fix: resolve Pulse v0.1.0 release blockers

[zekageri]

Summary

• replace the unsafe shutdown path with a four-state lifecycle, generation-aware concurrent end() handling, task-owned cleanup, intrusive implementation lifetime, and join-style destruction
• separate shutdown wakeup from the bounded control queue, synchronize all queue access, and make timer creation and control operations transactional
• enforce terminal callback semantics and process callback-generated controls before another already-due timer, with mutation generations preserving pause/restart results
• correct ESP-IDF stack high-water diagnostics to report bytes directly
• add deterministic host runtime tests, a device smoke sketch, no-exceptions compilation, version consistency checks, and release gating
• prepare 0.1.0 package metadata and curated changelog-based release notes

Why

The previous implementation could free PulseImpl while its task was still active, race queue deletion against public operations, report incorrect stack diagnostics, and accept callback controls that could later disappear or be overwritten. Compilation-only CI could not detect those behaviors.

Behavioral impact

• end() called from a Pulse callback returns Busy
• an end() timeout leaves Pulse safely in Stopping; a later call may continue waiting
• external destruction waits for the task to become quiescent
• timer creation and controls return Busy during shutdown
• shutdown remains available even when the command queue is full
• timeouts and final countdowns are terminal before their callback begins
• callback-side pause/restart operations on nonterminal timers are authoritative

Validation

Local validation completed before publishing:

• deterministic host runtime suite passed repeatedly, including 20 consecutive full-suite runs
• shutdown timeout/retry, concurrent waiters, callback-side end(), external destruction, self-destruction, and full-queue shutdown tests passed
• callback ordering, terminal countdown behavior, self-pause, and self-restart tests passed
• -fno-exceptions -Wall -Wextra -Werror production-source compilation passed
• source audit, version checks in branch and tag modes, changelog extraction, YAML parsing, and whitespace validation passed

CI / hardware

The existing PIOArduino and Arduino CLI board matrices remain required. CI also compiles the new device smoke sketch. Physical internal-stack and PSRAM-stack smoke runs remain part of the final release checklist when no hardware runner is connected.

[zekageri]

Closing this draft as an accidental duplicate of #1. The Arduino Lint LD003 issue was corrected by moving the smoke sketch to extras/PulseSmoke, but #1 is the canonical implementation and now has a fully successful lint and CI run.

[zekageri]

Summary

• replace the unsafe shutdown path with a four-state lifecycle, generation-aware concurrent end() handling, task-owned cleanup, intrusive implementation lifetime, and join-style destruction
• separate shutdown wakeup from the bounded control queue, synchronize all queue access, and make timer creation and control operations transactional
• enforce terminal callback semantics and process callback-generated controls before another already-due timer, with mutation generations preserving pause/restart results
• correct ESP-IDF stack high-water diagnostics to report bytes directly
• add deterministic host runtime tests, a device smoke sketch, no-exceptions compilation, version consistency checks, and release gating
• prepare 0.1.0 package metadata and curated changelog-based release notes

Why

The previous implementation could free PulseImpl while its task was still active, race queue deletion against public operations, report incorrect stack diagnostics, and accept callback controls that could later disappear or be overwritten. Compilation-only CI could not detect those behaviors.

Behavioral impact

• end() called from a Pulse callback returns Busy
• an end() timeout leaves Pulse safely in Stopping; a later call may continue waiting
• external destruction waits for the task to become quiescent
• timer creation and controls return Busy during shutdown
• shutdown remains available even when the command queue is full
• timeouts and final countdowns are terminal before their callback begins
• callback-side pause/restart operations on nonterminal timers are authoritative

Validation

Local validation completed before publishing:

• deterministic host runtime suite passed repeatedly, including 20 consecutive full-suite runs
• shutdown timeout/retry, concurrent waiters, callback-side end(), external destruction, self-destruction, and full-queue shutdown tests passed
• callback ordering, terminal countdown behavior, self-pause, and self-restart tests passed
• -fno-exceptions -Wall -Wextra -Werror production-source compilation passed
• source audit, version checks in branch and tag modes, changelog extraction, YAML parsing, and whitespace validation passed

CI / hardware

The existing PIOArduino and Arduino CLI board matrices remain required. CI also compiles the new device smoke sketch. Physical internal-stack and PSRAM-stack smoke runs remain part of the final release checklist when no hardware runner is connected.

[zekageri]

Closing this draft as an accidental duplicate of #1. The Arduino Lint LD003 issue was corrected by moving the smoke sketch to extras/PulseSmoke, but #1 is the canonical implementation and now has a fully successful lint and CI run.

