Probe failures follow the compile branch's transient rules, strict on the native esp8266 job

.github/workflows/ci.yml

@@ -1202,8 +1202,10 @@ jobs:
           echo ""
 
           # ESP8266 Arduino built directly (no PlatformIO); compile validates
-          # config first, so a separate config pass is redundant.
-          python3 script/test_build_components.py -e compile -t esp8266-ard -c "$TEST_COMPONENTS" -f --toolchain arduino --fail-on-no-tests
+          # config first, so a separate config pass is redundant. Strict pch:
+          # the docker smoke test covers the PlatformIO default, this job is
+          # the only CI exercise of the native ninja pch (and its probe edge).
+          ESPHOME_PCH_STRICT=1 python3 script/test_build_components.py -e compile -t esp8266-ard -c "$TEST_COMPONENTS" -f --toolchain arduino --fail-on-no-tests
 
   device-builder:
     name: Test downstream esphome/device-builder

esphome/build_helpers/pch.py

@@ -442,8 +442,9 @@ def prepare_pch(
     # Load probe: some toolchains build a .gch they then refuse to load
     # (per-process ASLR); dep flags are already stripped from cmd, so no
     # -MF is needed
+    # cmd ends with the fixed "-x c++-header -c <header> -o <gch>" tail
     probe_cmd = [
-        *cmd[: cmd.index("-x")],
+        *cmd[:-6],
         "-Winvalid-pch",
         "-include",
         str(header),
@@ -467,12 +468,22 @@ def prepare_pch(
         discard_pch(build_dir)
         pch_degraded(f"probe did not run: {err}")
         return
+    if probe.returncode < 0:
+        # Killed by a signal (OOM, ^C): environmental, do not latch
+        _LOGGER.warning(
+            "Precompiled header probe was killed (signal %d); retrying next build",
+            -probe.returncode,
+        )
+        discard_pch(build_dir)
+        pch_degraded(f"probe killed by signal {-probe.returncode}")
+        return
     if probe.returncode != 0 or ".gch" in probe.stderr:
         error = f"toolchain cannot load the pch: {probe.stderr.strip()[:400]}"
         _LOGGER.warning("Precompiled header failed; compiling without it: %s", error)
         discard_pch(build_dir)
-        failed_marker.write_text(checksum + "\n", encoding="utf-8")
-        os.utime(header)
+        if not any(m in error for m in _TRANSIENT_ERRORS):
+            failed_marker.write_text(checksum + "\n", encoding="utf-8")
+            os.utime(header)
         pch_degraded(error)
         return
     failed_marker.unlink(missing_ok=True)

tests/unit_tests/build_gen/test_arduino8266.py

@@ -1878,6 +1878,10 @@ def test_write_project_pch_strict_emits_probe_edge(
     )
     content = _write_ninja(paths, ccache="/usr/bin/ccache")
     assert "pchprobe esphome_pch.h.gch\n  flags = " in content
-    for line in content.splitlines():
-        if line.startswith("build obj/src/main.cpp.o:"):
-            assert line.endswith("| esphome_pch.h.gch esphome_pch.probe")
+    edges = [
+        line
+        for line in content.splitlines()
+        if line.startswith("build obj/src/") and ".cpp.o:" in line
+    ]
+    assert edges
+    assert all(line.endswith("| esphome_pch.h.gch esphome_pch.probe") for line in edges)

tests/unit_tests/build_gen/test_espidf.py

@@ -903,6 +903,32 @@ def test_prepare_pch_probe_spawn_failure_degrades(
         prepare_pch()
 
 
+def test_prepare_pch_probe_environmental_failures_do_not_latch(
+    tmp_path: Path,
+) -> None:
+    """A signal-killed or ENOSPC probe retries next build, no marker."""
+    from esphome.build_gen.espidf import prepare_pch
+
+    for stderr, code in (("", -9), ("fatal: No space left on device", 1)):
+        dev = _make_pch_device(tmp_path, f"dev_pe{code}")
+        CORE.build_path = dev
+        gch = dev / "build" / "esphome_pch.h.gch"
+
+        def env_probe(cmd, _gch=gch, _stderr=stderr, _code=code, **kwargs):
+            if "-fsyntax-only" in cmd:
+                return subprocess.CompletedProcess(cmd, _code, "", _stderr)
+            _gch.write_bytes(b"gch")
+            return subprocess.CompletedProcess(cmd, 0, "", "")
+
+        with (
+            patch.object(CORE, "name", "test"),
+            patch("esphome.build_helpers.pch.subprocess.run", side_effect=env_probe),
+        ):
+            prepare_pch()
+        assert not gch.exists()
+        assert not (dev / "build" / "esphome_pch.h.gch.failed").exists()
+
+
 def test_prepare_pch_signal_kill_strict_raises(
     tmp_path: Path, monkeypatch: pytest.MonkeyPatch
 ) -> None: