pcbjam/tests/apps/standalone/sched-context/sched_context_test.cpp

1013 lines
35 KiB
C++
Raw Normal View History

design-b D1b/c: context harness + the bounded-and-measured memory gate Doc 20 §6 D1, second slice: the dedicated test app D1 calls for, plus the doc 20 §7 risk-1 gate. tests/apps/standalone/sched-context/ drives the primitives with no wx linked (a failure can only be the contexts layer) and assertions ON — unlike the races harness, an emscripten "cannot start an async operation" state here IS the bug, not a tolerated state. Nine scenarios, each pinning one invariant; the load-bearing ones: - parked_does_not_block: a parked context does not stop others running. This is doc 19's freeze made unrepresentable — there a parked activity held a global interlock and the UI died; here "parked" simply means "not runnable", and three workers run past it before it resumes. - one_transition_in_flight: a context calling drain() gets a no-op, with a second ready context queued so a buggy nested drain would actually run something and be caught. - async_wake: a real macrotask hop (setTimeout → mark_ready → drain), the shape every production bridge has. - deep_park_sizing: parks 64 live frames deep to measure what a park actually costs. The memory gate asserts the ceiling (peak live contexts, peak bytes), that nothing leaked (live=0, bytes=0, created==finished), that nothing was left mid-transition, and that refusals occurred (zero would mean the illegal-operation scenarios stopped provoking). Measured here: ~34 B/frame, 2200 B for a 64-frame park, 1 MB peak for 4 concurrent contexts. Recorded with the caveat that the harness's frames carry three locals each, so this is a FLOOR — real bridges save far more per frame (libcontext runs 512 K after a 64 K buffer silently overflowed). The apparatus and its units are validated; the sizing DECISION needs deep-park numbers from real bridges at D3/D4. playwright.config: asyncify-firefox now matches every spec in ./asyncify instead of only asyncify-races* — a new harness there is covered by construction rather than by remembering to widen the pattern. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
2026-08-05 21:41:24 +02:00
/*
* Scheduler-context harness Design B D1 gate
* (pcbjam docs/features/async/20-design-b-core-plan.md §6 D1).
*
* Exercises wx/wasm/private/sched_context.h create / yield_park /
* mark_ready / drain against the invariants that make a resume something the registry
design-b D1b/c: context harness + the bounded-and-measured memory gate Doc 20 §6 D1, second slice: the dedicated test app D1 calls for, plus the doc 20 §7 risk-1 gate. tests/apps/standalone/sched-context/ drives the primitives with no wx linked (a failure can only be the contexts layer) and assertions ON — unlike the races harness, an emscripten "cannot start an async operation" state here IS the bug, not a tolerated state. Nine scenarios, each pinning one invariant; the load-bearing ones: - parked_does_not_block: a parked context does not stop others running. This is doc 19's freeze made unrepresentable — there a parked activity held a global interlock and the UI died; here "parked" simply means "not runnable", and three workers run past it before it resumes. - one_transition_in_flight: a context calling drain() gets a no-op, with a second ready context queued so a buggy nested drain would actually run something and be caught. - async_wake: a real macrotask hop (setTimeout → mark_ready → drain), the shape every production bridge has. - deep_park_sizing: parks 64 live frames deep to measure what a park actually costs. The memory gate asserts the ceiling (peak live contexts, peak bytes), that nothing leaked (live=0, bytes=0, created==finished), that nothing was left mid-transition, and that refusals occurred (zero would mean the illegal-operation scenarios stopped provoking). Measured here: ~34 B/frame, 2200 B for a 64-frame park, 1 MB peak for 4 concurrent contexts. Recorded with the caveat that the harness's frames carry three locals each, so this is a FLOOR — real bridges save far more per frame (libcontext runs 512 K after a 64 K buffer silently overflowed). The apparatus and its units are validated; the sizing DECISION needs deep-park numbers from real bridges at D3/D4. playwright.config: asyncify-firefox now matches every spec in ./asyncify instead of only asyncify-races* — a new harness there is covered by construction rather than by remembering to widen the pattern. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
2026-08-05 21:41:24 +02:00
* KNOWS rather than something a guard guesses (doc 19's disease):
*
* 1. a context runs only when the scheduler drains it;
* 2. a park yields the whole context; its result arrives on resume;
* 3. the registry is truth (status transitions, illegal ops refused);
* 4. nothing parks "in place" yield_park off a context is refused;
* 5. a PARKED context does not block others (doc 20 §5: "parked ones do
* not count") — the property that makes doc 19's freeze unrepresentable;
* 6. FIFO readiness, no starvation;
* 7. at most one transition in flight (drain is not re-entrant);
* 8. a real async wake (JS timeout mark_ready drain) resumes correctly;
* 9. memory is bounded and MEASURED (peak bytes, asyncify high-water)
* the doc 20 risk-1 gate.
