From 847d7313b1dc5c72799c7a89693e36f95266ee42 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Tue, 14 Apr 2026 00:37:31 +0000
Subject: [PATCH] Fix RPI breaking all mod-tap holds: revert to manual
 last_key_time
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

last_input_activity_elapsed() includes the mod-tap key's OWN matrix
event. Since process_record_user fires ~tapping_term ms after the
physical press (after action_tapping resolves tap/hold), the elapsed
time is always ≈ tapping_term < RPI thresholds, so holds never work.

Revert to manual last_key_time updated at the END of process_record_user.
Remove SPLIT_ACTIVITY_ENABLE (not needed — master processes all keys
from both halves).

Agent-Logs-Url: https://github.com/timfee/qmk_userspace/sessions/14d8fea9-7e5d-40c4-9b47-abab68e5d4e2

Co-authored-by: timfee <3246342+timfee@users.noreply.github.com>
---
 users/timfee/config.h |  1 -
 users/timfee/timfee.c | 22 ++++++++++++++--------
 2 files changed, 14 insertions(+), 9 deletions(-)

diff --git a/users/timfee/config.h b/users/timfee/config.h
index f28ca664..75a91967 100644
--- a/users/timfee/config.h
+++ b/users/timfee/config.h
@@ -29,6 +29,5 @@
 // ── Split transport ──
 #define SPLIT_TRANSPORT_MIRROR
 #define SPLIT_WPM_ENABLE
-#define SPLIT_ACTIVITY_ENABLE
 #define SPLIT_WATCHDOG_TIMEOUT 4000
 #define SPLIT_TRANSACTION_IDS_USER USER_SYNC_OLED_STATE, USER_SYNC_LASTKEY, USER_SYNC_PRESSES
diff --git a/users/timfee/timfee.c b/users/timfee/timfee.c
index 10969ba6..8ca0d203 100644
--- a/users/timfee/timfee.c
+++ b/users/timfee/timfee.c
@@ -6,8 +6,16 @@
 #endif
 
 // ── State for require-prior-idle ──
-// (last_key_time removed: using last_input_activity_elapsed() instead, which
-//  is synced from both halves via SPLIT_ACTIVITY_ENABLE, fixing cross-half RPI)
+// Manual timestamp of the last key processed through process_record_user.
+// This is intentionally NOT last_input_activity_elapsed() — that API reflects
+// the most recent matrix scan change, which includes the current mod-tap key's
+// OWN press. Since process_record_user fires ~tapping_term ms after the
+// physical press (after tap/hold resolution), the elapsed time would always be
+// ≈ tapping_term, which is less than the RPI thresholds — killing every hold.
+// A manual variable updated at the END of process_record_user correctly
+// captures the previous key's timing. This works for both halves because QMK
+// split processes all key events on the master side.
+static uint16_t last_key_time = 0;
 
 // ── Combos (matching Vial config) ──
 const uint16_t PROGMEM lparen_combo[] = {KC_R, KC_T, COMBO_END};
@@ -214,12 +222,8 @@ bool process_record_user(uint16_t keycode, keyrecord_t *record) {
         }
 #endif
 
-        // Require-prior-idle: force tap if pressed too soon after any recent
-        // activity on either half. SPLIT_ACTIVITY_ENABLE synchronizes the
-        // underlying activity timestamp from the slave to the master at the
-        // transport level, so last_input_activity_elapsed() reflects the most
-        // recent keypress on either half — not just the master side.
-        uint32_t elapsed = last_input_activity_elapsed();
+        // Require-prior-idle: force tap if pressed too soon after last key
+        uint16_t elapsed = timer_elapsed(last_key_time);
 
         switch (keycode) {
             case GU_SPC:
@@ -271,6 +275,8 @@ bool process_record_user(uint16_t keycode, keyrecord_t *record) {
                 }
                 break;
         }
+
+        last_key_time = timer_read();
     }
     return true;
 }