bpo-33499: Add PYTHONPYCACHEPREFIX env var for alt bytecode cache location. (GH-6834)

carljm · ncoghlan · commit b193fa996a74 · 2018-06-16T14:40:56.000+10:00
In some development setups it is inconvenient or impossible to write bytecode
caches to the code tree, but the bytecode caches are still useful. The
PYTHONPYCACHEPREFIX environment variable allows specifying an alternate
location for cached bytecode files, within which a directory tree mirroring the code
tree will be created. This cache tree is then used (for both reading and writing)
instead of the local `__pycache__` subdirectory within each source directory.

Exposed at runtime as sys.pycache_prefix (defaulting to None), and can
be set from the CLI as "-X pycache_prefix=path".

Patch by Carl Meyer.
diff --git a/Doc/library/compileall.rst b/Doc/library/compileall.rst
@@ -109,6 +109,11 @@ There is no command-line option to control the optimization level used by the
 :func:`compile` function, because the Python interpreter itself already
 provides the option: :program:`python -O -m compileall`.
 
+Similarly, the :func:`compile` function respects the :attr:`sys.pycache_prefix`
+setting. The generated bytecode cache will only be useful if :func:`compile` is
+run with the same :attr:`sys.pycache_prefix` (if any) that will be used at
+runtime.
+
 Public functions
 ----------------
 
diff --git a/Doc/library/sys.rst b/Doc/library/sys.rst
@@ -209,6 +209,26 @@ always available.
    yourself to control bytecode file generation.
 
 
+.. data:: pycache_prefix
+
+   If this is set (not ``None``), Python will write bytecode-cache ``.pyc``
+   files to (and read them from) a parallel directory tree rooted at this
+   directory, rather than from ``__pycache__`` directories in the source code
+   tree. Any ``__pycache__`` directories in the source code tree will be ignored
+   and new `.pyc` files written within the pycache prefix. Thus if you use
+   :mod:`compileall` as a pre-build step, you must ensure you run it with the
+   same pycache prefix (if any) that you will use at runtime.
+
+   A relative path is interpreted relative to the current working directory.
+
+   This value is initially set based on the value of the :option:`-X`
+   ``pycache_prefix=PATH`` command-line option or the
+   :envvar:`PYTHONPYCACHEPREFIX` environment variable (command-line takes
+   precedence). If neither are set, it is ``None``.
+
+   .. versionadded:: 3.8
+
+
 .. function:: excepthook(type, value, traceback)
 
    This function prints out a given traceback and exception to ``sys.stderr``.
diff --git a/Doc/using/cmdline.rst b/Doc/using/cmdline.rst
@@ -442,6 +442,9 @@ Miscellaneous options
      the default locale-aware mode. ``-X utf8=0`` explicitly disables UTF-8
      mode (even when it would otherwise activate automatically).
      See :envvar:`PYTHONUTF8` for more details.
+   * ``-X pycache_prefix=PATH`` enables writing ``.pyc`` files to a parallel
+     tree rooted at the given directory instead of to the code tree. See also
+     :envvar:`PYTHONPYCACHEPREFIX`.
 
    It also allows passing arbitrary values and retrieving them through the
    :data:`sys._xoptions` dictionary.
@@ -461,6 +464,9 @@ Miscellaneous options
    .. versionadded:: 3.7
       The ``-X importtime``, ``-X dev`` and ``-X utf8`` options.
 
