Reproduction for sentry-python#7346

sentry-python/7346/.gitignore

@@ -0,0 +1,3 @@
+.venv/
+uv.lock
+__pycache__/

sentry-python/7346/README.md

@@ -0,0 +1,87 @@
+# Reproduction for sentry-python#7346
+
+**Issue:** https://github.com/getsentry/sentry-python/issues/7346
+
+## Description
+
+`_experiments={"data_collection": None}` type checks, because `data_collection` is
+typed `Optional[DataCollectionUserOptions]`, but the two functions that read the
+option disagree about what `None` means:
+
+- `has_data_collection_enabled()` (`sentry_sdk/utils.py:2108`) gates on **key
+  presence**: `"data_collection" in options.get("_experiments", {})` → `True`
+- `_resolve_data_collection()` (`sentry_sdk/data_collection.py:287`) only treats a
+  **non-`None`** value as user provided: `if user_dc is not None` → `provided_by_user=False`
+
+So `data_collection=None` lands in a state where neither side of the option is
+active: every caller of `has_data_collection_enabled()` takes the data collection
+branch, while the resolved config is the legacy `send_default_pii=False` mapping.
+
+## Steps to Reproduce
+
+```bash
+uv run repro.py
+```
+
+The script runs against this checkout (`[tool.uv.sources]` points at `../..`),
+because `data_collection` is an unreleased `_experiments` option. No DSN is
+needed — events are inspected in `before_send`. Set `export SENTRY_DSN=...` if
+you also want them delivered.
+
+It exits `0` when the bug reproduces and `1` when it is fixed.
+
+## Expected Behavior
+
+`data_collection=None` should behave like one of its two neighbours, not fall
+between them:
+
+- the default `EventScrubber` stays installed, so `extra["password"]` is `[Filtered]`
+- an explicitly configured `event_scrubber` is kept, with no warning
+- WSGI request attributes attach no `http.query` / `url.path` / `url.full`, since
+  `should_send_default_pii()` is `False`
+
+## Actual Behavior
+
+```
+0. The two functions disagree
+config                                has_data_collection_enabled()  provided_by_user
+------------------------------------  -----------------------------  ----------------
+no _experiments (baseline)            False                          False
+data_collection={} (explicit opt in)  True                           True
+data_collection=None (the bug)        True                           False
+
+1. Event scrubbing
+config                                extra[password]  extra[token]  event_scrubber installed
+------------------------------------  ---------------  ------------  ------------------------
+no _experiments (baseline)            [Filtered]       [Filtered]    True
+data_collection={} (explicit opt in)  hunter2          abc123        False
+data_collection=None (the bug)        hunter2          abc123        False
+
+2. WSGI request attributes
+config                                http.query                     url.path   url.full
+------------------------------------  -----------------------------  ---------  ------------------------------------------------------------
+no _experiments (baseline)            None                           None       None
+data_collection={} (explicit opt in)  email=a%40b.com&coupon=SAVE10  /checkout  http://localhost:8000/checkout?email=a%40b.com&coupon=SAVE10
+data_collection=None (the bug)        email=a%40b.com&coupon=SAVE10  /checkout  http://localhost:8000/checkout?email=a%40b.com&coupon=SAVE10
+
+3. Explicitly configured event_scrubber
+   options["event_scrubber"] after init: None
+   warning: Event scrubbers are not enabled when data collection configuration is provided. Ignoring event_scrubber...
+
+Bug reproduced: True
+```
+
+## Note on the issue's scrubbing table
+
+The issue reports `extra["password"] == "[Filtered]"` for `data_collection={}`.
+On this checkout that config also yields `"hunter2"`, because skipping the
+`EventScrubber` is the intended behaviour whenever data collection is configured
+(`sentry_sdk/client.py:357`). That does not affect the report: the divergence is
+between `has_data_collection_enabled()` and `provided_by_user`, which section 0
+shows directly, and `data_collection=None` is the only config where they disagree.
+
+## Environment
+
+- Python: 3.14.5 (the script supports >= 3.9)
+- sentry-sdk: this checkout (2.69.1 at the time of writing)
+- OS: macOS

