Count GPUs without waking the one that is asleep

bin/omarchy-hw-hybrid-gpu

@@ -2,8 +2,29 @@
 
 # omarchy:summary=Detect whether the system has an active hybrid GPU configuration
 
+# Read the cached sysfs IDs rather than lspci, which reads PCI config space and
+# resumes a runtime-suspended GPU. This gate is what the Hybrid GPU menu row is
+# guarded by, so asking it used to wake the discrete card the row is about --
+# 2.6s on this laptop, and the card then stays powered. omarchy-hw-nvidia
+# already reads the same files for the same reason.
+#
+# Counting by class also stops "Non-VGA unclassified device" being counted as a
+# GPU: the old pattern matched the VGA inside Non-VGA, so a machine with one
+# graphics card and one such device counted two and called itself hybrid.
+pci_devices_path="${OMARCHY_PCI_DEVICES_PATH:-/sys/bus/pci/devices}"
+
 multiple_gpus() {
-  (($(lspci | grep -cE 'VGA|3D|Display') >= 2))
+  local device count=0
+
+  shopt -s nullglob
+
+  for device in "$pci_devices_path"/*; do
+    [[ -r $device/class ]] || continue
+    # 0x03 is the PCI display-controller class: VGA, XGA, 3D and other.
+    [[ $(< "$device/class") == 0x03* ]] && ((count++))
+  done
+
+  ((count >= 2))
 }
 
 if omarchy-cmd-present supergfxctl; then

test/shell.d/hw-hybrid-gpu-test.sh

@@ -28,18 +28,30 @@ esac
 printf '%s\n' "${SUPPORTED_MODES:-Integrated Hybrid}"
 STUB
 
-cat >"$fake_bin/lspci" <<'STUB'
-#!/bin/bash
+chmod +x "$fake_bin"/*
 
-for _ in $(seq "${GPU_COUNT:-1}"); do
-  echo "0000:00:02.0 VGA compatible controller: Stub GPU"
-done
-STUB
+# The detector counts display controllers out of sysfs rather than parsing
+# lspci, so the GPUs are faked as PCI device directories. Every tree also holds
+# an unclassified device, which lspci prints as "Non-VGA unclassified device":
+# the old pattern matched the VGA inside Non-VGA and counted it as a GPU.
+make_pci_tree() {
+  local count=$1 tree="$test_tmp/pci-$1" i
 
-chmod +x "$fake_bin"/*
+  rm -rf "$tree"
+  for ((i = 0; i < count; i++)); do
+    mkdir -p "$tree/0000:0$i:02.0"
+    printf '0x030000\n' >"$tree/0000:0$i:02.0/class"
+  done
+
+  mkdir -p "$tree/0000:80:14.5"
+  printf '0x000000\n' >"$tree/0000:80:14.5/class"
+
+  printf '%s' "$tree"
+}
 
 hybrid_gpu() {
-  PATH="$fake_bin:$PATH" timeout --kill-after=1s 10s bash "$ROOT/bin/omarchy-hw-hybrid-gpu"
+  PATH="$fake_bin:$PATH" OMARCHY_PCI_DEVICES_PATH="$(make_pci_tree "${GPU_COUNT:-1}")" \
+    timeout --kill-after=1s 10s bash "$ROOT/bin/omarchy-hw-hybrid-gpu"
 }
 
 hybrid_gpu ||
@@ -89,3 +101,24 @@ chmod +x "$fake_bin/omarchy-cmd-present"
 GPU_COUNT=2 hybrid_gpu ||
   fail "hybrid GPU detection counts GPUs without supergfxctl"
 pass "hybrid GPU detection counts GPUs without supergfxctl"
+
+# The unclassified device in every tree above is the regression: one graphics
+# card plus one of those used to count as two and report hybrid.
+cat >"$fake_bin/omarchy-cmd-present" <<'STUB'
+#!/bin/bash
+exit 1
+STUB
+chmod +x "$fake_bin/omarchy-cmd-present"
+
+GPU_COUNT=1 hybrid_gpu
+status=$?
+((status == 1)) ||
+  fail "one graphics card beside an unclassified device is not hybrid" "exit status: $status"
+pass "an unclassified PCI device is not counted as a graphics card"
+
+# Reading a device's class file does not touch PCI config space, so a suspended
+# discrete GPU is not resumed just to answer the menu guard this gate sits in.
+# Comments are stripped: the script explains at length why it avoids lspci.
+! grep -vE '^[[:space:]]*#' "$ROOT/bin/omarchy-hw-hybrid-gpu" | grep -q 'lspci' ||
+  fail "the detector must not resume a suspended GPU to count GPUs"
+pass "the detector counts GPUs without waking one"