*
* Output contract (parsed by tests/asyncify/sched-context.spec.ts):
* [SCHED_CTX] PASS <name> per scenario
* [SCHED_CTX] FAIL <name>: <detail>
* [SCHED_CTX] STATS <json> the memory/registry snapshot
* [SCHED_CTX] SUMMARY total=N passed=N failed=N
*/
#include "wx/wasm/private/sched_context.h"
design-b D1b/c: context harness + the bounded-and-measured memory gate Doc 20 §6 D1, second slice: the dedicated test app D1 calls for, plus the doc 20 §7 risk-1 gate. tests/apps/standalone/sched-context/ drives the primitives with no wx linked (a failure can only be the contexts layer) and assertions ON — unlike the races harness, an emscripten "cannot start an async operation" state here IS the bug, not a tolerated state. Nine scenarios, each pinning one invariant; the load-bearing ones: - parked_does_not_block: a parked context does not stop others running. This is doc 19's freeze made unrepresentable — there a parked activity held a global interlock and the UI died; here "parked" simply means "not runnable", and three workers run past it before it resumes. - one_transition_in_flight: a context calling drain() gets a no-op, with a second ready context queued so a buggy nested drain would actually run something and be caught. - async_wake: a real macrotask hop (setTimeout → mark_ready → drain), the shape every production bridge has. - deep_park_sizing: parks 64 live frames deep to measure what a park actually costs. The memory gate asserts the ceiling (peak live contexts, peak bytes), that nothing leaked (live=0, bytes=0, created==finished), that nothing was left mid-transition, and that refusals occurred (zero would mean the illegal-operation scenarios stopped provoking). Measured here: ~34 B/frame, 2200 B for a 64-frame park, 1 MB peak for 4 concurrent contexts. Recorded with the caveat that the harness's frames carry three locals each, so this is a FLOOR — real bridges save far more per frame (libcontext runs 512 K after a 64 K buffer silently overflowed). The apparatus and its units are validated; the sizing DECISION needs deep-park numbers from real bridges at D3/D4. playwright.config: asyncify-firefox now matches every spec in ./asyncify instead of only asyncify-races* — a new harness there is covered by construction rather than by remembering to widen the pattern. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
2026-08-05 21:41:24 +02:00
#include <emscripten/emscripten.h>
#include <cstdio>
#include <string>
#include <vector>
using namespace pcbjam_sched;
namespace
{
int g_total = 0;
int g_passed = 0;
void log_line( const std::string& aLine )
{
std::printf( "%s\n", aLine.c_str() );
std::fflush( stdout );
}
void report( const char* aName, bool aOk, const std::string& aDetail = "" )
{
++g_total;
if( aOk )
{
++g_passed;
log_line( std::string( "[SCHED_CTX] PASS " ) + aName );
}
else
{
log_line( std::string( "[SCHED_CTX] FAIL " ) + aName + ": " + aDetail );
}
}
// ---------------------------------------------------------------------------
// scenario 1: a created context runs its body — but only on drain()
// ---------------------------------------------------------------------------
int g_ran = 0;
void body_sets_flag( void* )
{
g_ran = 1;
}
void Scenario_CreateRunsOnDrain()
{
g_ran = 0;
const ContextId id = create( body_sets_flag, nullptr, "runs-on-drain" );
if( !id )
return report( "create_runs_on_drain", false, "create returned 0" );
// Creation alone must NOT run it: the scheduler decides when.
if( g_ran != 0 )
return report( "create_runs_on_drain", false, "body ran before drain()" );
const ContextId ran = drain();
if( ran != id )
return report( "create_runs_on_drain", false, "drain() returned a different id" );
if( g_ran != 1 )
return report( "create_runs_on_drain", false, "body did not run" );
if( status_of( id ) != Status::Finished )
return report( "create_runs_on_drain", false,
std::string( "status after return: " ) + status_name( status_of( id ) ) );
destroy( id );
report( "create_runs_on_drain", true );
}
// ---------------------------------------------------------------------------
// scenario 2: park yields the whole context; the result arrives on resume
// ---------------------------------------------------------------------------
int g_park_result = 0;
int g_park_stage = 0;
void body_parks_once( void* )
{
g_park_stage = 1;
g_park_result = yield_park( "test-park" );
g_park_stage = 2;
}
void Scenario_ParkAndResume()
{
g_park_stage = 0;
g_park_result = 0;
const ContextId id = create( body_parks_once, nullptr, "park-once" );
drain(); // runs until the park
if( g_park_stage != 1 )
return report( "park_and_resume", false, "body did not reach the park" );
if( status_of( id ) != Status::Parked )
return report( "park_and_resume", false,
std::string( "status: " ) + status_name( status_of( id ) ) );
// Draining again must do NOTHING: a parked context is not runnable until
// it is marked ready. (If this resumed it, "parked" would be a guess.)
if( drain() != 0 )
return report( "park_and_resume", false, "drain() resumed a parked context" );
if( !mark_ready( id, 4242 ) )
return report( "park_and_resume", false, "mark_ready failed" );
if( status_of( id ) != Status::Ready )
return report( "park_and_resume", false, "status is not ready after mark_ready" );
drain();
if( g_park_stage != 2 )
return report( "park_and_resume", false, "body did not resume" );
if( g_park_result != 4242 )
return report( "park_and_resume", false,
"wrong result: " + std::to_string( g_park_result ) );
destroy( id );
report( "park_and_resume", true );
}
// ---------------------------------------------------------------------------
// scenario 3: nothing parks in place — yield_park off a context is refused
// ---------------------------------------------------------------------------
void Scenario_NoParkInPlace()
{
// Called from the scheduler stack (this battery runs there).
const int rc = yield_park( "illegal" );
if( rc != -1 )
return report( "no_park_in_place", false, "yield_park did not refuse" );
if( current() != 0 )
return report( "no_park_in_place", false, "current() is not the scheduler" );
report( "no_park_in_place", true );
}
// ---------------------------------------------------------------------------
// scenario 4: a parked context does NOT block other contexts
// This is doc 19's freeze, made unrepresentable: there, a parked activity
// held a global interlock and everything else stopped. Here, a parked
// context is simply not runnable while every other context keeps running.
