The task interface needs the ABI switch, and a task program runs with it (0.1.1) (#1)

* The task interface needs the ABI switch, and a task program runs with it (0.1.1)

The threads feature declares requires_abi = { threads = true } instead of putting -pthread on this package's own units: the switch belongs to the artefact, and mcpp 2026.9.12.2 carries it as [target.'cfg(os = "emscripten")'.abi] threads = true in the consumer's root manifest. The README section that recorded a missing whole-graph channel is corrected, and CI asserts no task symbol without the feature, a refusal naming the feature without the switch, and with it eight symbols and a task program exiting 0 under node.

* ci: test against mcpp 2026.9.12.2, the engine that carries the ABI table

Author: Sunrisepeak

.github/workflows/ci.yml

@@ -40,7 +40,7 @@ env:
   # `em++`'s C compiler is `emcc` and not `em` (a C translation unit in the
   # conformance suite does not compile without that), and the wasm target rows
   # resolve the emsdk payload.
-  MCPP_VERSION: 2026.9.11.4
+  MCPP_VERSION: 2026.9.12.2
   XLINGS_VERSION: v2026.8.17.2
   XLINGS_NON_INTERACTIVE: '1'
 
@@ -167,8 +167,9 @@ jobs:
       # The set is named rather than `full`: clause 6.1 says an implementation
       # provides an interface in whole or not at all, and a group this platform
       # does not have is an absence rather than a deviation. `task` is not in
-      # the set -- see the task-gate step below for why it needs a link this
-      # suite cannot currently be given.
+      # the set: the suite's consumer manifest is generated by the
+      # specification's runner, which does not state the ABI switch threads
+      # need. The task-gate step below builds and runs a task program with it.
       - name: Every interface this platform has, examined
         run: |
           OPENKAL_CONFORMANCE_RUNNER="$NODE_BIN" \
@@ -288,20 +289,24 @@ jobs:
           done
           test "$fail" -eq 0
 
-      # THE TASK GATE, IN BOTH DIRECTIONS.
+      # THE TASK GATE, IN THREE DIRECTIONS.
       #
       # `openkal.task` needs `-pthread`, which selects a different C library
-      # build and a different memory model -- a property of the whole link. So
-      # the interface is carried by a feature: without it the translation unit
-      # is empty and the eight symbols do not exist, which is the same
-      # treatment the three absent interfaces get and for the same reason.
+      # build and a different memory model -- a property of the whole link, and
+      # therefore of the consumer's root manifest (`[target.<selector>.abi]
+      # threads = true`, mcpp 2026.9.12.2). The `threads` feature states that it
+      # needs the switch. Asserted:
       #
-      # Both directions are asserted, because a gate is only a gate if the
-      # other side of it differs. The conformance suite cannot yet be built
-      # with this switch -- it would have to reach the specification package's
-      # modules too, and mcpp has no whole-graph flag channel -- so what is
-      # asserted here is the gating, and the README records the rest.
-      - name: The task interface exists in a -pthread link and in no other
+      #   1. without the feature and the switch, no task symbol exists;
+      #   2. the feature without the switch is refused before anything compiles,
+      #      naming the feature;
+      #   3. with both, the eight symbols exist and a program that starts a task
+      #      and joins it exits 0 under node.
+      #
+      # The consumer names `openkal` by path, as the implementation's manifest
+      # now does (the step above): mcpp refuses one package reached both by path
+      # and by version.
+      - name: The task interface needs the ABI switch, is refused without it, and runs with it
         run: |
           set -euo pipefail
           rm -rf target
@@ -311,14 +316,55 @@ jobs:
           test "$n" -eq 0 || { echo "::error::$n kal_task symbols without the feature"; exit 1; }
           echo "  ok  no task symbol without the feature"
 
