Doug/laughing davinci stc796

.github/workflows/arduino-build.yml

@@ -13,13 +13,19 @@ on:
     paths:
       - "src/**"
       - "include/**"
+      - "vendor/**"
+      - "test/core_setup/**"
+      - "library.properties"
       - "examples/**"
       - ".github/workflows/arduino-build.yml"
   pull_request:
     branches: [main, master]
     paths:
       - "src/**"
       - "include/**"
+      - "vendor/**"
+      - "test/core_setup/**"
+      - "library.properties"
       - "examples/**"
   workflow_dispatch:
 
@@ -105,8 +111,26 @@ jobs:
       # xtensa default and the link fails with "libstdc++.a compiled for a big endian system and
       # target is little endian". (The PlatformIO build uses the older, statically-little-endian
       # xtensa-esp32-elf toolchain, so wrapping is safe there.) Each example does a full compile.
+      # Arduino compiles a library's src/ and nothing else, and src/ is the only include path it
+      # gives the library. The tree keeps the entry header in include/, the vendor axis in vendor/
+      # and the HAL arms in test/core_setup/, so the library is staged in Arduino's layout: those
+      # three under src/, where every #include spelling in the tree resolves unchanged. It is the
+      # same set of sources the ESP-IDF component globs.
+      - name: Stage the library in Arduino layout
+        run: |
+          STAGE="$RUNNER_TEMP/ProtoCore"
+          mkdir -p "$STAGE/src/test"
+          cp library.properties "$STAGE/"
+          cp -r src/. "$STAGE/src/"
+          cp include/*.h "$STAGE/src/"
+          cp -r vendor "$STAGE/src/vendor"
+          cp -r test/core_setup "$STAGE/src/test/core_setup"
+          # Not firmware: idemIP is a host model of the IP stack, boot/ the bare-metal simulator's
+          # startup (its own gmtime_r, memcpy and vector table).
+          rm -rf "$STAGE/src/test/core_setup/idemIP" "$STAGE/src/test/core_setup/boot"
+          echo "PROTOCORE_LIB=$STAGE" >> "$GITHUB_ENV"
       - name: Compile ${{ matrix.example.name }} (Arduino IDE, esp32 3.x)
         # No --build-property / build_flags: the example's build_opt.h supplies the feature
         # flags, so this is byte-for-byte how the Arduino IDE builds it.
         run: |
-          arduino-cli compile --fqbn esp32:esp32:esp32 --library . "examples/${{ matrix.example.name }}"
+          arduino-cli compile --fqbn esp32:esp32:esp32 --library "$PROTOCORE_LIB" "examples/${{ matrix.example.name }}"

.github/workflows/esp32-build.yml

@@ -13,6 +13,9 @@ on:
     paths:
       - "src/**"
       - "include/**"
+      - "vendor/**"
+      - "test/core_setup/**"
+      - "library.json"
       - "examples/**"
       - "platformio.ini"
       - "tools/ci_tooling/generate/example_footprints.py"
@@ -22,6 +25,9 @@ on:
     paths:
       - "src/**"
       - "include/**"
+      - "vendor/**"
+      - "test/core_setup/**"
+      - "library.json"
       - "examples/**"
   workflow_dispatch:
   schedule:

CMakeLists.txt

@@ -38,6 +38,11 @@ option(PROTOCORE_WITH_ARDUINO "Build against arduino-esp32 (Arduino-as-component
 # (cmake -P) to extract REQUIRES, and CONFIGURE_DEPENDS is illegal there. The glob is therefore
 # evaluated at configure time - run `idf.py reconfigure` after adding or removing a source file.
 file(GLOB_RECURSE PROTOCORE_SRCS "src/*.c" "src/*.cpp" "test/core_setup/*.c" "test/core_setup/*.cpp")
+# Two arms under test/core_setup/ are not firmware and carry no guard: idemIP is a host model of the
+# IP stack (its headers resolve from test/core_setup/, which is not an include dir here), and boot/
+# is the bare-metal simulator's startup - its own gmtime_r, memcpy and vector table, which collide
+# with newlib and the IDF startup.
+list(FILTER PROTOCORE_SRCS EXCLUDE REGEX "/test/core_setup/(idemIP|boot)/")
 