// ---------------------------------------------------------------------------
int g_blocker_stage = 0;
int g_worker_runs = 0;
void body_blocker( void* )
{
g_blocker_stage = 1;
yield_park( "long-park" );
g_blocker_stage = 2;
}
void body_worker( void* )
{
++g_worker_runs;
}
void Scenario_ParkedDoesNotBlock()
{
g_blocker_stage = 0;
g_worker_runs = 0;
const ContextId blocker = create( body_blocker, nullptr, "blocker" );
drain(); // blocker parks, and stays parked for the rest of this scenario
if( status_of( blocker ) != Status::Parked )
return report( "parked_does_not_block", false, "blocker is not parked" );
// Three unrelated contexts must run to completion while it is parked.
std::vector<ContextId> workers;
for( int i = 0; i < 3; ++i )
workers.push_back( create( body_worker, nullptr, "worker" ) );
for( int i = 0; i < 3; ++i )
drain();
if( g_worker_runs != 3 )
return report( "parked_does_not_block", false,
"workers run: " + std::to_string( g_worker_runs ) + " (expected 3)" );
if( g_blocker_stage != 1 )
return report( "parked_does_not_block", false, "blocker resumed unexpectedly" );
// And the blocker still resumes correctly afterwards.
mark_ready( blocker, 0 );
drain();
if( g_blocker_stage != 2 )
return report( "parked_does_not_block", false, "blocker did not resume at the end" );
for( ContextId w : workers )
destroy( w );
destroy( blocker );
report( "parked_does_not_block", true );
}
// ---------------------------------------------------------------------------
// scenario 5: FIFO readiness — no starvation
// ---------------------------------------------------------------------------
std::vector<int> g_order;
void body_records( void* aArg )
{
g_order.push_back( static_cast<int>( reinterpret_cast<intptr_t>( aArg ) ) );
}
void Scenario_FifoOrder()
{
g_order.clear();
std::vector<ContextId> ids;
for( int i = 1; i <= 4; ++i )
ids.push_back( create( body_records, reinterpret_cast<void*>( (intptr_t) i ), "fifo" ) );
for( size_t i = 0; i < ids.size(); ++i )
drain();
if( g_order.size() != 4 )
return report( "fifo_order", false, "ran " + std::to_string( g_order.size() ) );
for( int i = 0; i < 4; ++i )
{
if( g_order[i] != i + 1 )
return report( "fifo_order", false, "out of order at " + std::to_string( i ) );
}
for( ContextId id : ids )
destroy( id );
report( "fifo_order", true );
}
// ---------------------------------------------------------------------------
// scenario 6: one transition in flight — drain() is not re-entrant
// ---------------------------------------------------------------------------
int g_reentrant_drain_rc = -99;
bool g_saw_transition = false;
void body_calls_drain( void* )
{
g_saw_transition = transition_in_flight();
// A context calling drain() would turn the star into a cycle; the guard
// must make it a no-op rather than a nested swap.
g_reentrant_drain_rc = static_cast<int>( drain() );
}
void Scenario_OneTransitionInFlight()
{
g_reentrant_drain_rc = -99;
g_saw_transition = false;
// A second, ready context exists — so a buggy re-entrant drain would have
// something to run and the failure would be observable, not vacuous.
const ContextId victim = create( body_worker, nullptr, "victim" );
const ContextId id = create( body_calls_drain, nullptr, "reentrant" );
drain(); // runs victim (FIFO)
drain(); // runs the re-entrant body
if( !g_saw_transition )
return report( "one_transition_in_flight", false,
"transition_in_flight() was false inside a running context" );
if( g_reentrant_drain_rc != 0 )
return report( "one_transition_in_flight", false,
"nested drain() ran a context: " + std::to_string( g_reentrant_drain_rc ) );
destroy( victim );
destroy( id );
report( "one_transition_in_flight", true );
}
// ---------------------------------------------------------------------------
// scenario 7: registry refusals — illegal operations are recorded, not obeyed
// ---------------------------------------------------------------------------
void Scenario_RegistryRefusals()
{
if( mark_ready( 999999, 0 ) )
return report( "registry_refusals", false, "mark_ready on an unknown id succeeded" );
const ContextId id = create( body_parks_once, nullptr, "refusals" );
drain(); // parks
// Destroying a parked context would strand its stack.
if( destroy( id ) )
return report( "registry_refusals", false, "destroy() freed a parked context" );
// Marking a ready context ready again must not double-queue it.
mark_ready( id, 1 );
if( mark_ready( id, 1 ) )
return report( "registry_refusals", false, "mark_ready twice succeeded" );
drain();
destroy( id );
report( "registry_refusals", true );
}
// ---------------------------------------------------------------------------
// scenario 8: DEEP park — how does asyncify buffer use scale with depth?
//
// This is the doc 20 risk-1 measurement that matters. A shallow park saves
// almost nothing (~100 B), which would tempt a tiny buffer; libcontext's
// 512 K exists precisely because a DEEP park (a collab apply suspending
// inside commit.Push → connectivity → font work) overflowed 64 K and
// silently corrupted the rewind state. So park at a known depth with live
// locals and report the cost, giving a per-frame number a future buffer
// size can be derived FROM rather than inherited.
// ---------------------------------------------------------------------------
constexpr int DEEP_PARK_FRAMES = 64;
int g_deep_result = 0;
int g_deep_reached = 0;
// noinline + volatile locals: the point is to keep frames genuinely live
// across the park so the unwind has something real to save.