+   .. versionadded:: 3.8
+      The ``-X pycache_prefix`` option.
+
 
 Options you shouldn't use
 ~~~~~~~~~~~~~~~~~~~~~~~~~
@@ -587,6 +593,16 @@ conflict.
    specifying the :option:`-B` option.
 
 
+.. envvar:: PYTHONPYCACHEPREFIX
+
+   If this is set, Python will write ``.pyc`` files in a mirror directory tree
+   at this path, instead of in ``__pycache__`` directories within the source
+   tree. This is equivalent to specifying the :option:`-X`
+   ``pycache_prefix=PATH`` option.
+
+   .. versionadded:: 3.8
+
+
 .. envvar:: PYTHONHASHSEED
 
    If this variable is not set or set to ``random``, a random value is used
diff --git a/Include/pystate.h b/Include/pystate.h
@@ -1,4 +1,3 @@
-
 /* Thread and interpreter state structures and their interfaces */
 
 
@@ -44,6 +43,7 @@ typedef struct {
     int coerce_c_locale;    /* PYTHONCOERCECLOCALE, -1 means unknown */
     int coerce_c_locale_warn; /* PYTHONCOERCECLOCALE=warn */
     int utf8_mode;          /* PYTHONUTF8, -X utf8; -1 means unknown */
+    wchar_t *pycache_prefix; /* PYTHONPYCACHEPREFIX, -X pycache_prefix=PATH */
 
     wchar_t *program_name;  /* Program name, see also Py_GetProgramName() */
     int argc;               /* Number of command line arguments,
@@ -101,6 +101,7 @@ typedef struct {
     PyObject *warnoptions;         /* sys.warnoptions list, can be NULL */
     PyObject *xoptions;            /* sys._xoptions dict, can be NULL */
     PyObject *module_search_path;  /* sys.path list */
+    PyObject *pycache_prefix;      /* sys.pycache_prefix str, can be NULL */
 } _PyMainInterpreterConfig;
 
 #define _PyMainInterpreterConfig_INIT \
diff --git a/Lib/importlib/_bootstrap_external.py b/Lib/importlib/_bootstrap_external.py
@@ -102,6 +102,15 @@ def _path_isdir(path):
     return _path_is_mode_type(path, 0o040000)
 
 
+def _path_isabs(path):
+    """Replacement for os.path.isabs.
+
+    Considers a Windows drive-relative path (no drive, but starts with slash) to
+    still be "absolute".
+    """
+    return path.startswith(path_separators) or path[1:3] in _pathseps_with_colon
+
+
 def _write_atomic(path, data, mode=0o666):
     """Best-effort function to write data to a path atomically.
     Be prepared to handle a FileExistsError if concurrent writing of the
@@ -312,7 +321,33 @@ def cache_from_source(path, debug_override=None, *, optimization=None):
         if not optimization.isalnum():
             raise ValueError('{!r} is not alphanumeric'.format(optimization))
         almost_filename = '{}.{}{}'.format(almost_filename, _OPT, optimization)
-    return _path_join(head, _PYCACHE, almost_filename + BYTECODE_SUFFIXES[0])
+    filename = almost_filename + BYTECODE_SUFFIXES[0]
+    if sys.pycache_prefix is not None:
+        # We need an absolute path to the py file to avoid the possibility of
+        # collisions within sys.pycache_prefix, if someone has two different
+        # `foo/bar.py` on their system and they import both of them using the
+        # same sys.pycache_prefix. Let's say sys.pycache_prefix is
+        # `C:\Bytecode`; the idea here is that if we get `Foo\Bar`, we first
+        # make it absolute (`C:\Somewhere\Foo\Bar`), then make it root-relative
+        # (`Somewhere\Foo\Bar`), so we end up placing the bytecode file in an
+        # unambiguous `C:\Bytecode\Somewhere\Foo\Bar\`.
+        if not _path_isabs(head):
+            head = _path_join(_os.getcwd(), head)
+
+        # Strip initial drive from a Windows path. We know we have an absolute
+        # path here, so the second part of the check rules out a POSIX path that
+        # happens to contain a colon at the second character.
+        if head[1] == ':' and head[0] not in path_separators:
+            head = head[2:]
+
+        # Strip initial path separator from `head` to complete the conversion
+        # back to a root-relative path before joining.
+        return _path_join(
+            sys.pycache_prefix,
+            head.lstrip(path_separators),
+            filename,
+        )
+    return _path_join(head, _PYCACHE, filename)
 
 
 def source_from_cache(path):
@@ -328,23 +363,29 @@ def source_from_cache(path):
         raise NotImplementedError('sys.implementation.cache_tag is None')
