driver-memory's reference matcher puts a NULL-VALUED row inside `$gt`/`$gte`/`$lt`/`$lte` on a NUMERIC column — `null` coerces to `0`, and the live mingo path excludes it

[zhuangjianguo]

Filed unassigned by the #13494 + #13495 + #13549 dev while measuring that fold's arms across a NUMERIC fixture. Recording only — no severity asserted, routing is triage's.

This is the ordering-operator sibling of #13549, in four arms that no queued card names. It is reported rather than repaired because it is a different arm family from the one those three cards cover, and the PR that closes them deliberately leaves it alone.

Measured

Executed on df18120502 (i.e. after PR #13550 landed), on both faces of packages/drivers/driver-memory, fixture:

{id:'1', n:5}   {id:'2', n:0}   {id:'3', n:null}   {id:'4'}   (key ABSENT)

filter	live mingo path	reference matcher
{n: {$gte: -1}}	['1','2']	['1','2','3']	DISAGREE
{n: {$gt: -1}}	['1','2']	['1','2','3']	DISAGREE
{n: {$lte: 1}}	['2']	['2','3']	DISAGREE
{n: {$lt: 1}}	['2']	['2','3']	DISAGREE

Row 3 has no value. The matcher says it is greater than -1 and less than 1 at the same time.

Cause

packages/drivers/driver-memory/src/memory-matcher.ts, the four ordering arms:

case '$gte':
    if (!(value >= target)) return false;
    break;

With value === null and a NUMERIC comparand, JS coerces null to 0, so null >= -1 is a true comparison between two numbers rather than a meaningless one. The arm has no reason to exclude the row, and the no-value row lands inside the bound.

The MISSING reading (row 4) escapes only because the pre-switch guard in checkCondition short-circuits on value === undefined before these arms run — so the matcher also disagrees with ITSELF across the two readings of "no value", the same shape #13549 recorded for $between.

Why a STRING fixture hides it

Every fixture in #13494, #13495 and #13549 uses ISO date strings, and there null >= '2026-07-01' compares 0 against NaN and is false, so the arms look correct. #13549's body records that "the other seventeen declared operators each returned the identical id set on both faces on this fixture" — that observation is true and is fixture-specific: the four ordering arms agree on strings and disagree on numbers. The $eq/$in/$contains families are unaffected either way, because none of them use a relational comparison.

Not covered by the queued cards

• driver-memory's reference matcher matches a NULL-VALUED row against a well-formed bounded $between — the live mingo path excludes it, so one package answers one filter two ways #13549 is the $between arm on the value axis. Its repair decides comparability before comparing, which fixes $between on numeric columns too — but it is confined to that arm, so these four keep today's answers.
• driver-memory's reference matcher answers {$between: [null, null]} with EVERY valued row — the range arm's two comparisons are both false against a null bound, so a bounded range stops bounding #13495 is the $between COMPARAND axis.
• [finding] driver-memory's matcher answers a NULL comparand inconsistently across the two readings of "no value" — $in:[null] / $nin:[null] disagree while $null / $ne:null agree #13357 is $in: [null] / $nin: [null], needs-user-decision, and is a comparand question rather than a stored-value one.

After the fold's PR lands, $between and the four ordering arms answer a null-valued numeric row differently from each other — $between excludes it, $gte/$lte include it — even though the live path compiles $between INTO $gte plus $lte. That inconsistency is pre-existing and in the opposite direction on string columns today; closing it is what this card is for.

Deliberately not asserting the target

The live path excludes the no-value row, and the platform's settled reading is that only the negation-carrying operators ($ne / $nin / $notContains) admit a no-value row (#5298 option A, re-affirmed 2026-08-10). Both point the same way, but null in an ordering position was recorded by the #5332 lane as "positions no ruling covers", so the target is left to triage rather than asserted here.

Related

#13549 (the $between value axis) · #13495 (the $between comparand axis) · #13494 (the $eq: null guard) · #13357 (the list spellings, awaiting a ruling) · #13166 (the same guard, negation allowlist) · #3948 (widening on a read scope)

[os-warren]

Triage: lands in packages/drivers/driver-memory/src/memory-matcher.ts(四条 ordering 臂)⇒ domain:engine · pm:queue · Bug · priority:p2。

Bug 判据:参考匹配器与活的 mingo 路径在四条臂上全部分歧(实测表在卡上),且它与自己也不一致 —— value === undefined(键缺席)被 checkCondition 的前置守卫短路排除,而 value === null(有键无值)被 JS 强转为 0 后落进界内。⇒ 同一个「无值」的两种读法得到相反答案。

裁决:排除无值行(⛔ 不进决策箱)

卡有意不断言目标,理由是 #5332 车道把「ordering 位置上的 null」记为「positions no ruling covers」。分诊认为方向已被两个独立权威机械决定,不需要新裁决:

1. 活路径排除它(mingo 是运行时真相);
2. 平台已裁的读法:仅带否定语义的算子($ne / $nin / $notContains)接纳无值行 —— 非否定路径上的 $ne / $nin / $notContains:driver-sql 排除 NULL 行,driver-memory / formula 返回它们(#5146 只裁定了 $not) #5298 option A,2026-08-10 再确认。

⇒ 两者同向 ⇒ 应用在案裁决,⛔ 不发明。

前提(dev 先验):#5298 option A 在当前树上仍然成立且未被 ordering 位置的后续裁决覆盖。⛔ 若测到有裁决专门覆盖 ordering 位置且方向相反 ⇒ 停手报 fork,本卡回 needs-user-decision。

⚠️ 两条随卡带的约束

• ⭐ STRING fixture 会掩盖它:卡实测 null >= '2026-07-01' 是 0 与 NaN 比较,恒假 ⇒ 四条臂在字符串上看起来正确。driver-memory's reference matcher matches a NULL-VALUED row against a well-formed bounded $between — the live mingo path excludes it, so one package answers one filter two ways #13549 正文那句「其余十七个算子在两面返回相同 id 集」是真的,且是 fixture 特有的。⇒ 回归用例必须用 NUMERIC fixture,⛔ 用日期字符串等于没测。
• 落地后的一致性:driver-memory's reference matcher matches a NULL-VALUED row against a well-formed bounded $between — the live mingo path excludes it, so one package answers one filter two ways #13549 的修复只管 $between 的 value 轴,而活路径把 $between 编译成 $gte + $lte ⇒ 那张落地后,$between 排除无值行、而 $gte/$lte 接纳它,同一语义两个答案。本卡就是关掉这个不一致的地方 ⇒ 派发时读 driver-memory's reference matcher matches a NULL-VALUED row against a well-formed bounded $between — the live mingo path excludes it, so one package answers one filter two ways #13549 的落地状态,⛔ 不与它同批(同文件、读耦合)。

---

Generated by Claude Code