POC: dynamically choose which asvs to run

scripts/choose_asvs.py

@@ -0,0 +1,132 @@
+"""
+Try to identify which asvs can be ignored for a PR.
+"""
+import os
+import shlex
+import subprocess
+
+here = os.path.dirname(__file__)  # assumes repo directory
+bmark_dir = os.path.join("asv_bench", "benchmarks")
+
+
+def find_changed():
+    """
+    Find the names of files changes (compared to master) in this branch.
+    """
+    cmd = "git diff upstream/master --name-only"
+
+    p = subprocess.Popen(
+        shlex.split(cmd),
+        stdout=subprocess.PIPE,
+        stderr=subprocess.PIPE,
+        close_fds=True,
+    )
+    stdout, stderr = p.communicate()
+
+    flist = stdout.decode("utf-8").splitlines()
+    return flist
+
+
+def find_asv_paths():
+    """
+    Find all of our asv benchmark files.
+    """
+    bmarks = os.path.join(here, bmark_dir)
+    walk = os.scandir(bmarks)
+    walk = [
+        x
+        for x in walk
+        if x.is_file()
+        and x.name.endswith(".py")
+        and x.name not in ["__init__.py", "pandas_vb_common.py"]
+    ]
+
+    paths = [x.path.replace(bmarks, "").lstrip(os.sep) for x in walk]
+    return paths
+
+
+def trim_asv_paths():
+    """
+    Exclude benchmark files we can be confident are not affected
+    by this PR.
+    """
+    flist = find_changed()
+
+    changed = [x for x in flist if "pandas/" in x and "pandas/tests/" not in x]
+
+    asv_paths = find_asv_paths()
+
+    if "setup.py" in flist:
+        # We can't exclude anything
+        return asv_paths
+    if any("_libs/src" in x for x in changed):
+        # We can't exclude anything
+        return asv_paths
+    if not any("_libs/tslibs" in x for x in changed):
+        # If nothing in tslibs changed, then none of the tslibs asvs should
+        #  be affected
+        asv_paths = [x for x in asv_paths if not x.startswith("tslibs/")]
+    if not any("_libs/" in x for x in changed):
+        asv_paths = [
+            x for x in asv_paths if x not in ["libs.py", "indexing_engines.py"]
+        ]
+        # TODO:
+        #  inference.MaybeConvertNumeric
+        #  dtypes.InferDtypes
+        #  algorithms.MaybeConvertObjects  # (not 100% disjoint)
+    return asv_paths
+
+
+def path_to_module(path: str) -> str:
+    """
+    >>> path = "asv_bench/benchmarks/ctors.py"
+    >>> path_to_module(path)
+    'ctors'
+
+    >>> path = "asv_bench/benchmarks/tslibs/fields.py"
+    >>> path_to_module(path)
+    'tslibs.fields'
+    """
+    assert path.startswith(bmark_dir)
+    assert path.endswith(".py")
+
+    name = path.replace(bmark_dir, "").strip(os.sep)
+    name = name[:-3]
+    name = name.replace(os.sep, ".")
+    return name
+
+
+def run_relevant_asvs():
+    paths = trim_asv_paths()
+
+    names = [path_to_module(x) for x in paths]
+
+    pattern = "-b ".join(names)
+    run_asvs(pattern)
+
+
+def run_asvs(pattern: str):
+
+    cmd = (
+        "asv continuous "
+        "-E virtualenv "
+        "-f 1.05 "
+        "--record-samples --append-samples "
+        f"master HEAD -b {pattern}"
+    )
+
+    os.chdir("asv_bench")
+    try:
+        p = subprocess.Popen(
+            shlex.split(cmd),
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+            close_fds=True,
+        )
+        stdout, stderr = p.communicate()
+    finally:
+        os.chdir(here)
+
+    stdout = stdout.decode("utf-8")
+    stderr = stderr.decode("utf-8")
+    return stdout, stderr