sentry-python/7346/pyproject.toml

@@ -0,0 +1,11 @@
+[project]
+name = "repro-sentry-python-7346"
+version = "0.1.0"
+description = "Reproduction for getsentry/sentry-python#7346"
+requires-python = ">=3.9"
+dependencies = ["sentry-sdk"]
+
+[tool.uv.sources]
+# `data_collection` is an unreleased `_experiments` option, so the
+# reproduction runs against this checkout rather than a PyPI release.
+sentry-sdk = { path = "../..", editable = true }

sentry-python/7346/repro.py

@@ -0,0 +1,208 @@
+"""
+Reproduction for getsentry/sentry-python#7346.
+
+`_experiments={"data_collection": None}` leaves the client in a state where
+neither side of the option is active:
+
+* `has_data_collection_enabled()` (sentry_sdk/utils.py) gates on **key
+  presence**, so it returns True.
+* `_resolve_data_collection()` (sentry_sdk/data_collection.py) only treats a
+  **non-None** value as user provided, so the resolved config falls back to the
+  legacy `send_default_pii` mapping with `provided_by_user=False`.
+
+Two observable consequences, each printed as a table below:
+
+1. The default `EventScrubber` is never installed, so `extra` values like
+   `password` are sent unfiltered.
+2. WSGI request attributes gain `http.query` / `url.path` / `url.full`, which
+   are not attached when `data_collection` is left unset.
+
+Set SENTRY_DSN if you also want the events delivered to a real project:
+
+    export SENTRY_DSN=...
+"""
+
+import os
+import warnings
+
+import sentry_sdk
+from sentry_sdk.integrations.wsgi import _get_request_attributes
+from sentry_sdk.scrubber import EventScrubber
+from sentry_sdk.utils import has_data_collection_enabled
+
+DSN = os.environ.get("SENTRY_DSN") or ""
+
+# Sentinel telling the repro to omit `_experiments` entirely, which is
+# different from passing `{"data_collection": None}`.
+UNSET = object()
+
+SCENARIOS = [
+    ("no _experiments (baseline)", UNSET),
+    ("data_collection={} (explicit opt in)", {}),
+    ("data_collection=None (the bug)", None),
+]
+
+WSGI_ENVIRON = {
+    "REQUEST_METHOD": "GET",
+    "PATH_INFO": "/checkout",
+    "QUERY_STRING": "email=a%40b.com&coupon=SAVE10",
+    "SERVER_NAME": "localhost",
+    "SERVER_PORT": "8000",
+    "wsgi.url_scheme": "http",
+}
+
+
+def init(data_collection):
+    experiments = (
+        {} if data_collection is UNSET else {"data_collection": data_collection}
+    )
+    captured = []
+
+    def before_send(event, hint):
+        # `before_send` runs after the event scrubber, so whatever lands here is
+        # what would have been sent to Sentry.
+        captured.append(event)
+        return None if not DSN else event
+
+    sentry_sdk.init(
+        dsn=DSN,
+        _experiments=experiments,
+        before_send=before_send,
+    )
+    return captured
+
+
+def scrubber_row(data_collection):
+    captured = init(data_collection)
+
+    scope = sentry_sdk.get_isolation_scope()
+    scope.set_extra("password", "hunter2")
+    scope.set_extra("token", "abc123")
+
+    sentry_sdk.capture_message("hi")
+    sentry_sdk.flush()
+
+    options = sentry_sdk.get_client().options
+    extra = captured[0].get("extra", {}) if captured else {}
+    return (
+        extra.get("password"),
+        extra.get("token"),
+        options["event_scrubber"] is not None,
+    )
+
+
+def gate_row(data_collection):
+    init(data_collection)
+    options = sentry_sdk.get_client().options
+    return (
+        has_data_collection_enabled(options),
+        options["data_collection"]["provided_by_user"],
+    )
+
+
+def request_attributes_row(data_collection):
+    init(data_collection)
+    attributes = _get_request_attributes(WSGI_ENVIRON)
+    return (
+        attributes.get("http.query"),