__attribute__( ( noinline ) ) int deep_recurse( int aDepth )
{
volatile int a = aDepth * 3;
volatile int b = aDepth ^ 0x5a5a;
volatile double c = aDepth * 1.5;
if( aDepth <= 0 )
{
g_deep_reached = 1;
return yield_park( "deep-park" );
}
const int inner = deep_recurse( aDepth - 1 );
return inner + static_cast<int>( a ) - static_cast<int>( a ) + static_cast<int>( b )
- static_cast<int>( b ) + static_cast<int>( c ) - static_cast<int>( c );
}
void body_deep_park( void* )
{
g_deep_result = deep_recurse( DEEP_PARK_FRAMES );
}
void Scenario_DeepPark()
{
g_deep_result = 0;
g_deep_reached = 0;
const ContextId id = create( body_deep_park, nullptr, "deep-park" );
drain();
if( !g_deep_reached )
return report( "deep_park_sizing", false, "never reached the deep park" );
if( status_of( id ) != Status::Parked )
return report( "deep_park_sizing", false, "not parked at depth" );
mark_ready( id, 31337 );
drain();
if( g_deep_result != 31337 )
return report( "deep_park_sizing", false,
"wrong result through " + std::to_string( DEEP_PARK_FRAMES )
+ " frames: " + std::to_string( g_deep_result ) );
if( status_of( id ) != Status::Finished )
return report( "deep_park_sizing", false, "did not finish after the deep resume" );
// The sizing evidence: bytes saved for a known frame count.
log_line( "[SCHED_CTX] DEEPPARK frames=" + std::to_string( DEEP_PARK_FRAMES )
+ " registry=" + registry_json() );
destroy( id );
report( "deep_park_sizing", true );
}
// ---------------------------------------------------------------------------
// scenario 9: a foreign stack may NOT yield someone else's context
//
// The D3 blocker, pinned. A KiCad tool coroutine is a libcontext fiber
// swapped in ON TOP of whatever dispatched it. If that fiber calls a wait,
// the registry still says "the dispatch context is running" while the live
// stack belongs to the tool fiber — so a naive yield would save the TOOL
// fiber's stack into the DISPATCH context's fiber struct and hand it to the
// next resume. Silent, total, and undetectable after the fact.
//
// Here a raw emscripten fiber (the same thing libcontext builds on) is
// swapped in above a context and calls yield_park. It must be REFUSED, and
// the context must remain usable afterwards.
// ---------------------------------------------------------------------------
emscripten_fiber_t g_foreign_fiber;
emscripten_fiber_t g_host_return;
alignas( 16 ) char g_foreign_stack[64 * 1024];
alignas( 16 ) char g_foreign_asyncify[16 * 1024];
// The host side of the swap needs its OWN asyncify buffer: emscripten_fiber_swap
// unwinds the outgoing fiber into it. A zero-initialised emscripten_fiber_t has
// a null asyncify stack and traps on the first swap.
alignas( 16 ) char g_host_asyncify[16 * 1024];
int g_foreign_yield_rc = -99;
int g_foreign_ran = 0;
ContextId g_host_ctx = 0;
// PROBE_YIELD=0 isolates plain fiber-on-context nesting (no yield at all), so
// "does nesting itself work?" is answerable separately from "is the yield
// refused?". Both answers matter to D3 and they are different questions.
int g_probe_yield = 1;
void foreign_fiber_entry( void* )
{
g_foreign_ran = 1;
// We are on the foreign fiber's stack, but the registry still says the
// host context is running. This must NOT be allowed to yield it.
if( g_probe_yield )
g_foreign_yield_rc = yield_park( "from-a-foreign-stack" );
emscripten_fiber_swap( &g_foreign_fiber, &g_host_return );
}
void body_hosts_foreign_fiber( void* )
{
// Exactly libcontext's shape: the current stack (this context's) becomes a
// fiber we can be swapped back into, and the tool-like fiber is entered
// from it.
emscripten_fiber_init_from_current_context( &g_host_return, g_host_asyncify,
sizeof( g_host_asyncify ) );
emscripten_fiber_init( &g_foreign_fiber, foreign_fiber_entry, nullptr,
g_foreign_stack, sizeof( g_foreign_stack ),
g_foreign_asyncify, sizeof( g_foreign_asyncify ) );
emscripten_fiber_swap( &g_host_return, &g_foreign_fiber );
}
// Step 1: can a fiber nest inside a context AT ALL? (No yield involved.)
void Scenario_FiberNestsInContext()
{
g_foreign_yield_rc = -99;
g_foreign_ran = 0;
g_probe_yield = 0;
const ContextId host = create( body_hosts_foreign_fiber, nullptr, "nest-probe" );
drain();
if( !g_foreign_ran )
return report( "fiber_nests_in_context", false, "the nested fiber never ran" );
if( status_of( host ) != Status::Finished )
return report( "fiber_nests_in_context", false,
std::string( "host status: " ) + status_name( status_of( host ) ) );
destroy( host );
report( "fiber_nests_in_context", true );
}
// Step 2: and does a yield FROM that nested fiber get refused?
void Scenario_ForeignStackRefused()
{
g_foreign_yield_rc = -99;
g_foreign_ran = 0;
g_probe_yield = 1;
g_host_ctx = create( body_hosts_foreign_fiber, nullptr, "foreign-host" );
drain();
if( !g_foreign_ran )
return report( "foreign_stack_refused", false, "the foreign fiber never ran" );
if( g_foreign_yield_rc != -1 )
return report( "foreign_stack_refused", false,
"yield_park did not refuse (rc=" + std::to_string( g_foreign_yield_rc )
+ ") - it would have saved the wrong stack" );
// The host context must be intact: it ran to completion after the foreign
// fiber returned, rather than being left in a half-yielded state.
