post-W7 cleanup

This commit is contained in:
root
2026-06-15 12:00:46 +00:00
parent 75bccb778c
commit 1a5c6547b9
28 changed files with 896 additions and 301 deletions
-90
View File
@@ -1,90 +0,0 @@
# CLI Unit Test Port Plan
## Objective
Replace the Python-based network test runner and plugins with UCE unit tests invoked through the runtime's CLI socket path, keeping equivalent coverage for HTTP smoke, site suites, security checks, starter parity, TCP/WebSocket listener probes, and wasm kill checks. The final invocation should be a bash script that calls UCE CLI units; Python test files should be removed only after UCE coverage is in place and validated.
## Success Criteria
- [x] A bash command runs the full test suite through the CLI socket and exits nonzero on failure.
- [x] UCE CLI tests cover every current Python plugin behavior or explicitly document a deliberate replacement.
- [x] Existing W5/WASM gates use the UCE CLI test runner instead of `tests/run_network_tests.py`.
- [x] Python test runner/plugins are deleted after parity validation.
- [x] Full suite passes on `uce-dev` with wasm-only unit execution.
## Current State
- Status: verifying
- Last updated: 2026-06-13
- Source of truth: `/root/mount_ssh/uce-dev-root-htdocs-uce`
- Runtime/live target: `uce-dev:/Code/uce.openfu.com/uce`
## Goal Tree
Legend: `[ ]` not started, `[~]` in progress, `[x]` done, `[!]` blocked, `[-]` superseded
- [x] G1: Inventory Python test coverage and CLI constraints
- Why: replacement must preserve coverage before deleting Python.
- Done when: every plugin has a mapped UCE equivalent or blocker.
- Verify: coverage matrix in this document.
- [x] G1.1: Delegate design/coverage review to Spark.
- [x] G1.2: Inspect delegates and reconcile plan.
- [x] G2: Build UCE CLI test harness
- Why: Python runner features need a UCE-native replacement.
- Done when: one CLI unit can list/run tests, print pass/fail summary, and return failing CLI status.
- Verify: `scripts/run_cli_tests.sh --list` and `scripts/run_cli_tests.sh --include-wasm-kill`.
- [x] G2.1: Add reusable UCE assertion/reporting helpers.
- [x] G2.2: Add HTTP/TCP helper functions using UCE socket APIs.
- [x] G2.3: Add bash wrapper under `scripts/`.
- [x] G3: Port current plugin cases to UCE
- Why: only delete Python after equivalent UCE coverage exists.
- Done when: UCE suite covers demo, HTTP docs/starter, site suites, security, starter parity, TCP, wasm kill.
- Verify: UCE CLI full run passes and output names match coverage matrix.
- [x] G4: Replace Python gate usage and delete Python tests
- Why: user explicitly requested eliminating the Python suite.
- Done when: scripts no longer call `tests/run_network_tests.py`, Python test files removed, validation green.
- Verify: `rg 'run_network_tests|tests/plugins|python3 tests'` has no obsolete gate references except historical docs and benchmark/audit utilities.
- [~] G5: Document and validate
- Why: future agents/operators need the new test workflow.
- Done when: docs/project notes and in-repo docs mention the CLI test command and validation artifact.
- Verify: docs committed, full suite run artifact recorded.
## Coverage Matrix
- `uce_demo_smoke.py``cli_run_demo_smoke()` in `site/tests/cli_runner.uce` (43 demo pages).
- `uce_http_smoke.py``cli_run_http_smoke()` (docs and starter route/body checks, 14 cases).
- `uce_site_suite.py``cli_run_site_suite()` (manifest-driven published site suite pages, 13 cases).
- `uce_security_smoke.py``cli_run_security_smoke()` (direct HTTP traversal/header spoofing, CRLF header sanitization, session hardening, 4 cases).
- `uce_starter_parity.py``cli_run_starter_parity()` (starter view title/404 checks, 7 cases).
- `uce_tcp_smoke.py``cli_run_tcp_smoke()` (port 80 and 8080 reachability, 2 cases).
- `uce_wasm_kill.py``cli_run_wasm_kill()` gated by `--include-wasm-kill` (trap/loop/recurse + post-kill health checks, 3 cases).