[zekageri]

Summary

• replace the unsafe shutdown path with a four-state lifecycle, generation-aware concurrent end() handling, task-owned cleanup, intrusive implementation lifetime, and join-style destruction
• separate shutdown wakeup from the bounded control queue, synchronize all queue access, and make timer creation and control operations transactional
• enforce terminal callback semantics and process callback-generated controls before another already-due timer, with mutation generations preserving pause/restart results
• correct ESP-IDF stack high-water diagnostics to report bytes directly
• add deterministic host runtime tests, a device smoke sketch, no-exceptions compilation, version consistency checks, and release gating
• prepare 0.1.0 package metadata and curated changelog-based release notes

Why

The previous implementation could free PulseImpl while its task was still active, race queue deletion against public operations, report incorrect stack diagnostics, and accept callback controls that could later disappear or be overwritten. Compilation-only CI could not detect those behaviors.

Behavioral impact

• end() called from a Pulse callback returns Busy
• an end() timeout leaves Pulse safely in Stopping; a later call may continue waiting
• external destruction waits for the task to become quiescent
• timer creation and controls return Busy during shutdown
• shutdown remains available even when the command queue is full
• timeouts and final countdowns are terminal before their callback begins
• callback-side pause/restart operations on nonterminal timers are authoritative

Validation

Local validation completed before publishing:

• deterministic host runtime suite passed repeatedly, including 20 consecutive full-suite runs
• shutdown timeout/retry, concurrent waiters, callback-side end(), external destruction, self-destruction, and full-queue shutdown tests passed
• callback ordering, terminal countdown behavior, self-pause, and self-restart tests passed
• -fno-exceptions -Wall -Wextra -Werror production-source compilation passed
• source audit, version checks in branch and tag modes, changelog extraction, YAML parsing, and whitespace validation passed

CI / hardware

The existing PIOArduino and Arduino CLI board matrices remain required. CI also compiles the new device smoke sketch. Physical internal-stack and PSRAM-stack smoke runs remain part of the final release checklist when no hardware runner is connected.

[zekageri]

Closing this draft as an accidental duplicate of #1. The Arduino Lint LD003 issue was corrected by moving the smoke sketch to extras/PulseSmoke, but #1 is the canonical implementation and now has a fully successful lint and CI run.

[zekageri]

Summary

• replace the unsafe shutdown path with a four-state lifecycle, generation-aware concurrent end() handling, task-owned cleanup, intrusive implementation lifetime, and join-style destruction
• separate shutdown wakeup from the bounded control queue, synchronize all queue access, and make timer creation and control operations transactional
• enforce terminal callback semantics and process callback-generated controls before another already-due timer, with mutation generations preserving pause/restart results
• correct ESP-IDF stack high-water diagnostics to report bytes directly
• add deterministic host runtime tests, a device smoke sketch, no-exceptions compilation, version consistency checks, and release gating
• prepare 0.1.0 package metadata and curated changelog-based release notes

Why

The previous implementation could free PulseImpl while its task was still active, race queue deletion against public operations, report incorrect stack diagnostics, and accept callback controls that could later disappear or be overwritten. Compilation-only CI could not detect those behaviors.

Behavioral impact

• end() called from a Pulse callback returns Busy
• an end() timeout leaves Pulse safely in Stopping; a later call may continue waiting
• external destruction waits for the task to become quiescent
• timer creation and controls return Busy during shutdown
• shutdown remains available even when the command queue is full
• timeouts and final countdowns are terminal before their callback begins
• callback-side pause/restart operations on nonterminal timers are authoritative

Validation

Local validation completed before publishing:

• deterministic host runtime suite passed repeatedly, including 20 consecutive full-suite runs
• shutdown timeout/retry, concurrent waiters, callback-side end(), external destruction, self-destruction, and full-queue shutdown tests passed
• callback ordering, terminal countdown behavior, self-pause, and self-restart tests passed
• -fno-exceptions -Wall -Wextra -Werror production-source compilation passed
• source audit, version checks in branch and tag modes, changelog extraction, YAML parsing, and whitespace validation passed

CI / hardware

The existing PIOArduino and Arduino CLI board matrices remain required. CI also compiles the new device smoke sketch. Physical internal-stack and PSRAM-stack smoke runs remain part of the final release checklist when no hardware runner is connected.

[zekageri]

Closing this draft as an accidental duplicate of #1. The Arduino Lint LD003 issue was corrected by moving the smoke sketch to extras/PulseSmoke, but #1 is the canonical implementation and now has a fully successful lint and CI run.

