Interaction with an unrelated library causes unexpected behavior of tms.tiles(...)

[bmcandr]

I'll preface my comment by saying that the behavior I describe below does not occur in v5+ so it doesn't seem to be a critical issue. I'm just reporting something odd that I don't fully understand.

---

I've stumbled upon unusual behavior that appears to result from the interaction between morecantile and the datadog-api-client library. TLDR; tms.tiles(...) returns tiles outside the supplied bounds when modules from datadog-api-client are imported before morecantile.

We're running the following versions:

morecantile==4.3.0
datadog-api-client==2.17.0

We're still running 4.3.0 because we have not migrated our project to Pydantic v2 yet.

Example code:

import json
import shapely.geometry
from datadog_api_client.v2 import ApiClient, ApiException, Configuration
import morecantile

feat = {
            "type": "Feature",
            "geometry": {
                "coordinates": [
                    [
                        [-98.547359, 29.522354],
                        [-98.547359, 29.335023],
                        [-98.27275, 29.335023],
                        [-98.27275, 29.522354],
                        [-98.547359, 29.522354],
                    ]
                ],
                "type": "Polygon",
            },
            "properties": {},
        }

tms = morecantile.tms.get("WebMercatorQuad")

g1 = shapely.geometry.shape(feat["geometry"])

west, south, east, north = g1.bounds

features = [tms.feature(tile) for tile in tms.tiles(west, south, east, north, zooms=[10], truncate=False)]

with open("tiles.geojson", "w") as f:
    json.dump(dict(type="FeatureCollection", features=features), f)

The above code produces a FeatureCollection containing tiles beginning at x=0 rather than only the tiles that intersect feat (shown in red near San Antonio):

If I move the morecantile import before the datadog-api-client import or comment out the line importing modules from datadog_api_client.v2, only the tiles that intersect feat are returned as expected:

Any idea why this happens? I first encountered this in our project repo and verified with a fresh venv with only the libraries required to run the snippet above.

[vincentsarago]

This is SOOOO interesting

For some reason the pydantic model is altered by the import

See how the union order is different, in the first (good) case it's number then integer and in the second case (bad) it's integer then number

The type is defined https://github.com/developmentseed/morecantile/blob/4.3.0/morecantile/models.py#L49

[bmcandr]

Weird! I confirm that there's something going on between Pydantic and Datadog that was revealed through our use of morecantile.

# from datadog_api_client.v2 import ApiClient, ApiException, Configuration
from typing import Union, Tuple
from pydantic import BaseModel

NumType = Union[float, int]
BoundsType = Tuple[NumType, NumType]

class Example(BaseModel):
    b: BoundsType
    
e = Example(b=(-1.0, 1))

print(e.schema()["properties"]["b"]["items"])

Without datadog: [{'anyOf': [{'type': 'number'}, {'type': 'integer'}]}, {'anyOf': [{'type': 'number'}, {'type': 'integer'}]}]

With datadog: [{'anyOf': [{'type': 'integer'}, {'type': 'number'}]}, {'anyOf': [{'type': 'integer'}, {'type': 'number'}]}]

I'll report it to the Pydantic devs to see if they have any ideas. Thanks for taking a look!

[bmcandr]

Looks like it may actually be related to this discussion on Python's runtime type caching. My example can be further reduced to demonstrate the effect with just type aliases. Also, it only seems to arise with a Tuple of Unions? 🤷

from typing import List, Tuple, Union
from datadog_api_client.v2 import ApiClient, ApiException, Configuration

NumType = Union[float, int]
BoundsType = Tuple[NumType, NumType]
BoundsListType = List[NumType]

print(NumType)
print(BoundsType)
print(BoundsListType)

>>> typing.Union[float, int] # right
>>> typing.Tuple[typing.Union[int, float], typing.Union[int, float]]  # wrong
>>> typing.List[typing.Union[float, int]]  # right