## Execution Queue
1. Run final no-Python-suite validation after removing stale references.
2. Commit UCE and project-doc updates.
## Decisions
- 2026-06-13: Use UCE CLI socket invocation as the test entrypoint; bash wrappers are acceptable, Python runner/plugins are not.
- 2026-06-13: Do not delete Python tests until UCE replacement validates equivalent coverage on `uce-dev`.
## Assumptions
- UCE socket APIs are sufficient for HTTP/1.0 probes, TCP connect checks, and security header injection checks.
- Bash can provide filtering/list convenience if exact Python CLI parity is not needed.
## Blockers and Risks
- [ ] The new runner intentionally does not preserve the old Python runner's dynamic plugin/tag/regex filtering; add UCE-side selectors later if operators miss them.
- [ ] `tests/wasm_benchmark.py` and `tests/wasm_site_audit.py` remain Python utility scripts, not the network test suite; port separately if a strict no-Python tools policy is desired.
## Evidence and Verification Log
- 2026-06-13: Prior to this plan, `b1856c1 chore: narrow wasm backend entrypoint API` was committed after build and focused services validation.
- 2026-06-13: `scripts/run_cli_tests.sh --include-wasm-kill` on `uce-dev` passed `86 passed, 0 failed, 0 skipped` (`/tmp/uce/cli-tests-final.txt`).
## Change Log
- 2026-06-13: Created initial goal tree for CLI unit test port.
- 2026-06-13: Added `site/tests/cli_runner.uce`, `scripts/run_cli_tests.sh`, removed Python network runner/plugins, and repointed W5 network gates to the CLI runner.
+565
View File
@@ -0,0 +1,565 @@
# UCE Runtime Setup
This guide describes how to run UCE behind nginx or Apache. UCE is a FastCGI application server for `.uce` units; the web server should serve static files directly and forward dynamic `.uce` requests to the UCE runtime.
## Deployment shape
A typical deployment has four pieces:
1. A checked-out or packaged UCE runtime tree.
2. `/etc/uce/settings.cfg`, read by the UCE runtime at startup.
3. `uce.service`, a systemd service that builds/starts/restarts the runtime.
4. nginx or Apache as the public HTTP server.
Recommended filesystem layout for a source checkout:
```text
/opt/uce/ UCE repository/runtime root
/var/www/html/ public web root served by nginx/Apache
/etc/uce/settings.cfg runtime configuration
/run/uce/fastcgi.sock FastCGI socket used by nginx/Apache
/run/uce/cli.sock local CLI/admin/test socket
/var/cache/uce/work generated source, wasm modules, caches
/var/lib/uce/uploads multipart upload scratch space
/var/lib/uce/sessions session files
```
For packaged installs, the runtime may live under `/usr/lib/uce` instead of `/opt/uce`. Keep the public web root at `/var/www/html` or another normal web-root path, not under the runtime source tree.
## Build requirements
On Debian/Ubuntu-like systems, install the distro packages first:
```bash
apt update
apt install -y clang build-essential libpcre2-dev mariadb-client libmariadb-dev curl rsync ca-certificates
```
UCE also requires two toolchains that are not vendored in this repository:
- **Wasmtime C API / C++ headers** at `/opt/wasmtime` by default. `scripts/build_linux.sh` expects:
- `/opt/wasmtime/include/wasmtime.hh`
- `/opt/wasmtime/include/wasmtime/*.h`
- `/opt/wasmtime/lib/libwasmtime.so`
- **WASI SDK** at `/opt/wasi-sdk` by default. `scripts/build_core_wasm.sh` and `scripts/compile_wasm_unit` expect:
- `/opt/wasi-sdk/bin/clang++`
- `/opt/wasi-sdk/bin/wasm-ld`
- `/opt/wasi-sdk/bin/llvm-objcopy`
You can use different install locations by setting environment variables before building and in the systemd service environment:
```bash
export WASMTIME_HOME=/path/to/wasmtime
export WASI_SDK=/path/to/wasi-sdk
```
Install one web server:
```bash
apt install -y nginx
# or
apt install -y apache2
```
Build UCE from the repository root:
```bash
cd /opt/uce
bash scripts/build_core_wasm.sh
bash scripts/build_linux.sh
```
Publish the starter site or your application files into the web root:
```bash
mkdir -p /var/www/html
rsync -a site/ /var/www/html/
```
The main binary is written to:
```text
bin/uce_fastcgi.linux.bin
```
### Installing Wasmtime and WASI SDK
The UCE build does not download these dependencies for you. Install them through your distro if it provides compatible development packages, or unpack pinned release archives under `/opt/wasmtime` and `/opt/wasi-sdk`.
