bugc: lay storage arrays out as Solidity does

[gnidan]

Storage arrays now have Solidity's layout.

• A fixed-size array is inline: its elements start at its own slot. Before, the code put every array's elements at keccak256(slot) + i, but the debug pointer of a fixed-size array described them inline at slot + i, so a debugger read the wrong slots.
• A dynamic array still keeps its length in its slot and its elements from keccak256(slot).
• In either kind of array, an element narrower than 16 bytes shares a slot with its neighbors, as many as fit, from the slot's low-order end, as Solidity packs them. For a dynamic array this changes the code too. The pointer already described narrow elements as packed (with the wrong offsets), but the code gave each element a slot, so a pointer and the code could not agree without one of them changing, and Solidity's layout packs them. Any other element starts a slot and takes as many slots as its type needs (a struct, or an inner fixed-size array).
• In a struct, a fixed-size array member takes all its slots, and the next member starts the slot after them. Before, it took one slot.

How it is built:

• types/storage.ts is the one statement of the layout: an elementary type's size, how many elements share a slot, and how many slots a type takes. The struct layout, the code, and the pointers all use it.
• In IR generation, emitArrayElement replaces the two copies of the array branch in the storage chain load and store, and the array literal's storage writes. A fixed-size array literal no longer writes a length word into its first element's slot. For an element that shares its slot, the index's slot and byte offset are computed at run time (div/mod/mul, which constant folding removes for a constant index).
• Code generation now accepts a storage read/write offset that is not a constant: it shifts by offset << 3 at run time.
• The storage pointer generator takes a slot expression, so a struct or an array inside a list gets the element's own slot (before, it used slot 0). Each nested list has its own index variable.

arrays.bug declared its scalars at slots 1 to 5, inside its 10-slot array. They now start at slot 10. The test that expected one element of an int16 array at keccak256(0) + 1, and the IR test that expected a compute_slot for a fixed-size array in a mapping, now expect the inline layout. The storage verification test needed a local node and was skipped. It now runs on the in-process executor, and it checks the inline slots it always expected. The BUG case study describes the layout.

Known limit, unchanged here: a struct member that is itself a struct still takes one slot.

Conflicts to expect: #355 changes generateStorageWrite next to the offset handling here, and #361 changes irgen/generate/storage.ts.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-10-07 03:11 UTC

[gnidan]

Included in #368.