Merge branch 'main' into pep750-expression-default

Author: davepeck

Lib/random.py

@@ -844,8 +844,8 @@ def binomialvariate(self, n=1, p=0.5):
         # BTRS: Transformed rejection with squeeze method by Wolfgang Hörmann
         # https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.47.8407&rep=rep1&type=pdf
         assert n*p >= 10.0 and p <= 0.5
-        setup_complete = False
 
+        setup_complete = False
         spq = _sqrt(n * p * (1.0 - p))  # Standard deviation of the distribution
         b = 1.15 + 2.53 * spq
         a = -0.0873 + 0.0248 * b + 0.01 * p
@@ -860,22 +860,23 @@ def binomialvariate(self, n=1, p=0.5):
             k = _floor((2.0 * a / us + b) * u + c)
             if k < 0 or k > n:
                 continue
+            v = random()
 
             # The early-out "squeeze" test substantially reduces
             # the number of acceptance condition evaluations.
-            v = random()
             if us >= 0.07 and v <= vr:
                 return k
 
-            # Acceptance-rejection test.
-            # Note, the original paper erroneously omits the call to log(v)
-            # when comparing to the log of the rescaled binomial distribution.
             if not setup_complete:
                 alpha = (2.83 + 5.1 / b) * spq
                 lpq = _log(p / (1.0 - p))
                 m = _floor((n + 1) * p)         # Mode of the distribution
                 h = _lgamma(m + 1) + _lgamma(n - m + 1)
                 setup_complete = True           # Only needs to be done once
+
+            # Acceptance-rejection test.
+            # Note, the original paper erroneously omits the call to log(v)
+            # when comparing to the log of the rescaled binomial distribution.
             v *= alpha / (a / (us * us) + b)
             if _log(v) <= h - _lgamma(k + 1) - _lgamma(n - k + 1) + (k - m) * lpq:
                 return k

Tools/jit/_targets.py

@@ -137,7 +137,15 @@ async def _compile(
             f"-I{CPYTHON / 'Include' / 'internal' / 'mimalloc'}",
             f"-I{CPYTHON / 'Python'}",
             f"-I{CPYTHON / 'Tools' / 'jit'}",
-            "-O3",
+            # -O2 and -O3 include some optimizations that make sense for
+            # standalone functions, but not for snippets of code that are going
+            # to be laid out end-to-end (like ours)... common examples include
+            # passes like tail-duplication, or aligning jump targets with nops.
+            # -Os is equivalent to -O2 with many of these problematic passes
+            # disabled. Based on manual review, for *our* purposes it usually
+            # generates better code than -O2 (and -O2 usually generates better
+            # code than -O3). As a nice benefit, it uses less memory too:
+            "-Os",
             "-S",
             # Shorten full absolute file paths in the generated code (like the
             # __FILE__ macro and assert failure messages) for reproducibility: