Attempt to speed up deepfreeze.py (#107887)

* Instead of calling get_identifiers_and_strings(), extract identifiers and strings from pycore_global_strings.h.
* Avoid ast.literal_eval(), it's very slow.

Makefile.pre.in

@@ -1253,7 +1253,7 @@ regen-frozen: Tools/build/freeze_modules.py $(FROZEN_FILES_IN)
 .PHONY: regen-deepfreeze
 regen-deepfreeze: $(DEEPFREEZE_OBJS)
 
-DEEPFREEZE_DEPS=$(srcdir)/Tools/build/deepfreeze.py $(FREEZE_MODULE_DEPS) $(FROZEN_FILES_OUT)
+DEEPFREEZE_DEPS=$(srcdir)/Tools/build/deepfreeze.py Include/internal/pycore_global_strings.h $(FREEZE_MODULE_DEPS) $(FROZEN_FILES_OUT)
 
 # BEGIN: deepfreeze modules
 Python/deepfreeze/deepfreeze.c: $(DEEPFREEZE_DEPS)

Tools/build/deepfreeze.py

@@ -6,7 +6,6 @@
 by Python 3.10, and 3.11 features are not available.
 """
 import argparse
-import ast
 import builtins
 import collections
 import contextlib
@@ -17,10 +16,10 @@
 from typing import Dict, FrozenSet, TextIO, Tuple
 
 import umarshal
-from generate_global_objects import get_identifiers_and_strings
+
+ROOT = os.path.dirname(os.path.dirname(os.path.dirname(__file__)))
 
 verbose = False
-identifiers, strings = get_identifiers_and_strings()
 
 # This must be kept in sync with opcode.py
 RESUME = 151
@@ -114,13 +113,27 @@ def __init__(self, file: TextIO) -> None:
         self.hits, self.misses = 0, 0
         self.finis: list[str] = []
         self.inits: list[str] = []
+        self.identifiers, self.strings = self.get_identifiers_and_strings()
         self.write('#include "Python.h"')
         self.write('#include "internal/pycore_gc.h"')
         self.write('#include "internal/pycore_code.h"')
         self.write('#include "internal/pycore_frame.h"')
         self.write('#include "internal/pycore_long.h"')
         self.write("")
 
+    def get_identifiers_and_strings(self) -> tuple[set[str], dict[str, str]]:
+        filename = os.path.join(ROOT, "Include", "internal", "pycore_global_strings.h")
+        with open(filename) as fp:
+            lines = fp.readlines()
+        identifiers: set[str] = set()
+        strings: dict[str, str] = {}
+        for line in lines:
+            if m := re.search(r"STRUCT_FOR_ID\((\w+)\)", line):
+                identifiers.add(m.group(1))
+            if m := re.search(r'STRUCT_FOR_STR\((\w+), "(.*?)"\)', line):
+                strings[m.group(2)] = m.group(1)
+        return identifiers, strings
+
     @contextlib.contextmanager
     def indent(self) -> None:
         save_level = self.level
@@ -171,9 +184,9 @@ def generate_bytes(self, name: str, b: bytes) -> str:
         return f"& {name}.ob_base.ob_base"
 
     def generate_unicode(self, name: str, s: str) -> str:
-        if s in strings:
-            return f"&_Py_STR({strings[s]})"
-        if s in identifiers:
+        if s in self.strings:
+            return f"&_Py_STR({self.strings[s]})"
+        if s in self.identifiers:
             return f"&_Py_ID({s})"
         if len(s) == 1:
             c = ord(s)
@@ -441,12 +454,10 @@ def is_frozen_header(source: str) -> bool:
 
 
 def decode_frozen_data(source: str) -> types.CodeType:
-    lines = source.splitlines()
-    while lines and re.match(FROZEN_DATA_LINE, lines[0]) is None:
-        del lines[0]
-    while lines and re.match(FROZEN_DATA_LINE, lines[-1]) is None:
-        del lines[-1]
-    values: Tuple[int, ...] = ast.literal_eval("".join(lines).strip())
+    values: list[int] = []
+    for line in source.splitlines():
+        if re.match(FROZEN_DATA_LINE, line):
+            values.extend([int(x) for x in line.split(",") if x.strip()])
     data = bytes(values)
     return umarshal.loads(data)