+        attributes.get("url.path"),
+        attributes.get("url.full"),
+    )
+
+
+def explicit_scrubber_warning():
+    """An explicitly configured `event_scrubber` is discarded with a warning
+    that says data collection configuration was provided, which it was not."""
+    with warnings.catch_warnings(record=True) as caught:
+        warnings.simplefilter("always")
+        sentry_sdk.init(
+            dsn=DSN,
+            _experiments={"data_collection": None},
+            event_scrubber=EventScrubber(denylist=["my_secret"]),
+        )
+    scrubber = sentry_sdk.get_client().options["event_scrubber"]
+    return scrubber, [str(w.message) for w in caught]
+
+
+def table(title, headers, rows):
+    print(f"\n{title}")
+    widths = [
+        max(len(str(row[i])) for row in [headers, *rows]) for i in range(len(headers))
+    ]
+    fmt = "  ".join("{:<%d}" % w for w in widths)
+    print(fmt.format(*headers))
+    print(fmt.format(*("-" * w for w in widths)))
+    for row in rows:
+        print(fmt.format(*(str(cell) for cell in row)))
+
+
+def main():
+    gate_rows = []
+    scrubber_rows = []
+    attribute_rows = []
+    for label, data_collection in SCENARIOS:
+        gate_rows.append((label, *gate_row(data_collection)))
+        password, token, has_scrubber = scrubber_row(data_collection)
+        scrubber_rows.append((label, password, token, has_scrubber))
+        attribute_rows.append((label, *request_attributes_row(data_collection)))
+
+    table(
+        "0. The two functions disagree",
+        ("config", "has_data_collection_enabled()", "provided_by_user"),
+        gate_rows,
+    )
+    print(
+        "\n   `data_collection=None` is the only config where the two disagree: "
+        "\n   every caller of `has_data_collection_enabled()` takes the data "
+        "\n   collection branch, but the resolved config is the legacy "
+        "\n   `send_default_pii` mapping."
+    )
+
+    table(
+        "1. Event scrubbing",
+        ("config", "extra[password]", "extra[token]", "event_scrubber installed"),
+        scrubber_rows,
+    )
+    print(
+        "\n   Expected: `data_collection=None` filters like the baseline, because "
+        "\n   `_resolve_data_collection` did not treat it as user provided."
+        "\n   Actual:   the scrubber is skipped and the raw secrets are sent."
+    )
+
+    table(
+        "2. WSGI request attributes",
+        ("config", "http.query", "url.path", "url.full"),
+        attribute_rows,
+    )
+    print(
+        "\n   Expected: `data_collection=None` attaches no query params, like the "
+        "\n   baseline, since `should_send_default_pii()` is False."
+        "\n   Actual:   the data collection branch is taken and the query string is "
+        "\n   attached."
+    )
+
+    scrubber, messages = explicit_scrubber_warning()
+    print("\n3. Explicitly configured event_scrubber")
+    print(f'   options["event_scrubber"] after init: {scrubber}')
+    for message in messages:
+        print(f"   warning: {message}")
+    print(
+        "\n   Expected: no warning, because no data collection configuration was "
+        "\n   provided; the explicit scrubber should be kept."
+        "\n   Actual:   the scrubber is dropped and the warning misreports why."
+    )
+
+    _, password, token, has_scrubber = scrubber_rows[-1]
+    _, http_query, _, _ = attribute_rows[-1]
+    reproduced = (
+        password == "hunter2"
+        and token == "abc123"
+        and not has_scrubber
+        and http_query is not None
+    )
+    print(f"\nBug reproduced: {reproduced}")
+    return 0 if reproduced else 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())