     path = _os.fspath(path)
     head, pycache_filename = _path_split(path)
-    head, pycache = _path_split(head)
-    if pycache != _PYCACHE:
-        raise ValueError('{} not bottom-level directory in '
-                         '{!r}'.format(_PYCACHE, path))
+    found_in_pycache_prefix = False
+    if sys.pycache_prefix is not None:
+        stripped_path = sys.pycache_prefix.rstrip(path_separators)
+        if head.startswith(stripped_path + path_sep):
+            head = head[len(stripped_path):]
+            found_in_pycache_prefix = True
+    if not found_in_pycache_prefix:
+        head, pycache = _path_split(head)
+        if pycache != _PYCACHE:
+            raise ValueError(f'{_PYCACHE} not bottom-level directory in '
+                             f'{path!r}')
     dot_count = pycache_filename.count('.')
     if dot_count not in {2, 3}:
-        raise ValueError('expected only 2 or 3 dots in '
-                         '{!r}'.format(pycache_filename))
+        raise ValueError(f'expected only 2 or 3 dots in {pycache_filename!r}')
     elif dot_count == 3:
         optimization = pycache_filename.rsplit('.', 2)[-2]
         if not optimization.startswith(_OPT):
             raise ValueError("optimization portion of filename does not start "
-                             "with {!r}".format(_OPT))
+                             f"with {_OPT!r}")
         opt_level = optimization[len(_OPT):]
         if not opt_level.isalnum():
-            raise ValueError("optimization level {!r} is not an alphanumeric "
-                             "value".format(optimization))
+            raise ValueError(f"optimization level {optimization!r} is not an "
+                             "alphanumeric value")
     base_filename = pycache_filename.partition('.')[0]
     return _path_join(head, base_filename + SOURCE_SUFFIXES[0])
 
@@ -1533,6 +1574,7 @@ def _setup(_bootstrap_module):
     setattr(self_module, '_os', os_module)
     setattr(self_module, 'path_sep', path_sep)
     setattr(self_module, 'path_separators', ''.join(path_separators))
+    setattr(self_module, '_pathseps_with_colon', {f':{s}' for s in path_separators})
 
     # Directly load the _thread module (needed during bootstrap).
     thread_module = _bootstrap._builtin_from_name('_thread')
diff --git a/Lib/test/test_cmd_line.py b/Lib/test/test_cmd_line.py
@@ -519,6 +519,32 @@ def test_sys_flags_set(self):
             with self.subTest(envar_value=value):
                 assert_python_ok('-c', code, **env_vars)
 
+    def test_set_pycache_prefix(self):
+        # sys.pycache_prefix can be set from either -X pycache_prefix or
+        # PYTHONPYCACHEPREFIX env var, with the former taking precedence.
+        NO_VALUE = object()  # `-X pycache_prefix` with no `=PATH`
+        cases = [
+            # (PYTHONPYCACHEPREFIX, -X pycache_prefix, sys.pycache_prefix)
+            (None, None, None),
+            ('foo', None, 'foo'),
+            (None, 'bar', 'bar'),
+            ('foo', 'bar', 'bar'),
+            ('foo', '', None),
+            ('foo', NO_VALUE, None),
+        ]
+        for envval, opt, expected in cases:
+            exp_clause = "is None" if expected is None else f'== "{expected}"'
+            code = f"import sys; sys.exit(not sys.pycache_prefix {exp_clause})"
+            args = ['-c', code]
+            env = {} if envval is None else {'PYTHONPYCACHEPREFIX': envval}
+            if opt is NO_VALUE:
+                args[:0] = ['-X', 'pycache_prefix']
+            elif opt is not None:
+                args[:0] = ['-X', f'pycache_prefix={opt}']
+            with self.subTest(envval=envval, opt=opt):
+                with support.temp_cwd():
+                    assert_python_ok(*args, **env)
+
     def run_xdev(self, *args, check_exitcode=True, xdev=True):
         env = dict(os.environ)
         env.pop('PYTHONWARNINGS', None)
diff --git a/Lib/test/test_importlib/test_util.py b/Lib/test/test_importlib/test_util.py
@@ -1,9 +1,10 @@
-from . import util
+﻿from . import util
 abc = util.import_importlib('importlib.abc')
 init = util.import_importlib('importlib')
 machinery = util.import_importlib('importlib.machinery')