[zekageri]

Summary

• replace the unsafe shutdown path with a four-state lifecycle, generation-aware concurrent end() handling, task-owned cleanup, intrusive implementation lifetime, and join-style destruction
• separate shutdown wakeup from the bounded control queue, synchronize all queue access, and make timer creation and control operations transactional
• enforce terminal callback semantics and process callback-generated controls before another already-due timer, with mutation generations preserving pause/restart results
• correct ESP-IDF stack high-water diagnostics to report bytes directly
• add deterministic host runtime tests, a device smoke sketch, no-exceptions compilation, version consistency checks, and release gating
• prepare 0.1.0 package metadata and curated changelog-based release notes

Why

The previous implementation could free PulseImpl while its task was still active, race queue deletion against public operations, report incorrect stack diagnostics, and accept callback controls that could later disappear or be overwritten. Compilation-only CI could not detect those behaviors.

Behavioral impact

• end() called from a Pulse callback returns Busy
• an end() timeout leaves Pulse safely in Stopping; a later call may continue waiting
• external destruction waits for the task to become quiescent
• timer creation and controls return Busy during shutdown
• shutdown remains available even when the command queue is full
• timeouts and final countdowns are terminal before their callback begins
• callback-side pause/restart operations on nonterminal timers are authoritative

Validation

Local validation completed before publishing:

• deterministic host runtime suite passed repeatedly, including 20 consecutive full-suite runs
• shutdown timeout/retry, concurrent waiters, callback-side end(), external destruction, self-destruction, and full-queue shutdown tests passed
• callback ordering, terminal countdown behavior, self-pause, and self-restart tests passed
• -fno-exceptions -Wall -Wextra -Werror production-source compilation passed
• source audit, version checks in branch and tag modes, changelog extraction, YAML parsing, and whitespace validation passed

CI / hardware

The existing PIOArduino and Arduino CLI board matrices remain required. CI also compiles the new device smoke sketch. Physical internal-stack and PSRAM-stack smoke runs remain part of the final release checklist when no hardware runner is connected.

[zekageri]

Closing this draft as an accidental duplicate of #1. The Arduino Lint LD003 issue was corrected by moving the smoke sketch to extras/PulseSmoke, but #1 is the canonical implementation and now has a fully successful lint and CI run.

[zekageri]

Summary

• replace the unsafe shutdown path with a four-state lifecycle, generation-aware concurrent end() handling, task-owned cleanup, intrusive implementation lifetime, and join-style destruction
• separate shutdown wakeup from the bounded control queue, synchronize all queue access, and make timer creation and control operations transactional
• enforce terminal callback semantics and process callback-generated controls before another already-due timer, with mutation generations preserving pause/restart results
• correct ESP-IDF stack high-water diagnostics to report bytes directly
• add deterministic host runtime tests, a device smoke sketch, no-exceptions compilation, version consistency checks, and release gating
• prepare 0.1.0 package metadata and curated changelog-based release notes

Why

The previous implementation could free PulseImpl while its task was still active, race queue deletion against public operations, report incorrect stack diagnostics, and accept callback controls that could later disappear or be overwritten. Compilation-only CI could not detect those behaviors.

Behavioral impact

• end() called from a Pulse callback returns Busy
• an end() timeout leaves Pulse safely in Stopping; a later call may continue waiting
• external destruction waits for the task to become quiescent
• timer creation and controls return Busy during shutdown
• shutdown remains available even when the command queue is full
• timeouts and final countdowns are terminal before their callback begins
• callback-side pause/restart operations on nonterminal timers are authoritative

Validation

Local validation completed before publishing:

• deterministic host runtime suite passed repeatedly, including 20 consecutive full-suite runs
• shutdown timeout/retry, concurrent waiters, callback-side end(), external destruction, self-destruction, and full-queue shutdown tests passed
• callback ordering, terminal countdown behavior, self-pause, and self-restart tests passed
• -fno-exceptions -Wall -Wextra -Werror production-source compilation passed
• source audit, version checks in branch and tag modes, changelog extraction, YAML parsing, and whitespace validation passed

CI / hardware

The existing PIOArduino and Arduino CLI board matrices remain required. CI also compiles the new device smoke sketch. Physical internal-stack and PSRAM-stack smoke runs remain part of the final release checklist when no hardware runner is connected.

[zekageri]

Closing this draft as an accidental duplicate of #1. The Arduino Lint LD003 issue was corrected by moving the smoke sketch to extras/PulseSmoke, but #1 is the canonical implementation and now has a fully successful lint and CI run.

