I have a directory with hundreds of thousands of text files. I wanted to explore one file, so I wrote the following code expecting it to happen basically instantaneously because of how generators work:

```python
from pathlib import Path

base_dir = Path("/path/to/lotta/files/")
files = base_dir.glob("*.txt")            # return immediately
first_file = next(files)                  # doesn't return immediately
```

to my surprise, this took a long time to finish since `next` on a generator should be O(1).

A colleague pointed me to the following code: https://github.com/python/cpython/blob/adcd2205565f91c6719f4141ab4e1da6d7086126/Lib/pathlib.py#L431

I assume calling this list is to "freeze" a potentially changing directory since `scandir` relies on `os.stat`, but this causes a huge penalty and makes the generator return-type a bit disingenuous. In any case, I think this is bug worthy in someo sense.

The reason is different. The scandir() iterator should be closed before we go recursively deep in the directory tree. Otherwise we can reach the limit of open file descriptors (especially if several glob()s are called in parallel). See issue22167.

I agree that from the outside is seems slightly bizarre to make an internal list to implement a function documented as returning an iterator.  However, list(scandir) was added by Serhiy in #26032 with the comment that it made globbing 1.5-4 times faster.  This is, of course, if one runs the iterator to completion, as is the normal use.

For your presented use case, I suggest something like the following:

next(f for f in scandir(path) if os.path.splitext(f)[1] == '.txt')