-          rm -rf target
-          mcpp build --target wasm32-emscripten --features threads
-          objs=$(find target -path '*/obj/*' -name '*.o' ! -name '*.m.o' | tr '\n' ' ')
+          work="$RUNNER_TEMP/tasks"
+          impl="$(pwd)"
+          rm -rf "$work"; mkdir -p "$work/src"
+          cat > "$work/src/main.cpp" << 'CPP'
+          import openkal.task;
+          static void entry(void* arg) { *static_cast<int*>(arg) = 1; }
+          int main() {
+            // KAL_TASK_PROP_PARALLEL is bit 1 (openkal/task.h); a macro does not
+            // cross a module.
+            if ((kal_task_props() & 2u) == 0) return 2;
+            int ran = 0;
+            kal_task t{};
+            if (kal_task_start(&entry, &ran, &t) != 0) return 3;
+            if (kal_task_join(t) != 0) return 4;
+            return ran == 1 ? 0 : 5;
+          }
+          CPP
+          manifest() {
+            cat > "$work/mcpp.toml" << TOML
+          [package]
+          name = "tasks"
+          version = "0.1.0"
+
+          [dependencies]
+          openkal = { path = "$impl/.spec" }
+          openkal-emscripten = { path = "$impl", features = ["threads"] }
+          $1
+          TOML
+          }
+
+          manifest ""
+          if ( cd "$work" && mcpp build --target wasm32-emscripten ) > refused.log 2>&1; then
+            cat refused.log; echo "::error::the threads feature built without the ABI switch"; exit 1
+          fi
+          grep -q "requires the artefact's ABI to have threads (feature \`threads\`)" refused.log \
+            || { cat refused.log; echo "::error::the refusal does not name the feature"; exit 1; }
+          echo "  ok  the feature without the switch is refused, naming the feature"
+
+          manifest "$(printf '\n[target.%s.abi]\nthreads = true\n' "'cfg(os = \"emscripten\")'")"
+          ( cd "$work" && mcpp build --target wasm32-emscripten )
+          objs=$(find "$work/target" -path '*/obj/*' -name '*.o' ! -name '*.m.o' | tr '\n' ' ')
           n=$("$NM" --defined-only $objs 2>/dev/null \
               | awk '$2=="T"||$2=="W"{print $3}' | sed 's/^_//' \
               | grep -c '^kal_task' || true)
-          test "$n" -eq 8 || { echo "::error::expected 8 kal_task symbols with the feature, found $n"; exit 1; }
-          echo "  ok  eight task symbols with the feature"
+          test "$n" -eq 8 || { echo "::error::expected 8 kal_task symbols with the switch, found $n"; exit 1; }
+          echo "  ok  eight task symbols with the switch"
+          ( cd "$work" && timeout 300 mcpp run --target wasm32-emscripten ) \
+            || { echo "::error::the task program did not exit 0 under node"; exit 1; }
+          echo "  ok  a task starts, runs and joins under node"
 
       - name: Restore the manifest
         if: always()

README.md

@@ -61,30 +61,48 @@ So the interface is carried by a feature:
 
 ```toml
 [dependencies]
-openkal-emscripten = { version = "0.1.0", features = ["threads"] }
+openkal-emscripten = { version = "0.1.1", features = ["threads"] }
 ```
 
 Without it, `src/threads/task.cpp` compiles to nothing, the eight `kal_task_*`
 symbols do not exist, and `kal_interfaces()` does not claim the interface --
 the same treatment the three absent interfaces get, for the same reason.
 
-**This feature is not yet usable end to end, and the limitation is the build
-tool's rather than this package's.** `-pthread` changes the module
-configuration of *every* translation unit in the link, including the
-specification package's, and mcpp has no channel for a flag that applies to a
-whole dependency graph: a feature contributes sources, defines and per-glob
-compile flags, and `[build] ldflags` reaches the root package only. Measured
-2026-09-11 with the feature active and `-pthread` on the consumer:
+**The switch belongs to the artefact, and the root manifest states it.**
+`-pthread` changes the module configuration of *every* translation unit in the
+link, including the specification package's, so it cannot be a flag of this
+package's own units. With mcpp 2026.9.12.2 or later, the consumer writes
+
+```toml
+[target.'cfg(os = "emscripten")'.abi]
+threads = true
+```
+
+and the switch reaches the standard library module, every translation unit of
+every package, and the link. The `threads` feature states that it needs the
+switch (`requires_abi = { threads = true }`), so a consumer that activates the
+feature without the table is refused before anything compiles, naming the
+feature and the table. Measured 2026-09-12 with emsdk 6.0.9 and a development
+build of mcpp 2026.9.12.2: with the table, a program that starts a task and
+joins it exits 0 under node, and without it the build stops at
+
+```
+error: `openkal-emscripten` requires the artefact's ABI to have threads (feature `threads`), and this build does not state it.
+```
+
+An earlier version of this section, measured 2026-09-11, concluded that mcpp had
+no channel for a flag that applies to a whole dependency graph. What had been
+measured was `-pthread` in the consumer's per-package `cxxflags`:
 
 ```
 error: POSIX thread support was disabled in precompiled file
        '.../pcm.cache/openkal.types.pcm' but is currently enabled
 ```
 