 importlib_util = util.import_importlib('importlib.util')
 
+import contextlib
 import importlib.util
 import os
 import pathlib
@@ -12,6 +13,7 @@
 from test import support
 import types
 import unittest
+import unittest.mock
 import warnings
 
 
@@ -557,8 +559,8 @@ class PEP3147Tests:
 
     tag = sys.implementation.cache_tag
 
-    @unittest.skipUnless(sys.implementation.cache_tag is not None,
-                         'requires sys.implementation.cache_tag not be None')
+    @unittest.skipIf(sys.implementation.cache_tag is None,
+                     'requires sys.implementation.cache_tag not be None')
     def test_cache_from_source(self):
         # Given the path to a .py file, return the path to its PEP 3147
         # defined .pyc file (i.e. under __pycache__).
@@ -678,18 +680,17 @@ def test_sep_altsep_and_sep_cache_from_source(self):
             self.util.cache_from_source('\\foo\\bar\\baz/qux.py', optimization=''),
             '\\foo\\bar\\baz\\__pycache__\\qux.{}.pyc'.format(self.tag))
 
-    @unittest.skipUnless(sys.implementation.cache_tag is not None,
-                         'requires sys.implementation.cache_tag not be None')
+    @unittest.skipIf(sys.implementation.cache_tag is None,
+                     'requires sys.implementation.cache_tag not be None')
     def test_source_from_cache_path_like_arg(self):
         path = pathlib.PurePath('foo', 'bar', 'baz', 'qux.py')
         expect = os.path.join('foo', 'bar', 'baz', '__pycache__',
                               'qux.{}.pyc'.format(self.tag))
         self.assertEqual(self.util.cache_from_source(path, optimization=''),
                          expect)
 
-    @unittest.skipUnless(sys.implementation.cache_tag is not None,
-                         'requires sys.implementation.cache_tag to not be '
-                         'None')
+    @unittest.skipIf(sys.implementation.cache_tag is None,
+                     'requires sys.implementation.cache_tag to not be None')
     def test_source_from_cache(self):
         # Given the path to a PEP 3147 defined .pyc file, return the path to
         # its source.  This tests the good path.
@@ -749,15 +750,87 @@ def test_source_from_cache_missing_optimization(self):
         with self.assertRaises(ValueError):
             self.util.source_from_cache(path)
 
-    @unittest.skipUnless(sys.implementation.cache_tag is not None,
-                         'requires sys.implementation.cache_tag to not be '
-                         'None')
+    @unittest.skipIf(sys.implementation.cache_tag is None,
+                     'requires sys.implementation.cache_tag to not be None')
     def test_source_from_cache_path_like_arg(self):
         path = pathlib.PurePath('foo', 'bar', 'baz', '__pycache__',
                                 'qux.{}.pyc'.format(self.tag))
         expect = os.path.join('foo', 'bar', 'baz', 'qux.py')
         self.assertEqual(self.util.source_from_cache(path), expect)
 
+    @unittest.skipIf(sys.implementation.cache_tag is None,
+                     'requires sys.implementation.cache_tag to not be None')
+    def test_cache_from_source_respects_pycache_prefix(self):
+        # If pycache_prefix is set, cache_from_source will return a bytecode
+        # path inside that directory (in a subdirectory mirroring the .py file's
+        # path) rather than in a __pycache__ dir next to the py file.
+        pycache_prefixes = [
+            os.path.join(os.path.sep, 'tmp', 'bytecode'),
+            os.path.join(os.path.sep, 'tmp', '\u2603'),  # non-ASCII in path!
+            os.path.join(os.path.sep, 'tmp', 'trailing-slash') + os.path.sep,
+        ]
+        drive = ''
+        if os.name == 'nt':
+            drive = 'C:'
+            pycache_prefixes = [
+                f'{drive}{prefix}' for prefix in pycache_prefixes]
+            pycache_prefixes += [r'\\?\C:\foo', r'\\localhost\c$\bar']
+        for pycache_prefix in pycache_prefixes:
+            with self.subTest(path=pycache_prefix):
+                path = drive + os.path.join(
+                    os.path.sep, 'foo', 'bar', 'baz', 'qux.py')
+                expect = os.path.join(
+                    pycache_prefix, 'foo', 'bar', 'baz',