if( status_of( g_host_ctx ) != Status::Finished )
return report( "foreign_stack_refused", false,
std::string( "host context status: " )
+ status_name( status_of( g_host_ctx ) ) );
// And the scheduler is still healthy enough to run something else.
g_ran = 0;
const ContextId after = create( body_sets_flag, nullptr, "after-foreign" );
drain();
if( g_ran != 1 )
return report( "foreign_stack_refused", false,
"the scheduler was left unusable after the refusal" );
destroy( after );
destroy( g_host_ctx );
report( "foreign_stack_refused", true );
}
// ---------------------------------------------------------------------------
// FIBER LANE scenarios (doc 22 Phase A) — the symmetric semantics libcontext's
// wasm backend adapts onto. These pin what the adapter relies on: a fresh
// fiber enters at its entry, a swap suspends the swapper, enterability is
// registry truth, releasing a suspended fiber is legal, a stale id refuses
// instead of use-after-free, and the star lane is undisturbed throughout.
// ---------------------------------------------------------------------------
ContextId g_fiber_root = 0;
ContextId fiber_root()
{
// Adopted once for the whole battery, like libcontext's main context.
if( !g_fiber_root )
g_fiber_root = fiber_adopt_current( 64 * 1024, "test-root" );
return g_fiber_root;
}
alignas( 16 ) char g_fiber_a_stack[64 * 1024];
ContextId g_fiber_a = 0;
int g_fiber_a_runs = 0;
int g_fiber_a_saw_current = -1;
int g_fiber_a_saw_root_enterable = -1;
int g_fiber_a_saw_self_enterable = -1;
void fiber_a_entry( void* )
{
// Never returns, like libcontext's trampoline: run, swap back, repeat.
for( ;; )
{
++g_fiber_a_runs;
g_fiber_a_saw_current = ( fiber_current() == g_fiber_a ) ? 1 : 0;
g_fiber_a_saw_root_enterable = fiber_enterable( g_fiber_root ) ? 1 : 0;
g_fiber_a_saw_self_enterable = fiber_enterable( g_fiber_a ) ? 1 : 0;
fiber_swap( g_fiber_a, g_fiber_root );
}
}
void Scenario_FiberRoundtrip()
{
if( !fiber_root() )
return report( "fiber_roundtrip", false, "could not adopt the root" );
g_fiber_a_runs = 0;
g_fiber_a = fiber_create( fiber_a_entry, nullptr, g_fiber_a_stack,
sizeof( g_fiber_a_stack ), 32 * 1024, "fiber-a" );
if( !g_fiber_a )
return report( "fiber_roundtrip", false, "fiber_create failed" );
if( status_of( g_fiber_a ) != Status::Fresh || !fiber_enterable( g_fiber_a ) )
return report( "fiber_roundtrip", false, "a fresh fiber should be enterable" );
if( g_fiber_a_runs != 0 )
return report( "fiber_roundtrip", false, "body ran before the first swap" );
if( !fiber_swap( g_fiber_root, g_fiber_a ) )
return report( "fiber_roundtrip", false, "first swap refused" );
// Back here: the fiber ran once and swapped back to the root.
if( g_fiber_a_runs != 1 )
return report( "fiber_roundtrip", false,
"runs=" + std::to_string( g_fiber_a_runs ) + " after first swap" );
if( fiber_current() != g_fiber_root )
return report( "fiber_roundtrip", false, "root is not fiber_current() again" );
if( status_of( g_fiber_a ) != Status::Suspended || !fiber_enterable( g_fiber_a ) )
return report( "fiber_roundtrip", false,
std::string( "suspended fiber status: " )
+ status_name( status_of( g_fiber_a ) ) );
// What the fiber observed mid-run: it was current, the suspended root was
// enterable, and it itself (Running) was not — registry truth, no guessing.
if( g_fiber_a_saw_current != 1 )
return report( "fiber_roundtrip", false, "fiber did not see itself as current" );
if( g_fiber_a_saw_root_enterable != 1 )
return report( "fiber_roundtrip", false, "suspended root was not enterable" );
if( g_fiber_a_saw_self_enterable != 0 )
return report( "fiber_roundtrip", false, "a RUNNING fiber claimed to be enterable" );
// Re-entry takes the loop again — the second swap resumes, not restarts.
if( !fiber_swap( g_fiber_root, g_fiber_a ) || g_fiber_a_runs != 2 )
return report( "fiber_roundtrip", false, "second swap did not re-run the body" );
report( "fiber_roundtrip", true );
}
void Scenario_FiberReleaseSuspended()
{
// Fiber A is Suspended after the roundtrip. Releasing it mid-suspend is
// LEGAL — libcontext refcounts drop never-finished coroutines — and the
// stale id must then refuse instead of resuming freed state.