Do not use `curl | sh` installers in production setup scripts. Download archives from the upstream release pages, verify checksums/signatures when available, and keep the exact versions recorded in your deployment notes.
The expected directory shape is:
```text
/opt/wasmtime/include/wasmtime.hh
/opt/wasmtime/include/wasmtime/store.h
/opt/wasmtime/lib/libwasmtime.so
/opt/wasi-sdk/bin/clang++
/opt/wasi-sdk/bin/wasm-ld
/opt/wasi-sdk/bin/llvm-objcopy
```
If your paths differ, export the variables for manual builds:
```bash
WASMTIME_HOME=/usr/local/wasmtime WASI_SDK=/usr/local/wasi-sdk bash scripts/build_core_wasm.sh
WASMTIME_HOME=/usr/local/wasmtime WASI_SDK=/usr/local/wasi-sdk bash scripts/build_linux.sh
```
For systemd, add an override:
```bash
systemctl edit uce.service
```
```ini
[Service]
Environment=WASMTIME_HOME=/usr/local/wasmtime
Environment=WASI_SDK=/usr/local/wasi-sdk
```
Then reload and restart:
```bash
systemctl daemon-reload
systemctl restart uce.service
```
## Runtime configuration
Create `/etc/uce/settings.cfg` from `etc/uce/settings.cfg` and adjust paths if your runtime is not under `/opt/uce`.
Minimum useful settings:
```ini
BIN_DIRECTORY=/var/cache/uce/work
TMP_UPLOAD_PATH=/var/lib/uce/uploads
SESSION_PATH=/var/lib/uce/sessions
FCGI_SOCKET_PATH=/run/uce/fastcgi.sock
CLI_SOCKET_PATH=/run/uce/cli.sock
SITE_DIRECTORY=/var/www/html
JIT_COMPILE_ON_REQUEST=1
PROACTIVE_COMPILE_ENABLED=1
PROACTIVE_COMPILE_CHECK_INTERVAL=60
WASM_COMPILE_SCRIPT=scripts/compile_wasm_unit
WASM_BACKEND_VERBOSE=0
WASM_CORE_PATH=/opt/uce/bin/wasm/core.wasm
WASM_MEMORY_LIMIT_BYTES=536870912
WASM_EPOCH_DEADLINE_TICKS=200
WASM_EPOCH_PERIOD_MS=50
WORKER_COUNT=4
MAX_MEMORY=16777216
SESSION_TIME=2592000
HTTP_PORT=8080
```
Important settings:
- `FCGI_SOCKET_PATH` is the Unix socket used for normal `.uce` requests.
- `CLI_SOCKET_PATH` is a local HTTP-over-Unix socket used by `scripts/uce-cli` and test/admin units.
- `SITE_DIRECTORY` is the public site tree to scan for `.uce` files. Use `/var/www/html` when the web root is outside the runtime tree; relative paths are resolved from the runtime working directory.
- `BIN_DIRECTORY` stores generated C++, wasm artifacts, compile output, and runtime caches.
- `TMP_UPLOAD_PATH` and `SESSION_PATH` must be writable by the runtime.
- `HTTP_PORT` is the built-in HTTP/WebSocket listener used for WebSocket upgrade traffic and direct local probes.
- `WASM_CORE_PATH` must point at the built `core.wasm` file.
After editing settings, restart UCE:
```bash
systemctl restart uce.service
```
## systemd service
For source-checkout deployments, install the provided service helper:
```bash
cd /opt/uce
scripts/systemd/manage-uce-service.sh setup
```
That helper installs `scripts/systemd/uce.service`, creates runtime directories, enables the service, and starts it.
