Incorrect behavior of geod.npts

[FactorizeD]

Hello,

This is my first post here, first and foremost I would like to thank you all for doing a fantastic work with this package:)

I am working on a project where pyproj is used and came across a strange behavior. I haven't checked every version to check where exactly did the issue start happening but it seems to be happening for the newest 3.5.0.

Consider the following:

import Geod

geod = Geod(ellps="WGS84")
geod.npts(np.nan, 1.0, 2.0, np.nan, npts=1)  # returns [(np.nan, 2.0)] ?!
geod.npts(1.0, 2.0, 3.0, np.nan, npts=1)  # returns [(1.0, 2.0)] ?!

Problem description

I would expect .npts to always return a tuple of nans when called on input containing nans. This however doesn't seem to be the case. I checked other possibilities with 2x np.nan or 1x np.nan in the input but only these two seem to be problematic.

Expected Output

[(np.nan, np.nan)] for both

Environment Information

pyproj info:
pyproj: 3.5.0
PROJ: 9.2.0
data dir: /usr/local/lib/python3.8/dist-packages/pyproj/proj_dir/share/proj
user_data_dir: /home/developer/.local/share/proj
PROJ DATA (recommended version): 1.13
PROJ Database: 1.2
EPSG Database: v10.082 [2023-02-06]
ESRI Database: ArcGIS Pro 3.1 [2023-19-01]
IGNF Database: 3.1.0 [2019-05-24]

System:
python: 3.8.10 (default, Mar 13 2023, 10:26:41) [GCC 9.4.0]
executable: /usr/bin/python3
machine: Linux-5.15.90.1-microsoft-standard-WSL2-x86_64-with-glibc2.29

Python deps:
certifi: 2022.6.15
Cython: 0.29.14
setuptools: 58.4.0
pip: 23.1

Installation method

poetry inside docker

What do you think?

[snowman2]

This change appears to be connected to #841.

With pyproj 3.0.1:

>>> import math
>>> from pyproj import Geod
>>> geod = Geod(ellps="WGS84")
>>> geod.npts(math.nan, 1.0, 2.0, math.nan, npts=1)
[(nan, nan)]

With pyproj 3.1.0:

>>> import math
>>> from pyproj import Geod
>>> geod = Geod(ellps="WGS84")
>>> geod.npts(math.nan, 1.0, 2.0, math.nan, npts=1)
[(nan, 0.9999999999999998)]

cc: @idanmiara

[snowman2]

Possibly connected to this change as well: #825

[snowman2]

e0b9453d29e7c9475467268d6838dac8503ab145 is the first bad commit
commit e0b9453d29e7c9475467268d6838dac8503ab145
Author: Idan Miara <idan@miara.com>
Date:   Mon May 10 13:45:25 2021 +0300

    pyproj/[geod.py, _geod.pyi, _geod.pyx] - add `fwd_intermediate`, `inv_intermediate` (improved version of `npts()`) functions; add `out_lons`, `out_lats` params to `npts()`; add examples and some more docs
    pyproj/enums.py, docs/api/enums.rst - add `GeodIntermediateFlag` flag
    test/test_geod.py - add tests for new functions

[snowman2]

Seems the change started in #841

[snowman2]

It seems the issue has to do with is_fwd being defined on whether the input is NaN:

pyproj/pyproj/_geod.pyx

Line 400 in 3d0fb75

bint is_fwd = isnan(lat2_or_nan)

[snowman2]

Fix in #1288