build_utils: key the caches on the tree, not just the argumentsclaude/ci-select-rule17

The eight lru_cache layers take repo-RELATIVE paths - 'hw/bsp/<fam>',
'examples/<ex>/skip.txt', the literal 'hw/bsp' glob - while ci_select._in_repo()
chdirs around every call so one process can classify more than one tree. With no
cwd in the key the second tree gets the first tree's answers.

Reproduced: skip_example('host/bare_api','metro_m0_express') is False at the repo
root and STILL False after chdir into a tree where that board does not exist; only
cache_clear() gave the right answer. It bites the code-size skill's base-vs-branch
worktree compare, /pre-pr, and the first test that points classify_build at a
fixture tree. Master had no caching here, so the hazard arrived with it.

_cwd_cache puts os.getcwd() in the key. The 199-test suite passed before only
because every test happens to pass the real REPO; the new
TestCachesAreKeyedOnTheTree crosses trees deliberately.

Also adds the drift guard the class rule was missing. Ports, hw/mcu, get_deps
tokens and bsp families each have one; the class rule had only a comment claiming
vendor_host.c was the sole "enabled by no example config" case until its removal -
which src/class/bth falsifies today. TestClassesWithNoEnablingExample pins the set
to {bth}, so a class added before its first example, or an example config flipped
to 0, fails here instead of silently selecting nothing on both axes. Verified it
fires by adding a class dir nothing enables.

2 files changed, 92 insertions, 11 deletions

diff --git a/test/hil/test/test_ci_select.py b/test/hil/test/test_ci_select.py
index 66b20b2e4..9b2c61ef2 100644
--- a/test/hil/test/test_ci_select.py
+++ b/test/hil/test/test_ci_select.py
@@ -840,6 +840,62 @@ class TestRostersDoNotOverlap(unittest.TestCase):
                 seen[b['name']] = b.get('tests')
 
 
+class TestCachesAreKeyedOnTheTree(unittest.TestCase):
+    """build_utils caches on repo-RELATIVE paths while ci_select._in_repo() chdirs
+    between trees, so the cwd has to be part of every cache key. Without it a second
+    tree gets the first tree's skip.txt/only.txt and FAMILY_MCUS - which is exactly the
+    base-vs-branch comparison the code-size skill does in one process."""
+
+    def test_a_second_tree_is_not_answered_from_the_first(self):
+        import build_utils, tempfile
+        old = os.getcwd()
+        try:
+            os.chdir(REPO)
+            self.assertFalse(build_utils.skip_example('host/bare_api', 'metro_m0_express'))
+            with tempfile.TemporaryDirectory() as d:
+                os.makedirs(os.path.join(d, 'hw/bsp'), exist_ok=True)
+                os.chdir(d)
+                # the board does not exist in this tree at all -> unknown board -> skip
+                self.assertTrue(build_utils.skip_example('host/bare_api', 'metro_m0_express'),
+                                'the empty tree was answered from the repo tree cache')
+            os.chdir(REPO)
+            self.assertFalse(build_utils.skip_example('host/bare_api', 'metro_m0_express'),
+                             'and the repo answer must survive the excursion')
+        finally:
+            os.chdir(old)
+
+
+class TestClassesWithNoEnablingExample(unittest.TestCase):
+    """The class rule is the one rule with no drift guard: ports, hw/mcu, get_deps
+    tokens and bsp families all have one. A class dir that no example config enables
+    selects NOTHING on both axes (the maintainer's empty-means-empty ruling), which is
+    right - but it must be a listed state, not a surprise, or a class added before its
+    first example silently stops being built."""
+
+    # class dirs no example's tusb_config.h turns on, for either role. Must only shrink:
+    # a new entry means a class nothing compiles, so a break in it reaches master.
+    NO_EXAMPLE = {'bth'}
+
+    def test_only_the_known_classes_select_nothing(self):
+        import glob as _glob
+        dead = set()
+        for d in sorted(_glob.glob(os.path.join(REPO, 'src/class/*'))):
+            if not os.path.isdir(d):
+                continue
+            cls = os.path.basename(d)
+            hit = False
+            for base in sorted(os.path.basename(f) for f in _glob.glob(os.path.join(d, '*.[ch]'))):
+                roles = ci_select._class_roles(base)
+                if ci_select._build_class_examples(cls, base, roles, REPO):
+                    hit = True
+                    break
+            if not hit:
+                dead.add(cls)
+        self.assertEqual(dead, self.NO_EXAMPLE,
+                         'a class dir enabled by no example config: it selects nothing on '
+                         'both axes, so nothing compiles it until the next master push')
+
+
 class TestNoTrackedFileIsUnclassified(unittest.TestCase):
     """Rule 17 (unclassified -> full on both axes) is the fail-open net for paths nobody
     anticipated. It must stay that way - a wrong `full` costs runner minutes and is
@@ -1373,11 +1429,12 @@ class TestBuildPostFilter(unittest.TestCase):
         self.assertTrue(any('gone from tree' in r for r in s['reasons']), s['reasons'])
 