Useful commands:
```bash
scripts/systemd/manage-uce-service.sh status
scripts/systemd/manage-uce-service.sh restart
scripts/systemd/manage-uce-service.sh logs 200
```
Equivalent manual systemd service for a source checkout:
```ini
[Unit]
Description=UCE FastCGI Runtime
After=network-online.target mariadb.service memcached.service
Wants=network-online.target
[Service]
Type=simple
WorkingDirectory=/opt/uce
RuntimeDirectory=uce
StateDirectory=uce
CacheDirectory=uce
ExecStartPre=/usr/bin/mkdir -p /var/cache/uce/work /var/lib/uce/uploads /var/lib/uce/sessions
ExecStartPre=/usr/bin/rm -f /run/uce/fastcgi.sock
ExecStartPre=/usr/bin/bash /opt/uce/scripts/build_linux.sh
ExecStart=/opt/uce/bin/uce_fastcgi.linux.bin
ExecStopPost=/usr/bin/rm -f /run/uce/fastcgi.sock
Restart=always
RestartSec=2
TimeoutStopSec=15
KillMode=mixed
StandardOutput=journal
StandardError=journal
[Install]
WantedBy=multi-user.target
```
Install it as `/etc/systemd/system/uce.service` and run:
```bash
systemctl daemon-reload
systemctl enable --now uce.service
```
## How request routing works
### Static files
The web server should serve ordinary static files directly from the public web root, for example `/var/www/html`.
Examples:
```text
/style.css
/images/logo.png
/examples/uce-starter/js/site.js
```
These should not touch the UCE runtime.
### Normal `.uce` page requests
For a request such as:
```text
GET /doc/index.uce?p=component
```
nginx/Apache forwards the request to `FCGI_SOCKET_PATH` as FastCGI. The web server must provide CGI/FastCGI variables including:
- `SCRIPT_FILENAME` — full filesystem path to the `.uce` file.
- `DOCUMENT_ROOT` — public web root, normally `<runtime-root>/site`.
- `SCRIPT_NAME` — URL path to the script, such as `/doc/index.uce`.
- `DOCUMENT_URI` — normalized URI path without query string.
- `REQUEST_URI` — original request URI including query string.
- standard request variables such as method, query string, content type, body length, cookies, and headers.
UCE resolves the unit, compiles it to wasm if needed, creates a request workspace, and calls:
```cpp
RENDER(Request& context)
```
The unit writes output with template literals or `print()`. Response headers and status are set through `context.header` and `context.set_status()`.
### Component and sub-render calls
Inside a request, UCE code can call other units:
```cpp
component("components/card", props, context);
unit_render("other-page.uce", context);
```
These calls stay inside the UCE runtime. They are not new HTTP requests and do not go back through nginx or Apache.
### WebSocket pages
Any `.uce` unit can provide both an ordinary page render and WebSocket message handling:
```cpp
RENDER(Request& context) { ... } // normal page load
WS(Request& context) { ... } // later WebSocket messages
```
The nginx and Apache examples below split traffic by checking for a WebSocket upgrade request on `.uce` paths. A file such as `chat.uce` or `events.uce` can expose `WS(Request& context)`.
Routing split:
- Plain `GET /demo/chat.uce` should use FastCGI, just like any other page render.
- WebSocket upgrade requests for `/demo/chat.uce` should proxy to the UCE built-in HTTP/WebSocket listener at `HTTP_PORT`.
The built-in listener owns the socket lifecycle. When a message arrives, the broker forwards a render-style invocation back to the worker pool so `WS(Request& context)` runs inside the same wasm runtime model as normal pages.
### CLI requests
`CLI(Request& context)` handlers are not public web endpoints. They are invoked over `CLI_SOCKET_PATH`:
```bash
scripts/uce-cli /tests/cli.uce action=echo message=hello
curl --unix-socket /run/uce/cli.sock http://localhost/tests/cli.uce
```
Use CLI units for local tests, admin commands, and maintenance tools. Do not expose the CLI socket through nginx or Apache.
### Custom runtime HTTP servers
UCE code can start local custom HTTP listeners with `server_start_http()`. Those are runtime-managed listeners for app-specific local services. They are separate from the public nginx/Apache entry point and should be firewalled or bound locally unless you explicitly want them reachable.
