emdashes begone (#10847)

This commit is contained in:
Tom
2026-07-01 19:01:27 -05:00
committed by GitHub
parent dee94e0758
commit 3becaf2d95
276 changed files with 1795 additions and 1793 deletions
+5 -5
View File
@@ -33,7 +33,7 @@ class TrafficManagementModule : public MeshModule, private concurrency::OSThread
TrafficManagementModule();
~TrafficManagementModule();
// Singleton — no copying or moving
// Singleton - no copying or moving
TrafficManagementModule(const TrafficManagementModule &) = delete;
TrafficManagementModule &operator=(const TrafficManagementModule &) = delete;
@@ -116,10 +116,10 @@ class TrafficManagementModule : public MeshModule, private concurrency::OSThread
// unknown active: getUnknownCount() != 0 (low 6 bits only)
//
// next_hop: routing hint written only from ACK-confirmed NextHopRouter decisions.
// No TTL — keeps the slot alive across maintenance sweeps.
// No TTL - keeps the slot alive across maintenance sweeps.
//
#if _meshtastic_Config_DeviceConfig_Role_MAX > 15
#warning "Device role enum max exceeds 15 — TMM 4-bit role cache (rate_count[7:6]/unknown_count[7:6]) will truncate new values"
#warning "Device role enum max exceeds 15 - TMM 4-bit role cache (rate_count[7:6]/unknown_count[7:6]) will truncate new values"
#endif
struct __attribute__((packed)) UnifiedCacheEntry {
NodeNum node;
@@ -152,7 +152,7 @@ class TrafficManagementModule : public MeshModule, private concurrency::OSThread
// =========================================================================
//
// Plain array, linear scan (same idiom as WarmNodeStore). A lookup walks at
// most cacheSize() × 10 B — microseconds at LoRa packet rates, not worth a
// most cacheSize() × 10 B - microseconds at LoRa packet rates, not worth a
// hash table. Insertion on a full cache evicts the stalest entry,
// preferring entries without a next_hop hint (those are the long-tail
// routing state this cache exists to keep).
@@ -270,7 +270,7 @@ class TrafficManagementModule : public MeshModule, private concurrency::OSThread
// Resolve a sender's advertised device role for the position hot path. The tier-3
// cache (this entry's getCachedRole) is authoritative and is kept fresh by
// updateCachedRoleFromNodeInfo() — updated when NodeDB learns a role, not re-derived
// updateCachedRoleFromNodeInfo() - updated when NodeDB learns a role, not re-derived
// per packet. Only on first tracking (isNew) do we scan NodeDB (hot store → warm
// store, via getNodeRole) to seed the cache, so a resident special-role node is
// correct from its first position; after that the read is O(1) and survives the node