Fix #5499: Include in pip's User-Agent whether it looks like pip is in CI

news/5499.feature

@@ -0,0 +1,2 @@
+Include in pip's User-Agent string whether it looks like pip is running
+under CI.

src/pip/_internal/download.py

@@ -72,6 +72,34 @@
 logger = logging.getLogger(__name__)
 
 
+# These are environment variables present when running under various
+# CI systems.  For each variable, some CI systems that use the variable
+# are indicated.  The collection was chosen so that for each of a number
+# of popular systems, at least one of the environment variables is used.
+# This list is used to provide some indication of and lower bound for
+# CI traffic to PyPI.  Thus, it is okay if the list is not comprehensive.
+# For more background, see: https://github.com/pypa/pip/issues/5499
+CI_ENVIRONMENT_VARIABLES = (
+    # Azure Pipelines
+    'BUILD_BUILDID',
+    # Jenkins
+    'BUILD_ID',
+    # AppVeyor, CircleCI, Codeship, Gitlab CI, Shippable, Travis CI
+    'CI',
+)
+
+
+def looks_like_ci():
+    # type: () -> bool
+    """
+    Return whether it looks like pip is running under CI.
+    """
+    # We don't use the method of checking for a tty (e.g. using isatty())
+    # because some CI systems mimic a tty (e.g. Travis CI).  Thus that
+    # method doesn't provide definitive information in either direction.
+    return any(name in os.environ for name in CI_ENVIRONMENT_VARIABLES)
+
+
 def user_agent():
     """
     Return a string representing the user agent.
@@ -135,6 +163,12 @@ def user_agent():
     if setuptools_version is not None:
         data["setuptools_version"] = setuptools_version
 
+    # Use None rather than False so as not to give the impression that
+    # pip knows it is not being run under CI.  Rather, it is a null or
+    # inconclusive result.  Also, we include some value rather than no
+    # value to make it easier to know that the check has been run.
+    data["ci"] = True if looks_like_ci() else None
+
     return "{data[installer][name]}/{data[installer][version]} {json}".format(
         data=data,
         json=json.dumps(data, separators=(",", ":"), sort_keys=True),

tests/unit/test_download.py

@@ -10,8 +10,8 @@
 
 import pip
 from pip._internal.download import (
-    PipSession, SafeFileCache, path_to_url, unpack_file_url, unpack_http_url,
-    url_to_path,
+    CI_ENVIRONMENT_VARIABLES, PipSession, SafeFileCache, path_to_url,
+    unpack_file_url, unpack_http_url, url_to_path,
 )
 from pip._internal.exceptions import HashMismatch
 from pip._internal.models.link import Link
@@ -50,8 +50,41 @@ def _fake_session_get(*args, **kwargs):
         rmtree(temp_dir)
 
 
+def get_user_agent():
+    return PipSession().headers["User-Agent"]
+
+
 def test_user_agent():
-    PipSession().headers["User-Agent"].startswith("pip/%s" % pip.__version__)
+    user_agent = get_user_agent()
+
+    assert user_agent.startswith("pip/%s" % pip.__version__)
+
+
+@pytest.mark.parametrize('name, expected_like_ci', [
+    ('BUILD_BUILDID', True),
+    ('BUILD_ID', True),
+    ('CI', True),
+    # Test a prefix substring of one of the variable names we use.
+    ('BUILD', False),
+])
+def test_user_agent__ci(monkeypatch, name, expected_like_ci):
+    # Delete the variable names we use to check for CI to prevent the
+    # detection from always returning True in case the tests are being run
+    # under actual CI.  It is okay to depend on CI_ENVIRONMENT_VARIABLES
+    # here (part of the code under test) because this setup step can only
+    # prevent false test failures.  It can't cause a false test passage.
+    for ci_name in CI_ENVIRONMENT_VARIABLES:
+        monkeypatch.delenv(ci_name, raising=False)
+
+    # Confirm the baseline before setting the environment variable.
+    user_agent = get_user_agent()
+    assert '"ci":null' in user_agent
+    assert '"ci":true' not in user_agent
+
+    monkeypatch.setenv(name, 'true')
+    user_agent = get_user_agent()
+    assert ('"ci":true' in user_agent) == expected_like_ci
+    assert ('"ci":null' in user_agent) == (not expected_like_ci)
 
 
 class FakeStream(object):