if( !g_fiber_a || status_of( g_fiber_a ) != Status::Suspended )
return report( "fiber_release_suspended", false, "precondition: fiber A suspended" );
if( !fiber_release( g_fiber_a ) )
return report( "fiber_release_suspended", false, "release refused" );
if( fiber_enterable( g_fiber_a ) )
return report( "fiber_release_suspended", false, "released fiber still enterable" );
if( fiber_swap( g_fiber_root, g_fiber_a ) )
return report( "fiber_release_suspended", false,
"swap into a released fiber was allowed (use-after-free)" );
report( "fiber_release_suspended", true );
}
alignas( 16 ) char g_fiber_b_stack[64 * 1024];
ContextId g_fiber_b = 0;
int g_fiber_b_runs = 0;
int g_coexist_stage = 0;
int g_coexist_result = 0;
void fiber_b_entry( void* )
{
for( ;; )
{
++g_fiber_b_runs;
fiber_swap( g_fiber_b, g_fiber_root );
}
}
void body_coexist_park( void* )
{
g_coexist_stage = 1;
g_coexist_result = yield_park( "coexist-park" );
g_coexist_stage = 2;
}
void Scenario_FiberAndStarCoexist()
{
// One registry, two lanes: a PARKED star context must sit undisturbed
// while symmetric fiber swaps happen, and resume cleanly afterwards —
// production Phase A runs tool fibers while D1's star layer idles.
g_coexist_stage = 0;
g_fiber_b_runs = 0;
const ContextId star = create( body_coexist_park, nullptr, "coexist-star" );
drain(); // parks
if( g_coexist_stage != 1 || status_of( star ) != Status::Parked )
return report( "fiber_and_star_coexist", false, "star context did not park" );
g_fiber_b = fiber_create( fiber_b_entry, nullptr, g_fiber_b_stack,
sizeof( g_fiber_b_stack ), 32 * 1024, "fiber-b" );
if( !g_fiber_b || !fiber_swap( g_fiber_root, g_fiber_b ) || g_fiber_b_runs != 1 )
return report( "fiber_and_star_coexist", false, "fiber roundtrip failed" );
if( status_of( star ) != Status::Parked )
return report( "fiber_and_star_coexist", false,
"fiber swaps disturbed the parked star context" );
if( !mark_ready( star, 7 ) )
return report( "fiber_and_star_coexist", false, "mark_ready failed" );
drain();
if( g_coexist_stage != 2 || g_coexist_result != 7 )
return report( "fiber_and_star_coexist", false, "star context did not resume" );
destroy( star );
fiber_release( g_fiber_b );
report( "fiber_and_star_coexist", true );
}
// ---------------------------------------------------------------------------
// STAR-MODE TRANSFERS (doc 22 Phase B) — libcontext's symmetric contract kept,
// but every entry performed by the scheduler.
//
// The load-bearing claim: KiCad's TOOL_MANAGER calls cofunc->Call(evt) and
// reads Running() on the NEXT LINE, so a "call" must return only after the
// coroutine has yielded or finished. Under the star the caller parks and the
// scheduler resumes it — from the caller's C++ frame that is indistinguishable
// from a synchronous return, which is what makes the flip possible at all.
// ---------------------------------------------------------------------------
// A fiber entry must NEVER return — "if entry_func returns, the entire program
// will end" (emscripten fiber.h). libcontext's trampoline loops forever for
// exactly this reason; these harness bodies do the same by parking for good
// once their script is finished. (Getting this wrong wedges the battery with
// no output at all, which is how it was found.)
[[noreturn]] void park_forever()
{
for( ;; )
yield_park( "fiber-finished" );
}
alignas( 16 ) char g_tool_stack[64 * 1024];
alignas( 16 ) char g_caller_stack[64 * 1024];
ContextId g_star_caller = 0;
ContextId g_star_tool = 0;
int g_tool_entries = 0;
int g_tool_yields = 0;
intptr_t g_tool_saw_value = 0;
int g_caller_stage = 0;
intptr_t g_call_returned = 0;
int g_tool_running_after_call = -1;
int g_call_order_violation = 0;
// The "coroutine": runs, yields once, then finishes — libcontext's shape.
void star_tool_entry( void* )
{
for( ;; )
{
++g_tool_entries;
g_tool_saw_value = 0;
// KiYield(): hand control back to whoever called us.
++g_tool_yields;
fiber_transfer( g_star_tool, g_star_caller, 0xB1 );
// Resumed: finish and go back for good.
fiber_transfer( g_star_tool, g_star_caller, 0xF1 );
}
}
// The "TOOL_MANAGER": Call(), then read state on the next line.
void star_caller_entry( void* )
{
g_caller_stage = 1;
// Call(): the tool must have run and yielded by the time this returns.
g_call_returned = fiber_transfer( g_star_caller, g_star_tool, 0xA1 );
// THE assertion that decides whether Phase B is possible.
g_tool_running_after_call = ( g_tool_entries == 1 && g_tool_yields == 1 ) ? 1 : 0;
if( g_tool_entries != 1 )
g_call_order_violation = 1;
g_caller_stage = 2;
// Resume(): drive it to completion.
fiber_transfer( g_star_caller, g_star_tool, 0xA2 );
g_caller_stage = 3;
park_forever();
}
void Scenario_StarTransferKeepsCallSynchronous()
{
g_tool_entries = 0;
g_tool_yields = 0;
g_caller_stage = 0;
g_call_returned = 0;
g_tool_running_after_call = -1;
g_call_order_violation = 0;
g_star_caller = fiber_create( star_caller_entry, nullptr, g_caller_stack,
sizeof( g_caller_stack ), 32 * 1024, "star-caller" );
g_star_tool = fiber_create( star_tool_entry, nullptr, g_tool_stack,
sizeof( g_tool_stack ), 32 * 1024, "star-tool" );
if( !g_star_caller || !g_star_tool )
return report( "star_transfer_call_is_synchronous", false, "fiber_create failed" );
// The scheduler kicks the caller, then pumps until quiescent — exactly the
// shape the production tick will have.