[zekageri]

Summary

• replace the unsafe shutdown path with a four-state lifecycle, generation-aware concurrent end() handling, task-owned cleanup, intrusive implementation lifetime, and join-style destruction
• separate shutdown wakeup from the bounded control queue, synchronize all queue access, and make timer creation and control operations transactional
• enforce terminal callback semantics and process callback-generated controls before another already-due timer, with mutation generations preserving pause/restart results
• correct ESP-IDF stack high-water diagnostics to report bytes directly
• add deterministic host runtime tests, a device smoke sketch, no-exceptions compilation, version consistency checks, and release gating
• prepare 0.1.0 package metadata and curated changelog-based release notes

Why

The previous implementation could free PulseImpl while its task was still active, race queue deletion against public operations, report incorrect stack diagnostics, and accept callback controls that could later disappear or be overwritten. Compilation-only CI could not detect those behaviors.

Behavioral impact

• end() called from a Pulse callback returns Busy
• an end() timeout leaves Pulse safely in Stopping; a later call may continue waiting
• external destruction waits for the task to become quiescent
• timer creation and controls return Busy during shutdown
• shutdown remains available even when the command queue is full
• timeouts and final countdowns are terminal before their callback begins
• callback-side pause/restart operations on nonterminal timers are authoritative

Validation

Local validation completed before publishing:

• deterministic host runtime suite passed repeatedly, including 20 consecutive full-suite runs
• shutdown timeout/retry, concurrent waiters, callback-side end(), external destruction, self-destruction, and full-queue shutdown tests passed
• callback ordering, terminal countdown behavior, self-pause, and self-restart tests passed
• -fno-exceptions -Wall -Wextra -Werror production-source compilation passed
• source audit, version checks in branch and tag modes, changelog extraction, YAML parsing, and whitespace validation passed

CI / hardware

The existing PIOArduino and Arduino CLI board matrices remain required. CI also compiles the new device smoke sketch. Physical internal-stack and PSRAM-stack smoke runs remain part of the final release checklist when no hardware runner is connected.

[zekageri]

Closing this draft as an accidental duplicate of #1. The Arduino Lint LD003 issue was corrected by moving the smoke sketch to extras/PulseSmoke, but #1 is the canonical implementation and now has a fully successful lint and CI run.

[zekageri]

Summary

• replace the unsafe shutdown path with a four-state lifecycle, generation-aware concurrent end() handling, task-owned cleanup, intrusive implementation lifetime, and join-style destruction
• separate shutdown wakeup from the bounded control queue, synchronize all queue access, and make timer creation and control operations transactional
• enforce terminal callback semantics and process callback-generated controls before another already-due timer, with mutation generations preserving pause/restart results
• correct ESP-IDF stack high-water diagnostics to report bytes directly
• add deterministic host runtime tests, a device smoke sketch, no-exceptions compilation, version consistency checks, and release gating
• prepare 0.1.0 package metadata and curated changelog-based release notes

Why

The previous implementation could free PulseImpl while its task was still active, race queue deletion against public operations, report incorrect stack diagnostics, and accept callback controls that could later disappear or be overwritten. Compilation-only CI could not detect those behaviors.

Behavioral impact

• end() called from a Pulse callback returns Busy
• an end() timeout leaves Pulse safely in Stopping; a later call may continue waiting
• external destruction waits for the task to become quiescent
• timer creation and controls return Busy during shutdown
• shutdown remains available even when the command queue is full
• timeouts and final countdowns are terminal before their callback begins
• callback-side pause/restart operations on nonterminal timers are authoritative

Validation

Local validation completed before publishing:

• deterministic host runtime suite passed repeatedly, including 20 consecutive full-suite runs
• shutdown timeout/retry, concurrent waiters, callback-side end(), external destruction, self-destruction, and full-queue shutdown tests passed
• callback ordering, terminal countdown behavior, self-pause, and self-restart tests passed
• -fno-exceptions -Wall -Wextra -Werror production-source compilation passed
• source audit, version checks in branch and tag modes, changelog extraction, YAML parsing, and whitespace validation passed

CI / hardware

The existing PIOArduino and Arduino CLI board matrices remain required. CI also compiles the new device smoke sketch. Physical internal-stack and PSRAM-stack smoke runs remain part of the final release checklist when no hardware runner is connected.

[zekageri]

Closing this draft as an accidental duplicate of #1. The Arduino Lint LD003 issue was corrected by moving the smoke sketch to extras/PulseSmoke, but #1 is the canonical implementation and now has a fully successful lint and CI run.