+                    'qux.{}.pyc'.format(self.tag))
+                with util.temporary_pycache_prefix(pycache_prefix):
+                    self.assertEqual(
+                        self.util.cache_from_source(path, optimization=''),
+                        expect)
+
+    @unittest.skipIf(sys.implementation.cache_tag is None,
+                     'requires sys.implementation.cache_tag to not be None')
+    def test_cache_from_source_respects_pycache_prefix_relative(self):
+        # If the .py path we are given is relative, we will resolve to an
+        # absolute path before prefixing with pycache_prefix, to avoid any
+        # possible ambiguity.
+        pycache_prefix = os.path.join(os.path.sep, 'tmp', 'bytecode')
+        path = os.path.join('foo', 'bar', 'baz', 'qux.py')
+        root = os.path.splitdrive(os.getcwd())[0] + os.path.sep
+        expect = os.path.join(
+            pycache_prefix,
+            os.path.relpath(os.getcwd(), root),
+            'foo', 'bar', 'baz', f'qux.{self.tag}.pyc')
+        with util.temporary_pycache_prefix(pycache_prefix):
+            self.assertEqual(
+                self.util.cache_from_source(path, optimization=''),
+                expect)
+
+    @unittest.skipIf(sys.implementation.cache_tag is None,
+                     'requires sys.implementation.cache_tag to not be None')
+    def test_source_from_cache_inside_pycache_prefix(self):
+        # If pycache_prefix is set and the cache path we get is inside it,
+        # we return an absolute path to the py file based on the remainder of
+        # the path within pycache_prefix.
+        pycache_prefix = os.path.join(os.path.sep, 'tmp', 'bytecode')
+        path = os.path.join(pycache_prefix, 'foo', 'bar', 'baz',
+                            f'qux.{self.tag}.pyc')
+        expect = os.path.join(os.path.sep, 'foo', 'bar', 'baz', 'qux.py')
+        with util.temporary_pycache_prefix(pycache_prefix):
+            self.assertEqual(self.util.source_from_cache(path), expect)
+
+    @unittest.skipIf(sys.implementation.cache_tag is None,
+                     'requires sys.implementation.cache_tag to not be None')
+    def test_source_from_cache_outside_pycache_prefix(self):
+        # If pycache_prefix is set but the cache path we get is not inside
+        # it, just ignore it and handle the cache path according to the default
+        # behavior.
+        pycache_prefix = os.path.join(os.path.sep, 'tmp', 'bytecode')
+        path = os.path.join('foo', 'bar', 'baz', '__pycache__',
+                            f'qux.{self.tag}.pyc')
+        expect = os.path.join('foo', 'bar', 'baz', 'qux.py')
+        with util.temporary_pycache_prefix(pycache_prefix):
+            self.assertEqual(self.util.source_from_cache(path), expect)
+
 
 (Frozen_PEP3147Tests,
  Source_PEP3147Tests
diff --git a/Lib/test/test_importlib/util.py b/Lib/test/test_importlib/util.py
@@ -319,6 +319,17 @@ def ensure_bytecode_path(bytecode_path):
             raise
 
 
+@contextlib.contextmanager
+def temporary_pycache_prefix(prefix):
+    """Adjust and restore sys.pycache_prefix."""
+    _orig_prefix = sys.pycache_prefix
+    sys.pycache_prefix = prefix
+    try:
+        yield
+    finally:
+        sys.pycache_prefix = _orig_prefix
+
+
 @contextlib.contextmanager
 def create_modules(*names):
     """Temporarily create each named module with an attribute (named 'attr')
diff --git a/Misc/NEWS.d/next/Core and Builtins/2018-05-15-10-48-47.bpo-33499.uBEc06.rst b/Misc/NEWS.d/next/Core and Builtins/2018-05-15-10-48-47.bpo-33499.uBEc06.rst
@@ -0,0 +1,3 @@
+Add :envvar:`PYTHONPYCACHEPREFIX` environment variable and :option:`-X`
+``pycache_prefix`` command-line option to set an alternate root directory for
+writing module bytecode cache files.
diff --git a/Modules/main.c b/Modules/main.c
diff --git a/Python/importlib_external.h b/Python/importlib_external.h
diff --git a/Python/sysmodule.c b/Python/sysmodule.c

Original file line number	Diff line number	Diff line change
`@@ -0,0 +1,3 @@`
	`1`	+Add :envvar:`PYTHONPYCACHEPREFIX` environment variable and :option:`-X`
	`2`	+``pycache_prefix`` command-line option to set an alternate root directory for
	`3`	`+writing module bytecode cache files.`