test: literal corpus × entry point against PostgreSQL (route B, step 3)

[CaffeinatedCoder]

Closes #123

Problem

test/oracles/axes.md listed literal corpus × entry point against PostgreSQL on the same text as a pair no test varies. LiteralGrammarParityTests asked the server about quoted, escaped and whitespace bounds, through TryParse only. Everything else PostgreSQL's range_in, multirange_in, array_in and the element input functions tell apart had never been put to the server, and neither had the other entry points. #114 and #115 were single cells of this product. Each was found once and never kept as a test.

Change

• LiteralCorpusProbeTests (integration project, route B step 3 of Consumer probing: port TypeKit's prober tooling (route A), or BLite's lightweight way with hand-written axes and MSTest probe sweeps (route B) #105). It has one [DynamicData] row per type: 10 range types, their 10 multiranges and 28 value set types. YearMonth is left out because its literal isn't the text of the daterange it's stored as.
  • Corpus. Generated per element kind, 65–135 literals per type, 4,715 in all (30,699 literal × entry point cells). The structural families are bounds, equal and inverted bounds for each bracket pair, empty in every case, blank input, quoting, escapes, stray punctuation, trailing junk, ASCII and Unicode whitespace, array dimensions and NULLs, and multirange canonicalization. Per element kind it adds sign, leading zeros, exponent, NaN/Infinity, non-decimal integers and digit separators, more than 28 digits, overflow, the extremes, infinity spellings, epoch/now/today/allballs, non-ISO dates, 24:00:00, the leap second, offsets and zone names, sub-microsecond digits, uuid spellings and NUL.
  • Authority. pg_input_is_valid($1, type) and the canonical text, with every literal of a type sent as a text[] in one statement (a fallback runs one statement per literal). An array is read without its dimension decoration.
  • Entry points. Parse(string), Parse(span), TryParse(string), TryParse(span), JSON (a range or multirange as its literal; a value set as [E] where JSON can spell the element), the TypeConverter and the EF Core JSON document reader. All of them must agree. A value the library reads is also written back through EF Core as a SQL constant and as a JSON document, and the server compares it with the original as typed values. That comparison is what catches a precision the parser dropped.
  • Counted, not reported. A refusal of a value outside the CLR domain, a refusal docs/serialization.md names (J07 strict ISO 8601, J06 the numeric grammar), and an extreme the store spells ±infinity (docs/efcore.md).
  • Without Docker the row is Inconclusive; under CI=true it fails. The whole class takes about 6 s.
• Registered. Added to oracles.json, with the seed seeds/empty-keyword-matched-as-prefix.patch. New claim J34 quotes "The text is the PostgreSQL literal, the same text the server itself reads and prints". The class is added to the heldBy of J06, J11, J12, J33 and S12. In axes.md the pair moves into "Varied by", and the two pairs it leaves open are listed.
• Candidates. 115 rows, keyed {type} {family} , each with an issue. No defect is fixed here.
• Docs. A paragraph each in docs/oracles.md (route B) and docs/testing.md.

Findings

Issue	Family	Types	Silent
#114 (known)	inverted bounds parse as empty	20 ranges and multiranges	yes
#115 (known)	more than 28 digits rounded, seen on the EF write-back	DecimalRange, its multirange, DecimalSet	yes
#78 (known)	whitespace in a uuid element accepted	GuidSet, GuidSet<T>
#117 (known, column path)	[x,infinity) loses the last day, and (-infinity,x] the first, on the literal and JSON-document paths too (scope added to the issue)	DateRange, LocalDateRange, their multiranges	yes
#126	sub-microsecond digits: kept to 100 ns, stored truncated; PostgreSQL rounds half-to-even	16 temporal types	yes
#127	0x/0o/0b and 1_000 refused for integers	10
#128	[1:1]={…}, [0:1]={…}, {{}} refused	28 value sets
#129	GuidSet follows Guid.Parse: accepts (…), refuses hyphen groups	2
#130	LocalDateTime refuses Z and an offset that timestamp input drops	4
#131	DateTimeOffset refuses offsets beyond ±14:00	4
#132	(,max] raises 22003 in PostgreSQL and isn't in the boundary table (documentation)	Int32Range, Int64Range
#133	a NUL character accepted into a text element	2

I confirmed each one by hand against postgres:17-alpine with psql (repros are in the issues). No 24:00:00 cell disagrees, so nothing is registered against #108: the library refuses 24:00:00, and the server's value can't be a TimeOnly.

