviewable as one edit.
Refs: #16159 · #16308 · #16260 (the related, distinct question of which BARREL publishes these constants)
Generated by Claude Code
Filed by the
domain:engineexecution seat while implementing batch i of #16159 (PR #16308). ⛔ Observation, not a runtime defect — no assignee, nopm:queue; triage grades it.Blocked-by: #16308— the constants this is about do not exist onmainuntil that PR lands.The census
#16159 converts engine error codes from inline literals into importable
*_CODEconstants. The card's argument is about EXTERNAL consumers, but the same cost is already being paid inside this repo: three first-party call sites match these refusals bycodeand therefore carry a second spelling of the wire string, kept equal to the engine's by nothing but a grep.packages/rest/src/error-response.ts:1024and:1029ERR_DATASOURCE_UNAVAILABLEpackages/rest/src/import-runner.ts:537ERR_SUMMARY_RECOMPUTEpackages/metadata-protocol/src/seed-loader.ts:2081ERR_SUMMARY_RECOMPUTEMeasured on
mainat2756e07d1plus PR #16308's branch. Both codes are now exported:DATASOURCE_UNAVAILABLE_CODEandSUMMARY_RECOMPUTE_CODEfrom@objectstack/objectql.packages/restalready declares"@objectstack/objectql": "workspace:*", so its two sites can import the constants today. That half is mechanical.packages/metadata-protocoldoes not depend on@objectstack/objectqlat all. Converting its site would add a package dependency from the protocol layer onto the query engine, which may well be the wrong direction on purpose. ⛔ Do not treat it as the same edit. Either it stays a deliberate wire-string match (in which case a comment saying so is the deliverable, not an import), or the layering question is answered first.Why it is worth a card rather than a shrug
check:error-code-provenancecounts each of these as a stamp site in the consumer's own package — they are listed and legal today, and they stay legal after conversion; nothing is red either way. What changes is drift exposure: a spelling that lives in three packages can diverge from what the engine throws with no compile error to say so, and that is the exact failure #16159 exists to remove. Converting the producers without ever converting the first-party consumers leaves the repo demonstrating the pattern it tells customers not to use.⛔ Deliberately excluded from PR #16308: that PR is a producer-side sweep, this is a consumer-side change in two other packages, and no gate asks for it. Keeping them apart is what keeps each PR reviewable as one edit.
Refs: #16159 · #16308 · #16260 (the related, distinct question of which BARREL publishes these constants)
Generated by Claude Code