PromQL parser-fork gaps (umbrella) — 72% of corpus rejections, all in promql-parser

[zzylol]

Umbrella — PromQL parser-fork gaps (tracking)

233 of the ~322 PromQL corpus rejections (72%) are L1 parse failures in the private promql-parser fork (ProjectASAP/promql-parser, branch asap), before the lowerer ever runs. This repo consumes the parser via a git pin; the fixes land in the fork, and the only change here is bumping that pin once each lands.

Correction (2026-07-10): none of these are "mirror upstream"

An earlier read of this issue assumed some gaps (smoothed/anchored/start/end) were just a rebase onto GreptimeTeam upstream away. That was checked against the actual grammar and it does not hold — GreptimeTeam upstream lags Prometheus (Go) on every one of these:

• smoothed / anchored are stub tokens even upstream — src/parser/promql.y:124 marks them %expect-unused (declared, no production consumes them) on both the fork and GreptimeTeam main.
• trim operators, duration expressions, and callable start()/end() have no token / AST anywhere upstream.

The fork's asap commit (c51beaf) added the experimental functions (info, histogram_quantiles, min_of/max_of/step/range, the *_over_time reducers) but explicitly deferred the grammar/lexer features. So all five gaps are net-new parser work on the asap branch; there is nothing to mirror.

Split (all in ProjectASAP/promql-parser)

gap	corpus	kind	issue
smoothed / anchored	61	grammar production for existing stub tokens (lexer done)	ProjectASAP/promql-parser#1
trim </ >/	110	from-scratch operator port (lexer + grammar + AST)	ProjectASAP/promql-parser#2
duration-arithmetic exprs	51	DurationExpr AST + lexer + grammar	ProjectASAP/promql-parser#3
start() / end() callable	6	reserved-keyword lexer/grammar work	ProjectASAP/promql-parser#4
info() 2nd-arg matcher	5	validation bug fix	ProjectASAP/promql-parser#5

Ordered easiest-first: #5 (bug fix) → #1 (token exists) → #4 → #3 → #2 (biggest).

The remaining ~89 non-parser rejections are lowering gaps (tracked elsewhere) and deliberate rejections (out-of-range φ, __name__ regex per #67).

[zzylol]

Related open work

• Parser-fork failures contribute to corpus tracker Add corpus tests from DQC, BGP analytics, ClickBench, and business analytics workloads #132.
• PromQL: non-literal scalar parameters (topk(scalar(foo), v), quantile(scalar(f), v), limit_ratio(time()%17/17, v)) #120 is a downstream PromQL lowering/IR issue, not a parser gap, and should remain separate.