[sherenkov]

An application read its whole list of named arguments inside one try, so a failure anywhere in the list backtracked into the positional form, which then failed at the first ↦. A ) forgotten on line 6 of the program in #1941 was blamed on line 3, at the ↦ of text, which is valid code.

exTail now only looks ahead for the start of a named argument, an attribute or an α index followed by ↦, and once it sees one it parses the rest of the list without backtracking. The program of #1941 is now reported at 7:10, the ↦ of mode, the first token after the missing ) that cannot be read. ⟦ a ↦ Φ.foo( x ↦ 1, y ↦ ) ⟧ is reported at the empty value of y, and ⟦ a ↦ Φ.foo( x ↦ 1, 2 ) ⟧ at the 2, both instead of at the ↦ after x. Since #2259 an argument takes no inline voids, so the lookahead takes none either.

A ParserSpec block checks the three positions and fails without the change. A case in "parse applications with mixed bindings" checks that Q.f(b(c), d), a nested call, stays positional.

Fixes #1941

[github-actions]

Benchmark comparison

Median of 5 round(s) per side, interleaved on the same runner against the PR merge-base, with the bench fixture frozen across both builds. Noise is the spread between the fastest and the slowest round, taken from the noisier side.

Bench	base (μs)	head (μs)	delta	noise
morph/symbolic/accum/0	113.52	110.66	-2.5%	±3.0%
morph/symbolic/accum/400	133.57	131.41	-1.6%	±2.5%
morph/symbolic/demo/e1	132.29	131.36	-0.7%	±4.1%
morph/symbolic/demo/e2	134.16	131.59	-1.9%	±2.7%
morph/symbolic/demo/e3	134.48	132.30	-1.6%	±2.2%
morph/symbolic/demo/e4	135.09	131.03	-3.0%	±3.6%
morph/symbolic/demo/e5	137.00	133.04	-2.9%	±1.5%
morph/symbolic/native/e5	1190.42	1156.20	-2.9%	±43.1%
parse/phi	235516.49	234518.38	-0.4%	±1.7%
parse/xmir	948534.77	936682.03	-1.2%	±1.4%
print/salty/multiline	1782052.91	1771902.90	-0.6%	±2.9%
print/sweet/flat	68942.15	66033.29	-4.2%	±5.6%
print/sweet/multiline	535855.58	537550.94	+0.3%	±3.5%
rewrite/normalize	2.15	2.09	-3.0%	±5.3%

Scaling

Each sized case against its /0 sibling, the same work over nothing extra: a ratio near 1 means the extra size is free.

Bench	against	base	head
morph/symbolic/accum/400	morph/symbolic/accum/0	1.18×	1.19×

No case moved by more than 20% beyond its own noise.
This report is informational, a delta never fails the build.

[sherenkov]

@yegor256 CI is green and the branch is up to date with master, could you take a look when you have a moment? As you suggested in #1941, try now covers only the lookahead for the start of a named argument, so the program of the report is blamed at line 7, right after the missing ), instead of at the first ↦ on line 3, and ParserSpec checks that position.

[sherenkov]

@yegor256 CI is green and the branch is up to date with master, could you take a look when you have a moment? Since #2259 refused inline voids in an argument, the lookahead for a named argument takes none either, so it reads exactly the start argument reads, and a typo inside the list of #1941 is still reported where it is.

[yegor256]

The parser commits to named arguments only after lookahead sees an attribute or index followed by an arrow. Tests pin the improved error locations and retain nested positional calls.