## nginx configuration
### Required modules
A normal nginx build includes the needed FastCGI and proxy modules. Confirm nginx is installed and can load your config:
```bash
nginx -t
```
### WebSocket upgrade map
Put this in the nginx `http` block if using WebSockets:
```nginx
map $http_upgrade $connection_upgrade {
default upgrade;
'' close;
}
```
### Server block
Example site config:
```nginx
server {
listen 80;
server_name example.com;
root /var/www/html;
index index.uce index.html;
# Serve static files directly. Directory requests use index.uce when present.
location / {
try_files $uri $uri/ =404;
}
# UCE page requests use FastCGI. If the client asks to upgrade a .uce
# request to WebSocket, send that connection to the built-in listener.
location ~ \.uce$ {
error_page 418 = @uce_websocket;
if ($http_upgrade = "websocket") {
return 418;
}
include fastcgi_params;
fastcgi_param SCRIPT_FILENAME $document_root$fastcgi_script_name;
fastcgi_param DOCUMENT_ROOT $document_root;
fastcgi_param SCRIPT_NAME $fastcgi_script_name;
fastcgi_param DOCUMENT_URI $uri;
fastcgi_param REQUEST_URI $request_uri;
fastcgi_pass unix:/run/uce/fastcgi.sock;
}
location @uce_websocket {
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection $connection_upgrade;
proxy_pass http://127.0.0.1:8080;
}
# Defense in depth if the root is changed later.
location ~ ^/(src|scripts|etc|bin|work|dist|pkg|docs|changelog)/ {
return 404;
}
}
```
Notes:
- `fastcgi_pass` must match `FCGI_SOCKET_PATH`.
- `proxy_pass` must match `HTTP_PORT`.
- The example routes WebSocket upgrades for `.uce` paths to the HTTP/WebSocket listener.
- Ordinary `.uce` page loads continue to use FastCGI.
- Keep `root` pointed at `/var/www/html`, not the runtime repository root.
- If your app uses a front controller, replace `location /` with a `try_files` rule that ends at `/index.uce`.
Front-controller variant:
```nginx
location / {
try_files $uri $uri/ /index.uce?$query_string;
}
```
Reload nginx:
```bash
nginx -t
systemctl reload nginx
```
## Apache configuration
Apache can run UCE through `mod_proxy_fcgi` for FastCGI and `mod_proxy_wstunnel` or `mod_proxy_http` for WebSocket upgrades.
### Enable modules
On Debian/Ubuntu:
```bash
a2enmod proxy proxy_fcgi proxy_http proxy_wstunnel rewrite headers setenvif
systemctl restart apache2
```
### VirtualHost example
```apache
<VirtualHost *:80>
ServerName example.com
DocumentRoot /var/www/html
<Directory /var/www/html>
Require all granted
Options FollowSymLinks
AllowOverride None
DirectoryIndex index.uce index.html
</Directory>
# Do not expose repository internals if DocumentRoot changes later.
<LocationMatch "^/(src|scripts|etc|bin|work|dist|pkg|docs|changelog)/">
Require all denied
</LocationMatch>
RewriteEngine On
# WebSocket upgrade traffic for any .uce unit goes to UCE's built-in HTTP listener.
RewriteCond %{HTTP:Upgrade} =websocket [NC]
RewriteCond %{REQUEST_URI} \.uce(?:\?|$) [NC]
RewriteRule ^/(.*)$ ws://127.0.0.1:8080/$1 [P,L]
# Normal .uce page loads go to FastCGI.
<FilesMatch "\.uce$">
SetHandler "proxy:unix:/run/uce/fastcgi.sock|fcgi://localhost/"
</FilesMatch>
# Optional: make the key CGI variables explicit for UCE.
ProxyFCGISetEnvIf "true" DOCUMENT_ROOT "/var/www/html"
ProxyFCGISetEnvIf "true" DOCUMENT_URI "%{REQUEST_URI}"
</VirtualHost>
```
Apache notes:
- `SetHandler "proxy:unix:/run/uce/fastcgi.sock|fcgi://localhost/"` must use the same socket path as `FCGI_SOCKET_PATH`.
- The WebSocket rewrite rule must run before the FastCGI handler.
- Plain `.uce` page loads should not be proxied as WebSockets unless the client sends `Upgrade: websocket`.
- Apache's FastCGI environment differs by version and module configuration. If UCE cannot resolve a page, inspect the request environment and make sure `SCRIPT_FILENAME` points to the target file under the web root.
If your Apache version does not populate `SCRIPT_FILENAME` correctly through `SetHandler`, use `ProxyPassMatch` for `.uce` files instead:
```apache
ProxyPassMatch ^/(.*\.uce)$ unix:/run/uce/fastcgi.sock|fcgi://localhost/var/www/html/$1
```
Use only one FastCGI mapping style at a time (`SetHandler` or `ProxyPassMatch`) to avoid duplicate routing.
