Integer induction counters live in unsigned 64-bit slots, so every read
as a number was a uitofp i64 (a 6-7 instruction sequence on x86-64) and
every nested start an fptoui. The runtime's push path also round-tripped
the array length through a double for its bound check.
The values are at most 2^32 - 1: on 64-bit targets convert counters with
sitofp and loop starts with fptosi, convert array lengths and validated
indices through int64_t in scr_array.c, and test the push bound on the
integer length. 32-bit targets keep the unsigned conversions.
tsc-ts Excalidraw: uitofp i64 sequences 32.5 M -> 4.0 M, -0.66%
instructions on top of the throw-path change.
With runtime bitcode import, scr_throw_obj, scr_error_new and the other
throw entry points were inlined into the error paths of small compiled
functions. The inlined bodies needed callee-saved registers that every
call then saved and restored, and they made the helpers too large to
inline into their callers (1047 inlined throw bodies in tsc-ts).
Mark the 25 functions that only run while raising a JS exception
cold and noinline. The LLVM helper never imports noinline bodies, so
the throw paths stay calls. tsc-ts Excalidraw: -0.44% instructions.
Every SCR_ELEM_REF array stored its element type's retain, release and trace entry points in its own header (24 bytes), so ScrArr was 80 bytes: an 80-byte block for arrays of acyclic elements and a 96-byte block for cycle-capable ones. The three pointers now live in one constant ScrArrOps descriptor per element type, which the compiler emits once per module (@sc_arrops.*) and the runtime keeps static, and the header holds a pointer to it plus a 4-byte `cyc` flag beside `elem` that the inline RC fast paths test. ScrArr is 64 bytes: arrays take 64- and 80-byte blocks.
scr_arr_new_ref(retain, release, trace, cap) becomes scr_arr_new_ops(ops, cap); scr_set_to_arr_ref takes the result's descriptor from its caller instead of building an array from the set's key adapters. The backend names the %ScrArr fields in SCR_ARR_FIELD. The layout change bumps the runtime ABI marker to scr_runtime_abi_v13.
Program-lifetime instances were created with rc == SIZE_MAX, which every runtime contract reads as "published": the frozen-object guards throw on stores, postMessage and workerData share the object by pointer instead of cloning it, and the publish walker stops there. So programs that publish could not use ProgramLifetime (SC7007 refused it), and an unpublished instance posted to a worker would have been shared instead of cloned.
They are now created with the midpoint count 2^63 (2^31 on 32-bit targets). Only the class's empty retain and release helpers count references on these types, so nothing ever changes or frees that count; the midpoint keeps even an unpaired generic retain or release away from 0 and SIZE_MAX. Until the program publishes an instance it is an ordinary mutable, thread-local object to every runtime contract: stores succeed, postMessage clones it. Publication marks it SIZE_MAX like any object it reaches (the visited mark, logged and restored when a refusal undoes the publish), after which the frozen guards throw Node's TypeErrors and other threads read it by pointer. No header, extra word, runtime change or new instruction: the count itself is the published mark. The cycle analysis keeps the hierarchies untraced, so the collector never visits an instance, as for every untraced class.
SC7007 no longer refuses program-lifetime classes reachable from a publish() call; it still requires a root class in an executable. Without the marker nothing changes: tsc-ts best-v7 built with and without this change is byte-identical.
Tests: corpus 4634-contract-program-lifetime-publish (Node vs native): program-lifetime hierarchies with parent cycles, maps and sets of them, maps keyed by them and mortal scopes around them, published on the main thread and a worker and read on the other; writes after publication throw on both threads; unpublished instances stay writable; a refused publish leaves visited instances writable; a later publish reaches published and fresh instances; an unpublished instance in workerData is cloned. With the old SIZE_MAX count the program dies at its first constructor store. classes.test.ts covers the count on both pointer widths, the walker and the frozen guard. The contract-program-lifetime-publish diagnostics fixture is removed (the program now compiles). Docs: contracts and limitations pages, @scriptc/types.
Program-lifetime instances are never freed, and neither is anything they hold, so the RC audit of --sanitize builds reported every one of them, and every string, array and map they reference, as live at exit (4633-contract-program-lifetime failed the sanitized differential lane: 306 strings, 15 arrays, 338 objects). Like a program that abandons fibers, a program that keeps program-lifetime objects now skips the leak report with "scriptc RC audit skipped: N program-lifetime object(s) never freed", which the harnesses already filter.
Sanitized constructors call scr_obj_lifetime_note(obj), which counts the instance and registers it with LeakSanitizer as an intentional root, as publication does for published objects: once the audit no longer exits early, LeakSanitizer would otherwise report the instances the program drops. Plain builds emit nothing.
Chris Tate