 set(PROTOCORE_REQUIRES
     lwip

docs/ARCHITECTURE.md

@@ -145,7 +145,7 @@
 `http_parser_set_stream_hooks(begin, data, abort)` are global singletons
 (last-registered-wins, so OTA / upload / WebDAV streaming are still mutually
 exclusive per build). All three now take `HttpReq*`. A sink can keep
 per-connection state: WebDAV holds per-slot PUT state (`s_davput.put[MAX_CONNS]` in
 `src/server/io/webdav_handler/webdav_handler.c`) and each connection streams to its own file. This
 fixed the concurrent-PUT clobber (docs/BUGS.md) - HW: 4 parallel PUTs with distinct
 payloads, all byte-exact.
@@ -157,7 +157,7 @@
 | Socket TX            | transport `ConnPool`        | DONE                                               |
 | RX receive window    | transport                   | DONE (`ConnPool.ack_consumed`, ack-on-consume)     |
 | RX ring read/drain   | transport (read API)        | DONE (`ConnPool.read*`; consumers off the ring)    |
 | Streaming sink state | per-slot, slot-aware        | DONE (`s_davput.put[MAX_CONNS]`, slot-aware hooks) |
 | Event routing        | session (owner queue)       | DONE                                               |
 | Working memory       | mmgr `plain` (per slot)     | DONE (worker resets it per dispatch)               |
 | Key material         | mmgr `secure` (per slot)    | DONE (disjoint region; reclaiming wipes)           |
@@ -174,7 +174,7 @@
    stays the only place that reopens the window (per loop), so draining and ACKing
    each have exactly one owner. HW: 10/10 50 KB byte-exact, backpressure 0.
 3. **DONE - slot-aware streaming hooks** - `HttpStreamDataCb(HttpReq*, ...)` +
    per-slot WebDAV PUT state `s_davput.put[MAX_CONNS]`; fixed the concurrent-PUT bug
    (HW: 4 parallel PUTs, distinct payloads, all byte-exact).
 4. **DONE - outbound clients** - all clients (http_client / mqtt / ws_client) share
    `TcpClient`, the unified client transport; brought to the same ack-on-consume
@@ -219,7 +219,7 @@
 HTTP/2, and HTTP/3), and the response funnels back through the symmetric **TX seam** - a
 per-connection `protocore_resp_sink_fn` function pointer (`http_resp_sink[slot]`, session.h) that
 HTTP/2 installs at ALPN and HTTP/3 at dispatch. `send_text()` / `send_empty()` call
-`http_resp_sink[slot](...)`
+`http_resp_sink[slot](slot, code, content_type, payload, body_len)`
 when it is set (h2 frames the reply as HEADERS+DATA on the stream; h3 as an HTTP/3 response on
 its QUIC stream) and otherwise build the HTTP/1.1 message - so the response methods name no
 protocol. It is the RX `ProtoHandler` seam's TX counterpart: request decode and response encode

docs/HARDWARE_HOOKUP.md

@@ -907,8 +907,8 @@ clocking stalls that read instead of the main loop.
 
 - **LD2410 mmWave radar, `PROTOCORE_ENABLE_LD2410`** - a 24 GHz presence sensor that sees a
   still person (breathing), in the dark, through thin walls - over a UART at **256000
-  baud**. Cross the data wires (module TX -> ESP32 RX). `protocore_ld2410_poll()` decodes each
-  frame; `protocore_ld2410_present()` / `protocore_ld2410_distance_cm()` act on it. Example Ld2410.
+  baud**. Cross the data wires (module TX -> ESP32 RX). `Ld2410.poll()` decodes each
+  frame; `Ld2410.present()` / `Ld2410.distance_cm()` act on it. Example Ld2410.
 - **MPR121 capacitive touch, `PROTOCORE_ENABLE_MPR121`** - turns 12 wires or pads into
   touch buttons (I2C 0x5A). `protocore_mpr121_read_touched()` returns a 12-bit mask. Example
   Mpr121.