## Permissions
The web server needs permission to connect to `/run/uce/fastcgi.sock`. Common approaches:
- run UCE and the web server under compatible groups;
- add the web server user (`www-data` on Debian/Ubuntu) to the socket's group;
- adjust the service or runtime socket mode if needed.
The runtime creates the FastCGI socket and CLI socket under `/run/uce`. The CLI socket should remain local-only and should not be reachable from the public web server.
Writable paths for the runtime:
```text
/var/cache/uce/work
/var/lib/uce/uploads
/var/lib/uce/sessions
/run/uce
```
## Verification
Check service state:
```bash
systemctl status uce.service
journalctl -u uce.service -n 100 --no-pager
```
Check the FastCGI/web-server path:
```bash
curl -i http://127.0.0.1/doc/index.uce -H 'Host: example.com'
curl -i http://127.0.0.1/examples/uce-starter/ -H 'Host: example.com'
```
Check the local CLI path:
```bash
cd /opt/uce
scripts/uce-cli /tests/cli.uce action=echo message=hello
scripts/run_cli_tests.sh
```
Check WebSocket routing with a WebSocket client against a `.uce` endpoint that defines `WS(Request& context)` through nginx/Apache, not directly against `HTTP_PORT`.
## Troubleshooting
### 502 Bad Gateway
Check:
- `systemctl status uce.service`
- `journalctl -u uce.service -n 200 --no-pager`
- socket path in web server config equals `FCGI_SOCKET_PATH`
- web server user can connect to the Unix socket
- `SCRIPT_FILENAME` resolves to an existing `.uce` file
### Raw `.uce` source is downloaded or displayed
The `.uce` request did not match the FastCGI rule. Check location/order rules and confirm the public root is `/var/www/html` or your chosen web-root path.
### Static files 404
Confirm the web server `root`/`DocumentRoot` is `/var/www/html` or your chosen web-root path and that `location /` or Apache directory rules allow static file reads.
### WebSocket page renders but upgrade fails
Check:
- the client sends `Upgrade: websocket`
- `.uce` upgrade traffic reaches `HTTP_PORT`
- nginx/Apache preserves `Upgrade` and `Connection` headers
- firewall/network policy allows localhost access to `HTTP_PORT`
### Page compiles fail
Check the compile artifact paths shown in the UCE error response and service logs. Generated files and compile output live under `BIN_DIRECTORY`. A safe last-resort recovery step is to stop UCE, move the affected generated artifact directory aside, and restart so the runtime recompiles from source.
### CLI commands fail
Check:
- `CLI_SOCKET_PATH` in `/etc/uce/settings.cfg`
- `/run/uce/cli.sock` exists
- `scripts/uce-cli --socket /run/uce/cli.sock /ping` works
- the target unit defines `CLI(Request& context)`
-110
View File
@@ -1,110 +0,0 @@
# WASM Phase 1: DValue C ABI and UCEB1
Phase 1 freezes the native DValue ABI that the future WASM core and units use
as their shared structured-value contract. The implementation is in
`src/lib/dvalue.{h,cpp}` and is available in the native runtime before any WASM
backend is enabled.
## Opaque handle
```c
typedef struct DValue uce_dvalue;
```
`uce_dvalue*` is a borrowed pointer owned by the active request/workspace. It
must not be freed by ABI callers and it must not be retained beyond that
workspace lifetime.
## Accessors
```c
uce_dvalue* uce_dv_root(void);
uce_dvalue* uce_dv_get(uce_dvalue* value, const char* key, size_t key_len);
uce_dvalue* uce_dv_find(uce_dvalue* value, const char* key, size_t key_len);
const char* uce_dv_value(uce_dvalue* value, size_t* len_out);
void uce_dv_set_value(uce_dvalue* value, const char* bytes, size_t len);
size_t uce_dv_count(uce_dvalue* value);
int uce_dv_is_list(uce_dvalue* value);
```
- `uce_dv_root()` returns the current native request's `context.call` root.
The WASM core will later map this to the decoded request context root.
- `uce_dv_get()` creates the child if absent. `uce_dv_find()` returns `NULL`
if absent.
- String inputs and outputs are length-delimited and binary-safe.
- `uce_dv_value()` returns a borrowed pointer valid until the next ABI value
call on the same thread.
