fix(line-items): F/P expression tags, null guard, shared price-fill (#514 items 8, 3, 15)

[yinlianghui]

Description

Partial follow-up to #514, scoped deliberately to the two *_line_item objects (plus the one-line campaign_member tag fix that belongs to the same item). Each of the four defects was re-verified as still present on main at 5b88ad0 before being touched — the issue body predates several merged PRs.

This PR does not close #514; the remaining tier-3/tier-4 items are untouched.

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Code refactoring
• Performance improvement
• CI/CD update

Related Issues

Refs #514 — items 8, 3 (line-item half) and 15 (price-fill half) only. Not a fix-all; the issue stays open.

Changes Made

• Item 8 — formula-on-formula. crm_quote_line_item.total_price was record.subtotal * (1 + tax_rate/100), and subtotal is itself a formula, so the total depended on the platform hydrating another computed field first (the hazard written up at lead.object.ts:61-64). It is now composed from the same stored fields subtotal reads, with the tax multiplier on top. The arithmetic is unchanged — 4 × 100 at 10% line discount and 8% tax is still 388.80 — what changes is that it no longer depends on evaluation order.
• Item 3 (line-item half) — unguarded predicate. crm_opportunity_line_item's unit_price_positive compared record.unit_price < 0 with no null guard. Strict CEL aborts on null < 0 rather than evaluating it false, so the rule was inert on a blank price — it never fired at all. It now carries the same != null && guard its quote-side twin has always had. The guard only narrows the predicate, so no previously-accepted record starts failing and no seed data changes; a blank price is still caught by the field's own required.
• Item 8 — wrong expression tags. Formula fields on both line-item objects were authored with P (the predicate alias), and campaign_member's validation used a raw { dialect: 'cel', source } object because the file never imported P. F, P and cel are all aliases of the same tagged template, so this was invisible at runtime and only ever misled readers. Formulas now use F, conditions use P, and campaign_member.object.ts imports P.
• Item 15 (price-fill half) — near-verbatim twins. opportunity_line_item.hook.ts and quote_line_item.hook.ts carried copies of the same price-fill handler differing only in comments — the shape that lets a fix land on one object and silently skip the other. Both now build from a new src/objects/_line-item-price-fill.ts. The sharing is authoring-time only: the handler body closes over nothing but its own ctx, so it still lowers to a body-only sandbox callable, verified against the build artifact (the two lowered body.source strings in dist/objectstack.json are byte-identical and mention neither factory parameter). The rollup hooks next door look alike but compute genuinely different totals, so they are left alone.

Testing

• Unit tests pass (pnpm test — 22 files, 403 passed / 2 skipped)
• Linting passes (pnpm lint)
• Build succeeds (pnpm build)
• pnpm verify end to end, exit code 0
• New tests added

New guards live in test/line-item-conventions.test.ts — a new file on purpose, rather than more surface on test/metadata-references.test.ts, which is the repo's highest-churn test module and would conflict with parallel #514 work.

Written test-first: all six guards fail on main before the source changes, one per defect.

Guard	Asserted against
Every Field.formula expression uses F	source text of every *.object.ts
Every validation condition uses P (catches the raw {dialect:'cel'} literal too)	source text of every *.object.ts
No line-item formula reads another formula field	compiled metadata
Inlined quote total still yields 388.80 on the worked example	compiled metadata
Both unit_price_positive rules are null-guarded	compiled metadata
The two price-fill handlers are one implementation	handler source + a shared behavioural scenario table run against each object

The tag guards have to read source text because F/P/cel produce byte-identical Expression envelopes — the distinction exists only in the authored source. The other guards assert against compiled metadata. Both tag guards run repo-wide over src/objects/*.object.ts and are clean after this change; the null-guard check is scoped to the two line-item objects because product.object.ts has the same defect and is tracked by the rest of #514 item 3.

The pre-existing behavioural price-fill tests in hooks-runtime.test.ts and hooks-runtime-sales.test.ts still pass unchanged, which is the independent evidence that the de-duplication preserved behaviour on both objects.

Screenshots

n/a — no UI surface changes.

Checklist

• I have added a changeset — .changeset/line-item-expressions-and-price-fill.md
• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation — n/a, no documented behaviour changes
• My changes generate no new warnings (the two campaign_enrollment expression warnings are pre-existing on main)
• I have added tests that prove my fix is effective
• New and existing unit tests pass locally with my changes
• Any dependent changes have been merged and published

Additional Notes

The null guard on unit_price_positive is the one item here with a real behavioural footprint, and it is strictly narrowing: the rule fired on nothing before and now fires on a genuinely negative price. Worth a second look during review is the sandbox reasoning in _line-item-price-fill.ts — the factory is only safe as long as the handler body never reads a factory parameter, since that would resolve at authoring time and be undefined inside QuickJS. That constraint is stated in the module's doc comment.

---

Generated by Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
hotcrm	Ignored		Jul 31, 2026 8:38am