Repository navigation
lexer: numeric suffixes are not keyed by literal kind or applied on every ReadNumber path — 3.14L runs as float64 with the suffix dropped, .5f is SPY0103 while 0.5f runs #2339
Description
Class. The decimal ReadNumber path does not apply the numeric-suffix rule uniformly across literal kinds and lexing paths. The violated contract: a suffix is accepted exactly when the spec gives it to that literal's kind, and on every path that lexes that kind. The spec gives integer suffixes (L/u/ul/…) to integers in integer_literals.md, and f/d/m to floats in float_literals.md. float_literals.md also says .5 is the "Same as 0.5". Every other suffix is refused with SPY0023.
Found by the P68 numeric twin (NumericLiteralTwinTests, plan-d31acd, #2300) at 220259b. The lead confirmed it with the built sharpyc.
| # | Cell | Program | Observed | Expected | Lane |
|---|---|---|---|---|---|
| 1 | integer suffix on a decimal float | x = 3.14L (also 3.14u, 1e5L) |
runs. The literal types float64 (a: int64 = 3.14L → SPY0220 "Cannot assign type 'float64'"), and the suffix is silently dropped |
SPY0023 (C# refuses 3.14L too) |
missing-refusal |
| 2 | float suffix on a leading-dot float | x = .5f (also .5m, .5d, .5e3f) |
SPY0103 "Expected end of statement, got Identifier". The leading-dot branch of ReadNumber (Lexer.Literals.cs ~:723–788) returns before the suffix arm, so f lexes as an Identifier. 0.5f runs |
accepted as float32, the same as 0.5f |
false-refusal |
| 3 | sibling of 2 | .5j |
SPY0103 | SPY0023, the same as 1j |
message |
The sites are the suffix roster at Lexer.Literals.cs ~:923, which is not keyed by literal kind, and the leading-dot early return. The hex/bin/oct path already keys its suffixes by kind through IntegerSuffixes, from the #2208 fix.
Matrix to sweep at fix time. Cross the literal kind (decimal int, decimal float, leading-dot float, exponent-only float 1e5, hex/bin/oct) with each suffix in the roster, under both of its cases. The numeric twin's SharpyOnlyRows_LexAsTheSpecSays table is the place to pin the matrix.
refs #2300
Activity
docs(gap-discovery): roster rows for the P68 lexer harnesses (refs #2300
antonsynd commentedon Oct 22, 2026
Batching 2026-10-21 @ e7e0a0e — charter
.claude/plans/remediation-round-2026-09-20-batching.md§3.0, chores band (order 0, one commit, no plan; rides P21e's gate; lands BEFORE P22c, which rewrites the same literal reader). Ruling R-HX (owner, 2026-10-21): (a) by class — #2339 → a chore in the current band (LexerReadNumberonly); #2337 → P22c; #2338 → P67.Contract (
integer_literals.md/float_literals.md): a numeric suffix is keyed by the literal's KIND (integer vs float) and consulted on EVERYReadNumberpath — every suffix the spec lists is accepted on every spelling of its kind and refused on every other kind. Measured @ e7e0a0e (built CLI, lock-free):print(3.14L, 1e5u)RUNS printing3.14 100000.0— the integer suffix is silently dropped and the value isfloat64;y = .5f→ SPY0103 "Expected end of statement, got Identifier" while0.5fruns;.5j→ SPY0103 where the spec says SPY0023. Lane / reach: false-refusal + missing-refusal · program (the first cell is silent-wrong in effect). Class: the parallel-path meta-class (contract §1) — the hex/bin/oct path has been keyed by kind since #2208; the decimal and leading-dot paths are not (grep-only; the chore measures). Cure: one suffix table keyed by kind, consulted by every path after the digits; the twin's hand tableNumericLiteralTwinTests.SharpyOnlyRows_LexAsTheSpecSayspins kind {int, float} × spelling {0.5,.5,5.,1e5, hex/bin/oct, underscored} × suffix {every integer suffix, every float suffix,j, none} × case; direction record against the prior commit (3.14Lran → refused;.5fSPY0103 → runs;.5j→ SPY0023); mutation (unkey one path → its rows red) in the commit body; fixture-hit count for both spellings FIRST (zero expected). Blast radius: Compiler.Tests lexer +FrontEndParity; no snapshots. Class tracker: this issue.Also recorded here for the chore's convenience (not a requirement): R-HW's lone-surrogate message (
𝒳 = 1→ SPY0015 "Unexpected character: '�'") may print the full code point — one line in the same reader; the surface itself is adocs/deviations.yamlrow (identifier-letters-bmp-only). Rejected (R-HX): (b) a P73 residue group with #2337/#2338; (c) P22c (numeric, not the prefix table). Two-way.fix(lexer): a numeric suffix is keyed by the literal's kind and read …
antonsynd commentedon Oct 22, 2026
Landed (chores band 2026-10-22) on
dev@ 4d22212, PR #2346 to mainline. Whole-solution gate 81669/0/7 @ 4d22212 (measured), against 81093/0/7 @ be4664e.Commit 0b30fa9. One suffix table,
Lexer.NumericSuffixes, keyed by kind {Integer, Float}. One reader,ReadNumericSuffix, is called by all fiveReadNumberpaths (decimal, leading-dot, hex, binary, octal) after their digits.Direction (
run, e7e0a0e → this commit):print(3.14L, 1e5u)3.14 100000.0print(1.5ul)y = .5f0.5print(.5j)Matrix:
NumericLiteralTwinTests.SharpyOnlyRows_LexAsTheSpecSaysnow has 345 generated rows, up from 83.Mutations:
Every acceptance cell is met. Closing.
Merge pull request #2346 from antonsynd/dev