     def test_class_source_selecting_nothing_selects_nothing(self):
-        # synthetic class-with-no-enabling-config case (vendor_host.c was the live
-        # instance until its removal): no config enables CFG_TUH_VENDOR, so
+        # a class-with-no-enabling-config case: no config enables CFG_TUH_VENDOR, so
         # nothing exercises it and nothing builds - empty means empty (maintainer
         # decision; the file is still parsed by every full master-push build, which is
-        # the accepted net for a break outside its #if guard)
+        # the accepted net for a break outside its #if guard). src/class/bth is the
+        # live instance of this state today; TestClassesWithNoEnablingExample pins the
+        # whole set, so a new one cannot appear unnoticed.
         s = ci_select.classify_build(['src/class/vendor/vendor_host.c'], REPO)
         self.assertFalse(s['full'])
         self.assertEqual(s['families'], [])
diff --git a/tools/build_utils.py b/tools/build_utils.py
index 1eeef0269..1b81335e0 100755
--- a/tools/build_utils.py
+++ b/tools/build_utils.py
@@ -1,5 +1,6 @@
 #!/usr/bin/env python3
 import functools
+import os
 import subprocess
 import pathlib
 import re
@@ -24,7 +25,30 @@ _CMAKE_VAR_RE = re.compile(r'\$\{([A-Za-z_]\w*)\}')
 _CMAKE_CASE_RE = re.compile(r'string\s*\(\s*(TOUPPER|TOLOWER)\s+(\S+)\s+([A-Za-z_]\w*)\s*\)')
 
 
[email protected]_cache(maxsize=None)
+
+def _cwd_cache(fn):
+    """lru_cache, keyed on the working directory as well as the arguments.
+
+    Every cached helper below takes repo-RELATIVE paths ('hw/bsp/<fam>',
+    'examples/<ex>/skip.txt', or the literal 'hw/bsp' glob), while ci_select._in_repo()
+    chdirs around each call so one process can classify more than one tree - the
+    code-size skill's base-vs-branch worktrees, /pre-pr, a test pointing at a fixture.
+    Without the cwd in the key the second tree silently gets the first tree's
+    skip.txt/only.txt and FAMILY_MCUS answers. Master had no caching here, so this
+    hazard arrived with it."""
+    cache = {}
+
+    @functools.wraps(fn)
+    def wrapper(*args):
+        key = (os.getcwd(), args)
+        if key not in cache:
+            cache[key] = fn(*args)
+        return cache[key]
+
+    wrapper.cache_clear = cache.clear
+    return wrapper
+
+@_cwd_cache
 def _cmake_sets(path):
     """One cmake file's variable assignments as NAME -> first definition seen, as
     either a literal value or an ('TOUPPER'|'TOLOWER', source) pair. Only used to
@@ -87,7 +111,7 @@ def _cmake_expand(value, files, depth=0):
     return None if '${' in out else out
 
 
[email protected]_cache(maxsize=None)
+@_cwd_cache
 def _board_dirs(board):
     """(board_dir, family_dir) for a board name, or (None, None). Cached: skip_example
     is asked (board x example) times - 566k lstat calls per selector run without this,
@@ -98,7 +122,7 @@ def _board_dirs(board):
     return hits[0], hits[0].parent.parent
 
 
[email protected]_cache(maxsize=None)
+@_cwd_cache
 def _family_mcus(family_dir, board_dir):
     """The MCU names CMake's family_filter iterates. family_support.cmake:176/190
     loop `foreach(MCU IN LISTS FAMILY_MCUS)`, so a family-wide list (broadcom_64bit
@@ -175,7 +199,7 @@ def _family_mcus(family_dir, board_dir):
     return frozenset(out)
 
 
[email protected]_cache(maxsize=None)
+@_cwd_cache
 def _scrape_mcu(family_dir, board_dir, family):
     """(CFG_TUSB_MCU token of this board, the text it was read from), master's
     algorithm verbatim: family.mk (family.cmake when there is none) first, falling
@@ -215,7 +239,7 @@ def _scrape_mcu(family_dir, board_dir, family):
     return mcu, mk_contents
 
 
[email protected]_cache(maxsize=None)
+@_cwd_cache
 def _board_mcu(board_dir, family_dir, family):
     """(CFG_TUSB_MCU of this board, MAX3421_HOST enabled by its cmake BSP).
 
@@ -254,7 +278,7 @@ def _board_mcu(board_dir, family_dir, family):
     return mcu, max3421_enabled
 
 
[email protected]_cache(maxsize=None)
+@_cwd_cache
 def _filter_tokens(path):
     """skip.txt / only.txt as a token set, or None when the file does not exist."""
     f = pathlib.Path(path)
@@ -285,7 +309,7 @@ def skip_example(example, board, extra_defines=(), build_system='cmake'):
     return _skip_example(example, board, tuple(extra_defines), build_system)
 
 
[email protected]_cache(maxsize=None)
+@_cwd_cache
 def _skip_example_make(example, board):
     """master's skip_example, verbatim (tools/build_utils.py @ 9c202e8c6): the
     make build's own answer, derived from family.mk/board.mk with the single
@@ -333,7 +357,7 @@ def _skip_example_make(example, board):
     return False
 
 
[email protected]_cache(maxsize=None)
+@_cwd_cache
 def _skip_example(example, board, extra_defines, build_system):
     if build_system == 'make':
         return _skip_example_make(example, board)