MAINT: Add functions to parse shell-strings in the platform-native way

There are places in distutils where we accept a single string from the user, and interpret it as a set of command line arguments.
Previously, these were passed on as a string unmodified to exec_command, and interpreted by subprocess in a platform-specific way.

Recent changes to distutils now pass a list of arguments to subprocess, meaning we have to split the strings ourselves.
While `shlex.split` is perfect on posix systems, it is not a good approximation of either the old or the expected behavior on windows.

Provides the building blocks needed to fix gh-12979

1 files changed, 79 insertions, 0 deletions

diff --git a/numpy/distutils/tests/test_shell_utils.py b/numpy/distutils/tests/test_shell_utils.py
new file mode 100644
index 000000000..a0344244f
--- /dev/null
+++ b/numpy/distutils/tests/test_shell_utils.py
@@ -0,0 +1,79 @@
+from __future__ import division, absolute_import, print_function
+
+import pytest
+import subprocess
+import os
+import json
+import sys
+
+from numpy.distutils import _shell_utils
+
+argv_cases = [
+    [r'exe'],
+    [r'path/exe'],
+    [r'path\exe'],
+    [r'\\server\path\exe'],
+    [r'path to/exe'],
+    [r'path to\exe'],
+
+    [r'exe', '--flag'],
+    [r'path/exe', '--flag'],
+    [r'path\exe', '--flag'],
+    [r'path to/exe', '--flag'],
+    [r'path to\exe', '--flag'],
+
+    # flags containing literal quotes in their name
+    [r'path to/exe', '--flag-"quoted"'],
+    [r'path to\exe', '--flag-"quoted"'],
+    [r'path to/exe', '"--flag-quoted"'],
+    [r'path to\exe', '"--flag-quoted"'],
+]
+
+
+@pytest.fixture(params=[
+    _shell_utils.WindowsParser,
+    _shell_utils.PosixParser
+])
+def Parser(request):
+    return request.param
+
+
+@pytest.fixture
+def runner(Parser):
+    if Parser != _shell_utils.NativeParser:
+        pytest.skip('Unable to run with non-native parser')
+
+    if Parser == _shell_utils.WindowsParser:
+        return lambda cmd: subprocess.check_output(cmd)
+    elif Parser == _shell_utils.PosixParser:
+        # posix has no non-shell string parsing
+        return lambda cmd: subprocess.check_output(cmd, shell=True)
+    else:
+        raise NotImplementedError
+
+
+@pytest.mark.parametrize('argv', argv_cases)
+def test_join_matches_subprocess(Parser, runner, argv):
+    """
+    Test that join produces strings understood by subprocess
+    """
+    # invoke python to return its arguments as json
+    cmd = [
+        sys.executable, '-c',
+        'import json, sys; print(json.dumps(sys.argv[1:]))'
+    ]
+    joined = Parser.join(cmd + argv)
+    json_out = runner(joined).decode()
+    assert json.loads(json_out) == argv
+
+
+@pytest.mark.parametrize('argv', argv_cases)
+def test_roundtrip(Parser, argv):
+    """
+    Test that split is the inverse operation of join
+    """
+    try:
+        joined = Parser.join(argv)
+        assert argv == Parser.split(joined)
+    except NotImplementedError:
+        pytest.skip("Not implemented")