`next/kernel/http_server.erl` gains `start/1(Port)` + `start/2(Port, Cfg)`. Both spawn an Erlang process that hosts
the native `http:listen/2` accept loop with the Cfg-aware `route/2` as the handler.
The blocker — the BIF wrapper in `lib/erlang/runtime.sx` had no dict↔proplist marshaling, so Erlang handler funs
couldn't pattern-match on an opaque SX request dict — is resolved by a new family of helpers added next to `er-of-sx`
(which is left untouched so non-HTTP callers see no behavioural drift):
er-request-dict-to-proplist request dict -> [{method,<<>>},{path,<<>>},...] (atom keys)
er-of-sx-deep recursive marshal: dicts -> binary-keyed proplist
er-dict-to-header-proplist headers: [{<<"content-type">>,<<"text/plain">>},...]
(binary keys keep arbitrary user input out of the atom table)
er-proplist-to-dict response proplist -> SX dict for native serialiser
er-proplist-fill! dict-set! walker over a cons-of-2-tuples
er-to-sx-deep recursive marshal: cons-of-2-tuples -> nested dict
er-proplist-2tuple? predicate distinguishing a header proplist from a binary body
`er-bif-http-listen`'s body is updated to route through the new pair instead of `er-of-sx` / `er-to-sx`. Existing
`http_listen_bif.sh` (Step 8a) still passes — the BIF's external contract (port + handler validation, registration)
hasn't changed, only the request/response shape the handler sees.
This commit also lands a small pre-existing unstaged refactor that was sitting in the same file (er-binary->string
helper above er-bif-http-listen, a "Register everything at load time." comment move, and the binary_to_list /
list_to_binary / er-iolist-walk! defines reshuffled into the er-register-builtin-bifs! body). The refactor was
agreed-out-of-scope earlier in the loop but was unblocked this iteration when the user OK'd progress on 8b-start.
Bundling it here keeps the lib/erlang/runtime.sx diff coherent.
Tests:
- `next/tests/http_marshal.sh` (10 cases) — marshaling unit tests: request dict → cons proplist; method as
<<"GET">> via SX-side proplist walker; path-as-string roundtrip; nested headers reach through binary keys;
response status/body field marshaling; nested headers reconstruct dict; full round-trip preserves status.
- `next/tests/http_server_start.sh` (6 cases) — structural verification: http_server module loaded, start bound
in module env, marshalers defined as lambdas, http:listen BIF registered. Can't invoke spawn in an Erlang test
because the cooperative scheduler (`er-sched-run-all!`) drains every runnable process before returning to the
caller, and the listener's accept loop never exits.
- `next/tests/http_server_tcp.sh` (5 cases) — **first live end-to-end transport test in the milestone**: boots
sx_server in background with FIFO-held stdin (~10s boot for all lib/erlang/*.sx loads + module compile +
Unix.bind), then drives the listener via shell-side curl over real TCP. Verifies GET / → 200, GET
/.well-known/sx-capabilities → 200, GET unknown → 404, POST /activity → 401 with no/bad bearer. Doubles as the
smoke surface for 9a-tcp / 9b-tcp.
Erlang conformance **761/761** unchanged. All standing suites stay green (http_listen_bif 5/5, log_disk 12/12,
log_rotate 10/10, term_codec 18/18).
Step 8b-start ticked in plans/fed-sx-milestone-1.md. Remaining in the milestone: 9a-tcp / 9b-tcp — partly covered
by http_server_tcp.sh's smoke probes; the full curl-driven publish flows are the next iteration.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase 1 of the jit-perf-regression plan reproduced and quantified the alleged
30× substrate slowdown across 5 guests (tcl, lua, erlang, prolog, haskell). On
a quiet machine all five suites pass cleanly:
tcl test.sh 57.8s wall, 16.3s user, 376/376 ✓
lua test.sh 27.3s wall, 4.2s user, 185/185 ✓
erlang conformance 3m25s wall, 36.8s user, 530/530 ✓ (needs ≥600s budget)
prolog conformance 3m54s wall, 1m08s user, 590/590 ✓
haskell conformance 6m59s wall, 2m37s user, 156/156 ✓
Per-test user-time at architecture HEAD vs pre-substrate-merge baseline
(83dbb595) is essentially flat (tcl 0.83×, lua 1.4×, prolog 0.82×). The
symptoms reported in the plan (test timeouts, OOMs, 30-min hangs) were heavy
CPU contention from concurrent loops + one undersized internal `timeout 120`
in erlang's conformance script. There is no substrate regression to bisect.
Changes:
* lib/tcl/test.sh: `timeout 2400` → `timeout 300`. The original 180s deadline
is comfortable on a quiet machine (3.1× headroom); 300s gives some safety
margin for moderate contention without masking real regressions.
* lib/erlang/conformance.sh: `timeout 120` → `timeout 600`. The 120s budget
was actually too tight for the full 9-suite chain even before this work.
* lib/erlang/scoreboard.{json,md}: 0/0 → 530/530 — populated by a successful
conformance run with the new deadline. The previous 0/0 was a stale
artefact of the run timing out before parsing any markers.
* plans/jit-perf-regression.md: full Phase 1 progress log including
per-guest perf table, quiet-machine re-measurement, and conclusion.
Phases 2–4 (bisect, diagnose, fix) skipped — there is no substrate regression
to find. Phase 6 (perf-regression alarm) still planned to catch the next
quadratic blow-up early instead of via watchdog bumps.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