-which is the specification package's module, compiled without the switch. The
-interface word is already correct for a link that manages it; closing the gap
-is an engine change of the same shape as the existing whole-graph runtime
-flags, and it is recorded rather than worked around.
+That channel does not reach the specification package's module. A graph-wide
+channel did exist (`[build] dialect_cxxflags`); what was missing was a way to
+scope it to one target and to let this feature state its requirement, which the
+typed table provides.
 
 ## Conformance
 
@@ -128,7 +146,7 @@ implementation in its source:
 openkal = "0.12.0"
 
 [target.'cfg(os = "emscripten")'.dependencies]
-openkal-emscripten = "0.1.0"
+openkal-emscripten = "0.1.1"
 ```
 
 ## Licence

mcpp.toml

@@ -1,7 +1,7 @@
 [package]
 namespace   = "mcpplibs"
 name        = "openkal-emscripten"
-version     = "0.1.0"
+version     = "0.1.1"
 description = "An implementation of openkal for Emscripten, written ABOVE a C library rather than beneath one, because on this platform there is no kernel to issue calls to."
 license     = "Apache-2.0"
 
@@ -59,29 +59,24 @@ default = []
 # capability is not a property of the machine this runs on, and it is not
 # discoverable at run time by anything this package can ask.
 #
-# Declaring it as a feature puts the decision where the rest of the link is
-# decided, and `kal_task_props` reports the consequence: with this feature the
-# word carries KAL_TASK_PROP_PARALLEL, and without it the interface is still
-# provided and `kal_task_start` reports kal_err_not_supported -- which is
-# clause 6.2's third time, a capability word read before use, rather than the
-# shape the specification forbids.
+# THE SWITCH IS THE ARTEFACT'S, SO THE ROOT MANIFEST STATES IT. `-pthread`
+# changes the module configuration of every translation unit in the link, the
+# specification package's included, so it cannot be a flag of this package's
+# own units. mcpp 2026.9.12.2 carries it as a typed member of the root's target
+# table,
 #
-# `-pthread` ON THIS PACKAGE'S OWN UNITS, AND NOT A DEFINE OF ITS OWN. Passing
-# the switch is what makes Emscripten define `__EMSCRIPTEN_PTHREADS__`, so the
-# code keys on that rather than on a name this manifest invented: one
-# mechanism, and it is the compiler's own answer to "was this compiled with
-# threads" instead of a second word that could disagree with it.
+#   [target.'cfg(os = "emscripten")'.abi]
+#   threads = true
 #
-# THE LINK HALF IS THE CONSUMER'S, AND THAT IS A LIMITATION WORTH NAMING. A
-# feature contributes build INPUTS -- sources, defines, flags -- and mcpp's
-# feature axis has no `ldflags`, so this cannot put `-pthread` on the link
-# line. A project that activates this feature must also write
+# which reaches the standard library module, every translation unit and the
+# link. This feature states that it needs the switch: a build that activates it
+# without the table is refused before anything compiles, naming the feature and
+# the table.
 #
-#   [build]
-#   ldflags = ["-pthread"]
-#
-# and a project that forgets fails at LINK naming the pthread symbols, which
-# is a loud failure rather than a silent single-threaded program.
-threads = { flags = [
-  { glob = "src/**", cxxflags = ["-pthread"] },
-] }
+# `kal_task_props` reports the consequence: with the switch the word carries
+# KAL_TASK_PROP_PARALLEL, and without it `kal_task_start` reports
+# kal_err_not_supported -- clause 6.2's third time, a capability word read
+# before use. The code keys on `__EMSCRIPTEN_PTHREADS__`, which the compiler
+# defines when `-pthread` is passed: one mechanism, and it is the compiler's own
+# answer to "was this compiled with threads".
+threads = { requires_abi = { threads = true } }