The test from #1548 casts JSON.parse to Partial<Envelope> into a
block-local const, which continues as its parsed value since #1561, as
in Node. The path messages are now pinned with the cast passed to a
typed parameter, which is checked at the cast, and the block-local form
is pinned against Node.
Another module can import an exported const (`export const` or an
`export { name }` specifier) and reach its elements, so a spread of it
is not their only reference. Those spreads keep the element-layout
refusal instead of copying the elements.
A const initialized by an array literal and spread once can still run
that spread more than once: in a loop body, loop test or update, a
function, or a class field initializer. Each run inserts the same
elements in JavaScript, so copying them into the destination's layout
would break identity. Those spreads keep the element-layout refusal.
Each armed site lowers the code after it twice, and a cast inside another
site's tail is lowered in both copies, so n sequential casts produced 2^n
copies of the code after the last one. A site now arms only while fewer
than two splits enclose its tail or share its split point; later casts
keep the throwing cast.
The stream probes for untyped values (pipe destinations, pipeline
stages, readline input and output, and the checks against the stream
classes) read the vtable word of any boxed object. Instances with a
class id in the rc word have no vtable word, so a hierarchy instance
whose first field is a number crashed those probes. scr_stream_vt_of
answers NULL for them, and each probe treats that as no stream.
workloads.json listed every workload in one array, and its README table
listed them in one hand order, so changes that each added a workload
appended at the same place and conflicted. Each workload is now a JSON
file in benchmarks/runtime/workloads/ named after it, read through
benchmarks/runtime/workloads.mjs by the bench runner, the zero-cost
gate, the strace script and the dynamic-tier test, ordered by category
and then name. The loader still reads workloads.json in a checkout from
before this change, so the gate can compare against one. The README
table follows the same order.
The lib-fn sets and prefix lists in runtime-features.ts grew by
appending, so changes that added different entries to one list touched
the same last lines and conflicted. Each list's entries are now sorted
within its commented groups, and the all-off feature literal follows the
rules table's name order. Entry order is unobservable: the lists feed
membership maps and prefix tests that each set their own flag.
#1360 moved the content hash and row count into next.config.mjs and
deleted node-v24-compatibility-meta.json, but a later merge kept a
branch's copy. Nothing reads it, and its hash line conflicts on every
compatibility change. The local compatibility check now fails while the
file exists, so a merge can't bring it back unnoticed.
When an operand of a bitwise operator is a hoisted optional boolean, the
operands evaluate into temporaries before the operation. A left operand
that is safe to repeat, such as a variable read, kept its place in the
operation while the right operand's temporary ran first, so a right
operand that assigns that variable changed the value the left operand
read: `seen ^ (seen = !seen)` gave 0 where Node gives 1.
The left operand now takes a temporary whenever the right one does, so
both evaluate in source order. The corpus program covers an assignment
in the right operand of ^ and <<.
fs.realpathSync and fs.realpathSync.native read as function values
lower to the fs.realpathValue libCall. The library sidecar's determinism
attestation demotes on every fs.* spelling, so a fence on realpathSync
has to witness that one too. It is now an alias of both members, and
the surface-manifest attestation parity test passes again.
Chris Tate