codegen: a dependent varip aggregate initializer sees its dependency's value at declaration time - #263

What

A varip aggregate (UDT, array, map, matrix) whose initializer reads a value computed by an earlier statement now initializes at its declaration, in statement order, like the var form fixed in #226 (merged in #261). Before, it was initialized in the first-bar preamble, before that statement, and read the zero default.

//@version=6
strategy('varip gap', initial_capital = 100000)
type State
    float value
float base = strategy.initial_capital / 100
varip state = State.new(value=base)
observed = state.value

observed was 0; it is now 1000, the same as the inline form varip state = State.new(value=strategy.initial_capital / 100) and the var form.

How

The varip declaration takes the same declaration-site route as var when its initializer depends on an earlier statement (a two-line change in the statement-order check). Independent initializers and callable-scoped declarations keep their routes and their emission. varip storage and its once flag stay outside the calc_on_order_fills rollback checkpoint, as before: a value set on the first bar survives a fill recalculation and the initializer runs once.

Tests

• Emission order of the dependency and the declaration, the compiled run against the inline and var controls for each aggregate kind (UDT, array, map, matrix, UDT built by a helper), the issue's own script in the static and dynamic builds, and that the first value is kept when the dependency changes later. All of them fail on main (the value is 0 against 1000, or the declaration comes first).
• A guard that independent varip initializers keep their preamble emission byte for byte.
• With a native engine: a calc_on_order_fills strategy with a dependent varip aggregate keeps its first value across fill recalculations (it reads 0 on main), and a static test that the varip storage and its once flag stay out of the rollback checkpoint and the restore.
• The test that pinned the old behaviour as a known gap is removed; its shapes are covered by the new tests.
• Full suite with the paired engine: 10881 passed, 221 skipped (private fixtures and empty parameter sets, the same 221 as main), 0 failed.
• Over the 555 corpus and fixture scripts the emitted C++ is identical to main: none of them has a dependent varip aggregate initializer.

The CHANGELOG entry of #226 now says the varip form takes the same route.

Fixes #258

🤖 Generated with Claude Code