if( !fiber_start( g_star_caller, 0 ) )
return report( "star_transfer_call_is_synchronous", false, "fiber_start refused" );
const size_t ran = drain_all();
if( g_caller_stage != 3 )
return report( "star_transfer_call_is_synchronous", false,
"caller did not run to completion (stage="
+ std::to_string( g_caller_stage ) + ")" );
if( g_call_order_violation )
return report( "star_transfer_call_is_synchronous", false,
"the tool had not run when Call() returned - the synchronous "
"contract TOOL_MANAGER depends on is broken" );
if( g_tool_running_after_call != 1 )
return report( "star_transfer_call_is_synchronous", false,
"tool state wrong right after Call(): entries="
+ std::to_string( g_tool_entries ) + " yields="
+ std::to_string( g_tool_yields ) );
if( g_call_returned != 0xB1 )
return report( "star_transfer_call_is_synchronous", false,
"Call() returned the wrong transfer value: "
+ std::to_string( g_call_returned ) );
if( ran < 4 )
return report( "star_transfer_call_is_synchronous", false,
"too few transitions (" + std::to_string( ran ) + ")" );
fiber_release( g_star_tool );
fiber_release( g_star_caller );
report( "star_transfer_call_is_synchronous", true );
}
// A transfer chain three deep (root → A → B), the nested-Call shape, plus the
// proof that a parked star context in the OTHER lane is untouched throughout.
alignas( 16 ) char g_chain_a_stack[64 * 1024];
alignas( 16 ) char g_chain_b_stack[64 * 1024];
ContextId g_chain_root = 0;
ContextId g_chain_a = 0;
ContextId g_chain_b = 0;
int g_chain_trace[8];
int g_chain_len = 0;
void chain_note( int aMark )
{
if( g_chain_len < 8 )
g_chain_trace[g_chain_len++] = aMark;
}
void chain_b_entry( void* )
{
for( ;; )
{
chain_note( 3 );
fiber_transfer( g_chain_b, g_chain_a, 0 );
}
}
void chain_a_entry( void* )
{
for( ;; )
{
chain_note( 2 );
fiber_transfer( g_chain_a, g_chain_b, 0 ); // nested "Call"
chain_note( 4 );
fiber_transfer( g_chain_a, g_chain_root, 0 );
}
}
void chain_root_entry( void* )
{
chain_note( 1 );
fiber_transfer( g_chain_root, g_chain_a, 0 );
chain_note( 5 );
park_forever();
}
void Scenario_StarTransferChain()
{
g_chain_len = 0;
const ContextId parked_star = create( body_parks_once, nullptr, "coexist-during-chain" );
drain(); // it parks and must stay parked across every transfer below
g_chain_root = fiber_create( chain_root_entry, nullptr, g_caller_stack,
sizeof( g_caller_stack ), 32 * 1024, "chain-root" );
g_chain_a = fiber_create( chain_a_entry, nullptr, g_chain_a_stack,
sizeof( g_chain_a_stack ), 32 * 1024, "chain-a" );
g_chain_b = fiber_create( chain_b_entry, nullptr, g_chain_b_stack,
sizeof( g_chain_b_stack ), 32 * 1024, "chain-b" );
if( !g_chain_root || !g_chain_a || !g_chain_b )
return report( "star_transfer_chain", false, "fiber_create failed" );
fiber_start( g_chain_root, 0 );
drain_all();
// Strict order: root, A, B, back to A, back to root.
const int expected[5] = { 1, 2, 3, 4, 5 };
if( g_chain_len != 5 )
return report( "star_transfer_chain", false,
"trace length " + std::to_string( g_chain_len ) + ", expected 5" );
for( int i = 0; i < 5; ++i )
{
if( g_chain_trace[i] != expected[i] )
return report( "star_transfer_chain", false,
"wrong order at " + std::to_string( i ) + ": "
+ std::to_string( g_chain_trace[i] ) );
}
if( status_of( parked_star ) != Status::Parked )
return report( "star_transfer_chain", false,
"the parked star context was disturbed by the transfers" );
mark_ready( parked_star, 1 );
drain();
destroy( parked_star );
fiber_release( g_chain_b );
fiber_release( g_chain_a );
fiber_release( g_chain_root );
report( "star_transfer_chain", true );
}
// ---------------------------------------------------------------------------
// scenario 10: a REAL async wake — JS timeout resolves the park
design-b D1b/c: context harness + the bounded-and-measured memory gate Doc 20 §6 D1, second slice: the dedicated test app D1 calls for, plus the doc 20 §7 risk-1 gate. tests/apps/standalone/sched-context/ drives the primitives with no wx linked (a failure can only be the contexts layer) and assertions ON — unlike the races harness, an emscripten "cannot start an async operation" state here IS the bug, not a tolerated state. Nine scenarios, each pinning one invariant; the load-bearing ones: - parked_does_not_block: a parked context does not stop others running. This is doc 19's freeze made unrepresentable — there a parked activity held a global interlock and the UI died; here "parked" simply means "not runnable", and three workers run past it before it resumes. - one_transition_in_flight: a context calling drain() gets a no-op, with a second ready context queued so a buggy nested drain would actually run something and be caught. - async_wake: a real macrotask hop (setTimeout → mark_ready → drain), the shape every production bridge has. - deep_park_sizing: parks 64 live frames deep to measure what a park actually costs. The memory gate asserts the ceiling (peak live contexts, peak bytes), that nothing leaked (live=0, bytes=0, created==finished), that nothing was left mid-transition, and that refusals occurred (zero would mean the illegal-operation scenarios stopped provoking). Measured here: ~34 B/frame, 2200 B for a 64-frame park, 1 MB peak for 4 concurrent contexts. Recorded with the caveat that the harness's frames carry three locals each, so this is a FLOOR — real bridges save far more per frame (libcontext runs 512 K after a 64 K buffer silently overflowed). The apparatus and its units are validated; the sizing DECISION needs deep-park numbers from real bridges at D3/D4. playwright.config: asyncify-firefox now matches every spec in ./asyncify instead of only asyncify-races* — a new harness there is covered by construction rather than by remembering to widen the pattern. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
2026-08-05 21:41:24 +02:00
// The park's wake crosses a JS turn, which is the shape every production
// bridge has (doc 21 D4): the context must resume from a fresh task, and
// the wake must never rewind inline.