- Bad `NULL` inputs return `NULL`, zero, or no-op rather than trapping.
## Iteration
```c
typedef struct uce_dv_iter { size_t position; size_t reserved[3]; } uce_dv_iter;
uce_dv_iter uce_dv_iter_begin(uce_dvalue* value);
int uce_dv_iter_next(uce_dvalue* value, uce_dv_iter* iter,
const char** key_out, size_t* key_len_out,
uce_dvalue** child_out);
```
Map iteration follows DValue's native order. List-shaped maps iterate in numeric
index order (`0`, `1`, ...), matching `DValue::each()`, `dv_values()`, and the
serializers. The reserved iterator fields are caller-opaque and must be
zero-preserved by callers that copy the iterator; they provide ABI headroom for
future non-linear keyed-map iteration without changing the struct size.
## UCEB1 wire format
UCEB1 is the membrane/cross-instance binary DValue encoding.
```
document := "UCEB" version node
version := 0x01
node := flags scalar children
flags := u8 bitset; bit0 = list-shaped map
scalar := varuint length, bytes
children := varuint count, count * (key, node)
key := varuint length, bytes
```
Varuint is unsigned LEB128. Strings are byte sequences; the codec does not
assume NUL termination and preserves embedded NUL bytes. The Phase 1 layout stores scalar values as their native string representation
plus child nodes and the list-shape flag. Floating-point values use
`max_digits10` precision so numeric scalars can round-trip through the string
form. Pointer/reference identity is intentionally not part of the wire contract;
pointer nodes encode as an empty scalar rather than leaking process addresses.
An empty non-list map has no wire distinction from an empty scalar in UCEB1 v1.
Documents that contain both scalar bytes and child nodes are reserved for future
use; the v1 decoder accepts the children and ignores the scalar.
## Codec APIs
C++/UCE-visible helpers:
```cpp
String ucb_encode(const DValue& value);
DValue ucb_decode(const String& encoded);
bool ucb_decode(const String& encoded, DValue& out, String* error_out = 0);
```
C ABI helpers:
```c
size_t uce_dv_encode(uce_dvalue* value, char* buf, size_t cap);
uce_dvalue* uce_dv_decode(const char* buf, size_t len);
const char* uce_dv_last_error(void);
```
`uce_dv_encode()` returns the required byte length even when `buf` is `NULL` or
`cap` is zero. `uce_dv_decode()` returns a thread-local decoded root, or `NULL`
with `uce_dv_last_error()` populated. The returned decoded root is valid until
the next `uce_dv_decode()` call on the same thread. Decoding rejects documents
deeper than 1024 nested nodes so malformed input cannot recurse without bound.
## Test coverage
`site/tests/core.uce` covers:
- UCEB1 round-trip for maps, nested values, lists, empty lists, and embedded NUL
scalar bytes.
- C ABI get/find/value/count/list/iteration/encode/decode behavior.
+12 -26
View File
@@ -5,9 +5,7 @@ describes the **runtime architecture as built** — the process topology, the
wasm membrane, the unified request dispatch, and the central WebSocket broker.
Native `.so` unit execution/dlopen fallback has been removed; the parser and
preprocessor remain only as the front-end that emits C++ for wasm side-module
compilation. For the historical motivation and the phased migration plan, see
[`WASM-PROPOSAL.md`](../WASM-PROPOSAL.md); this file is the steady-state
reference that proposal points at.
compilation.
The guiding principle: **request code never shares an address space or an
allocator with the runtime.** Every unit runs as a WebAssembly module inside a
@@ -55,9 +53,9 @@ gets invoked*.
| **serve_http dispatcher** (×bind) | one custom-server bind address | no — forwards to the pool | `custom_server_http_dispatcher_loop()` |
| **Proactive compiler** | nothing; pre-compiles units | no | `run_proactive_compiler()` |
The key invariant: **only workers instantiate Wasmtime and run unit code.**
**only workers instantiate Wasmtime and run unit code.**
Every connection-owning process (broker, serve_http dispatcher) forwards the
real invocation back to a worker over `/run/uce.sock` using the minimal
request invocation back to a worker over `/run/uce.sock` using the minimal
FastCGI client in [`src/lib/fcgi_forward.h`](../src/lib/fcgi_forward.h). This is
forced by Wasmtime: an `Engine`/`Store` cannot be safely re-created across
`fork()`, and the brokers fork from the parent that already touched the
@@ -163,7 +161,7 @@ otherwise (page) → serve_via_wasm(entry_unit, "render")
The `UCE_*` params are set by whichever broker forwarded the request:
- **Page render**: nginx → `/run/uce.sock` directly; no `UCE_*` flags → `render`.
