Pyright complains incorrectly about object not being iterable.

[fkhawajagh]

So I came across this 'not a bug' ticket. #7916. I seem to be having a similar issue even when I'm using guards (both positive and negative)

If I add a default value for the get function, the alert goes away.

Any thoughts on what I'm doing wrong here?

"node_modules/pyright": {
  "version": "1.1.383",
  "resolved": "https://registry.npmjs.org/pyright/-/pyright-1.1.383.tgz",
  "integrity": "sha512-b540vUDWGXFlVwhxREgCrvKYT9bnUUPiDtSv5s7sUGxIokTxc06bPC2vfnGunUqaUu6hgIqlv1GRFdOKIEb09A==",

[erictraut]

I'm happy to help you diagnose the problem, but could you please post a minimal repro in text form so I don't need to retype your code from the screen shots?

[fkhawajagh]

Thank you .. here's the code

def some_func():
    x = {
        'data': [
            {'id': 1, 'name': 'test1'},
            {'id': 2, 'name': 'test2'},
            {'id': 3, 'name': 'test3'},
            {'id': 4, 'name': 'test4'}
        ],
        'other': []
    }

    if x.get('data'):
        for row in x.get('data', None):
            print(f'id: {row.get("id")}')

[erictraut]

Thanks for the code.

You didn't state this, but I presume that you're using strict type checking mode or have manually enabled the strictDictionaryInference configuration option. Without this, I'm not able to repro the issue.

The problem in this code is that you're assuming that a call to x.get("data") implies that a subsequent call to x.get("data", None") will never return None. A type checker cannot make such an assumption. It must assume that each call is independent of other calls, since functions in Python are not guaranteed to return the same value even if the same arguments are passed to a function multiple times.

Here's how to make this code type safe:

    data = x.get("data")
    if data:
        for row in data:
            print(f'id: {row.get("id")}')

Or if you prefer to use a walrus operator:

    if data := x.get("data"):
        for row in data:
            print(f'id: {row.get("id")}')

[fkhawajagh]

Thanks for the response. I haven't modified pyright at all. This is the default config as installed by Mason. That said, I tried this code (below) which also silences the alert. I don't like the assignment to data because it's the only place in my real function where this object is being referenced and it seemed unnecessary to create an intermediate variable.

def some_func():
    x = {
        'data': [
            {'id': 1, 'name': 'test1'},
            {'id': 2, 'name': 'test2'},
            {'id': 3, 'name': 'test3'},
            {'id': 4, 'name': 'test4'}
        ],
        'other': []
    }

    for row in x.get('data', []):
        print(f'id: {row.get("id")}')

So the question now is.. if None is not allowed, why does this not generate an alert?

    for row in x.get('data', None):
        print(f'id: {row.get("id")}')

[erictraut]

I'm not familiar with mason or how it configures pyright. You may need to reach out to its maintainers for additional details. Perhaps you have a pyproject.toml file with a [tools.pyright] section?

By default, pyright will not generate an error in your initial code sequence. You can see this in the pyright playground. That's because pyright's default inference rules for dictionary expressions infer a type of dict[str, Unknown] for x in this case. If you enable strict mode, the type of x will be inferred as dict[str, list[dict[str, int | str]]], and you will now see a type violation error.

You should see the same behaviors (with and without strict mode enabled) in your last code sample.