// ---------------------------------------------------------------------------
ContextId g_async_ctx = 0;
int g_async_result = 0;
int g_async_stage = 0;
void body_async_park( void* )
{
g_async_stage = 1;
g_async_result = yield_park( "js-promise" );
g_async_stage = 2;
}
void finish_battery()
{
log_line( std::string( "[SCHED_CTX] STATS " ) + stats_json() );
log_line( std::string( "[SCHED_CTX] REGISTRY " ) + registry_json() );
log_line( "[SCHED_CTX] SUMMARY total=" + std::to_string( g_total )
+ " passed=" + std::to_string( g_passed )
+ " failed=" + std::to_string( g_total - g_passed ) );
}
extern "C" EMSCRIPTEN_KEEPALIVE void sched_ctx_async_wake( int aResult )
{
if( !mark_ready( g_async_ctx, aResult ) )
{
report( "async_wake", false, "mark_ready from JS failed" );
finish_battery();
return;
}
// Resume from THIS fresh JS task — never inline in the resolver.
drain();
if( g_async_stage != 2 )
report( "async_wake", false, "context did not resume" );
else if( g_async_result != aResult )
report( "async_wake", false, "wrong result: " + std::to_string( g_async_result ) );
else if( status_of( g_async_ctx ) != Status::Finished )
report( "async_wake", false, "context did not finish" );
else
report( "async_wake", true );
destroy( g_async_ctx );
finish_battery();
}
void Scenario_AsyncWake()
{
g_async_stage = 0;
g_async_result = 0;
g_async_ctx = create( body_async_park, nullptr, "async-wake" );
drain(); // parks awaiting the JS wake
if( g_async_stage != 1 || status_of( g_async_ctx ) != Status::Parked )
{
report( "async_wake", false, "context did not park" );
finish_battery();
return;
}
// A real macrotask hop, exactly like a bridge promise settling.
EM_ASM( { setTimeout( function() { Module["_sched_ctx_async_wake"]( 777 ); }, 20 ); } );
}
} // namespace
int main()
{
log_line( "[SCHED_CTX] battery start" );
Scenario_CreateRunsOnDrain();
Scenario_ParkAndResume();
Scenario_NoParkInPlace();
Scenario_ParkedDoesNotBlock();
Scenario_FifoOrder();
Scenario_OneTransitionInFlight();
Scenario_RegistryRefusals();
Scenario_DeepPark();
Scenario_FiberNestsInContext();
Scenario_ForeignStackRefused();
Scenario_FiberRoundtrip();
Scenario_FiberReleaseSuspended();
Scenario_FiberAndStarCoexist();
Scenario_StarTransferKeepsCallSynchronous();
Scenario_StarTransferChain();
design-b D1b/c: context harness + the bounded-and-measured memory gate Doc 20 §6 D1, second slice: the dedicated test app D1 calls for, plus the doc 20 §7 risk-1 gate. tests/apps/standalone/sched-context/ drives the primitives with no wx linked (a failure can only be the contexts layer) and assertions ON — unlike the races harness, an emscripten "cannot start an async operation" state here IS the bug, not a tolerated state. Nine scenarios, each pinning one invariant; the load-bearing ones: - parked_does_not_block: a parked context does not stop others running. This is doc 19's freeze made unrepresentable — there a parked activity held a global interlock and the UI died; here "parked" simply means "not runnable", and three workers run past it before it resumes. - one_transition_in_flight: a context calling drain() gets a no-op, with a second ready context queued so a buggy nested drain would actually run something and be caught. - async_wake: a real macrotask hop (setTimeout → mark_ready → drain), the shape every production bridge has. - deep_park_sizing: parks 64 live frames deep to measure what a park actually costs. The memory gate asserts the ceiling (peak live contexts, peak bytes), that nothing leaked (live=0, bytes=0, created==finished), that nothing was left mid-transition, and that refusals occurred (zero would mean the illegal-operation scenarios stopped provoking). Measured here: ~34 B/frame, 2200 B for a 64-frame park, 1 MB peak for 4 concurrent contexts. Recorded with the caveat that the harness's frames carry three locals each, so this is a FLOOR — real bridges save far more per frame (libcontext runs 512 K after a 64 K buffer silently overflowed). The apparatus and its units are validated; the sizing DECISION needs deep-park numbers from real bridges at D3/D4. playwright.config: asyncify-firefox now matches every spec in ./asyncify instead of only asyncify-races* — a new harness there is covered by construction rather than by remembering to widen the pattern. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEHGiiXMShNXbBr7gSJ7iz
2026-08-05 21:41:24 +02:00
// Async last: it emits STATS + SUMMARY when its wake lands.
Scenario_AsyncWake();
return 0;
}