Tests

• dotnet test test/CodoMetis.ValueRanges.EFCore.PostgreSQL.IntegrationTests passes with Docker: 200 tests in 11 s, LiteralCorpusProbeTests included (49 tests, about 6 s).
• dotnet test test/CodoMetis.ValueRanges.Conventions.Tests passes (217 tests); Oracle*Tests hold the registry, claims, seed and candidates.
• dotnet run tools/oracles.cs seeds gives every seed is caught: 48 (seed, oracle) pairs over both route B steps. LiteralCorpusProbeTests catches empty-keyword-matched-as-prefix (10 of 49 rows; LiteralGrammarParityTests and RangeLiteralGrammarContractTests stay green under it), quoted-bound-whitespace-kept (14) and temporal-set-element-whitespace-kept (16).
• dotnet run tools/oracles.cs status reports 0 oracles without a seed and 0 candidates without an issue.
• The EF write-back leg can't go quiet unnoticed. The Decimal text beyond 28 significant digits is rounded silently when a literal or JSON array is read #115 and A temporal literal with sub-microsecond digits is a different value than PostgreSQL reads from the same text #126 rows for DecimalRange and TimeRange come only from that leg, so a leg that stopped comparing would fail as candidate fixed.

Bugs this sweep would have caught (retroactive check over every bug issue)

I added found-by/oracle where it was missing and posted a comment naming the row and how it was verified:

• Quoted range bound with a leading space is refused; PostgreSQL accepts it #45 (closed): row DateRange whitespace [" 2026-01-01",2026-01-02). I verified it by applying seeds/quoted-bound-whitespace-kept.patch (the revert of fix: strip whitespace at either end of a quoted or escaped range bound #73): 14 of 49 rows fail.
• Quoted value set element with surrounding whitespace is refused for temporal sets; PostgreSQL accepts it #70 (closed): row DateSet whitespace {" 2026-01-01"}. I verified it by applying seeds/temporal-set-element-whitespace-kept.patch (the revert of fix: strip whitespace at either end of a quoted or escaped temporal set element #81): 16 of 49 rows fail.
• Both seeds now name LiteralCorpusProbeTests.
• An inverted range literal parses as empty on every entry point; PostgreSQL rejects it #114, Decimal text beyond 28 significant digits is rounded silently when a literal or JSON array is read #115, GuidSet accepts whitespace inside an element that PostgreSQL's uuid_in rejects #78, A discrete range read as [x,infinity) loses 9999-12-31; UpperBound disagrees between memory and server #117 (open): the candidates.json rows are the proof. For A discrete range read as [x,infinity) loses 9999-12-31; UpperBound disagrees between memory and server #117 the catch is on the literal and JSON-document paths only; the column path isn't a cell of this sweep.
• A temporal literal with sub-microsecond digits is a different value than PostgreSQL reads from the same text #126–StringSet accepts a NUL character, which PostgreSQL text cannot hold #133: found by this sweep.
• Not caught: Minimal API binds a variant-typed parameter as null for a valid literal and answers 200 #116, IsAdjacentTo at the date extremes: memory says adjacent, PostgreSQL does not (MinValue/MaxValue are ±infinity there) #110, IsUnboundedStart/IsUnboundedEnd on the infinite range answer false in memory and true through EF Core (lower_inf) #109, RangeSet.Overlaps answers false on a continuous domain when the operand ends exactly where a later element begins #106, Any/Count over a captured list of ranges silently merge overlapping ranges and drop empty ones #76, list.Contains(rangeColumn) fails with 42809 'op ANY/ALL (array) requires array on right side' #71, Null value set in ExecuteUpdate on a JSON column is refused with the to_jsonb message #57, Wrapper set arities don't check that an element's text form round-trips #53, list.Contains(valueSetColumn) throws InvalidCastException in Npgsql #46, SECURITY.md names the version being prepared as supported, not the latest release #79. They are host binding, algebra, translation, wrapper element and release defects, and none of them is a literal read differently from the server. Document that a stored timerange bound of 24:00:00 fails to load #108 (documentation) agrees cell by cell.

Rebased onto main after route B step 2 (#125) merged; both steps' probe classes and seeds are kept.

🤖 Generated with Claude Code