- **Page render**: FastCGI nginx → `/run/uce.sock` directly; no `UCE_*` flags → `render`.
- **CLI**: the CLI socket sets `is_cli`.
- **serve_http**: the custom-server dispatcher sets `UCE_SERVE_HTTP=1` plus
`UCE_SERVE_HTTP_FUNCTION` and rewrites `SCRIPT_FILENAME` to the configured
@@ -181,21 +179,12 @@ code.
## 6. The central WebSocket broker
The broker is the architectural centerpiece. **One process owns the HTTP port
**One process owns the HTTP port
and every WebSocket connection**, so any unit's `ws_*` call can reach any or all
connections, and a unit-code crash (which happens in a worker) never drops live
connections.
### 6.1 Why central, and why it forwards
Earlier designs gave each worker its connections and captured `ws_*` output to
return inline. That was wrong: connections couldn't outlive a worker, and one
unit could only talk to connections it happened to own. The broker fixes both —
it is the sole connection registry and data broker between connected clients and
the units that handle them. It renders nothing itself; like every other
connection owner it forwards unit invocation to the worker pool.
### 6.2 Inbound: a WS frame → a worker render (non-blocking)
### 6.1 Inbound: a WS frame → a worker render (non-blocking)
`ws_broker_ws_message(request, message, opcode)` fires when a complete
(reassembled) WS message arrives on a connection. It does **not** block the
@@ -216,7 +205,7 @@ writing each queued request, then drains and discards the reply (the unit's
output comes back via the command socket, not this reply), closing the fd when
the worker closes its end.
### 6.3 Outbound: `ws_*` commands flushed back to the broker
### 6.2 Outbound: `ws_*` commands flushed back to the broker
Any unit code — not just WebSocket handlers — may call `ws_send` / `ws_send_to`
/ `ws_close`. In the workspace these **record dispatch commands** rather than
@@ -236,14 +225,14 @@ the full registry it owns: `broadcast` (by scope), `send_to` (by connection id),
`close`. If the batch carries `connection_state`, it persists that onto the
matching live connection's `websocket_state`.
### 6.4 Un-upgraded HTTP on the WS port
### 6.3 Un-upgraded HTTP on the WS port
The WS port can also receive ordinary (non-Upgrade) HTTP requests.
`ws_broker_complete()` routes by param: `UCE_WS_DISPATCH=1` → apply commands;
otherwise → `forward_request_to_worker()` — the *same* shared facility the
serve_http dispatcher uses, so there is no duplicated request-forwarding code.
### 6.5 The broker loop
### 6.4 The broker loop
`run_ws_broker()` drops the worker listeners it inherited
(`close_inherited_server_sockets`), installs permissive `on_request`/`on_data`
@@ -255,9 +244,7 @@ command socket, and loops `process(50)` + `drain_outbound()`. The design is
broker's single epoll loop** — the broker never blocks on a worker.
The parent respawns the broker if it dies (`ws_broker_alive` / `ensure_ws_broker`
in `main()`). A broker restart loses live connections (acceptable: crashes
happen in workers, so the broker stays up in practice), but no unit state is at
risk because the broker holds none.
in `main()`).
---
@@ -309,11 +296,10 @@ header free-functions are `inline`. The wasm backend exposes only declarations
- **Regression gate**: `scripts/run_cli_tests.sh --include-wasm-kill` runs the
in-runtime CLI suite (`site/tests/cli_runner.uce`) plus the site test pages.
Current baseline: **87 passed, 0 failed**.
- **WebSocket end-to-end**: a headless client performs a raw WS handshake to
`:HTTP_PORT` with path `/site/tests/websockets.ws.uce` (self-resolving
`SCRIPT_FILENAME`) and asserts the `hello-ack` frame — exercising the full
broker → worker → broker → client chain across process boundaries.
All builds, runs, and installs happen on the dev host (`k-uce` / uce-dev) over
SSH; the local checkout is edit-only.