-- Cross-engine error parity: at-OOB now reports the same rich message,
-- code (ILO-R009), and call-stack notes on tree, VM, and Cranelift.
--
-- Pre-fix (v0.11.7 db-analyst rerun8):
-- tree: ILO-R009 "at: index 99 out of range for list of length 3"
-- notes: ["called from 'main'", "called from 'g'"]
-- VM: ILO-R004 "at: index out of range" (no values)
-- notes: ["called from 'main'", "called from 'g'"]
-- Cranelift: ILO-R004 "at: index out of range" (no values)
-- notes: [] <-- empty
--
-- Post-fix: all three engines produce the tree shape bit-equal. An agent
-- that filters by code (e.g. retry-on-R009-only) now sees consistent
-- behaviour regardless of which engine the runner picked.
--
-- This file only exercises the *positive* paths so the examples harness
-- can pin against `-- out:` directly. The cross-engine error-shape
-- contracts (rich message, R009, call-stack notes) are asserted in
-- tests/regression_cross_engine_error_parity.rs which spawns the binary
-- and inspects --json stderr on every engine.
-- Safe `at`: positive index inside bounds works the same way on every engine.
safe-at>n;at [10,20,30] 1
-- Safe `lst`: replacing an in-bounds element works the same way too.
safe-lst>L n;lst [10,20,30] 1 99
-- Nested call: when this errors, all three engines now report
-- ["called from 'main'", "called from 'g'"] in notes. We don't assert
-- the error path here (harness limitation), but the test in
-- regression_cross_engine_error_parity.rs covers it.
g xs:L n>n;at xs 0
deepest>n;g [42]
-- run: safe-at
-- out: 20
-- run: safe-lst
-- out: [10, 99, 30]
-- run: deepest
-- out: 42