Add is_sorted utility, and use it.

[matteobachetti]

I realized that the code for periodograms was not robust when events were not sorted, so I forced it. I wrote a new is_sorted function, that goes through the array and gives False at the first instance of a non-sorted list (super fast in random arrays, O(n) in quasi-sorted arrays), but it is only fast if numba-compiled. I also used the standard (np.diff(array) >=0).all() implementation if Numba is not installed.
I took the chance to use the new function in a number of places where we used different checks of the same kind, and to fix the test data (that were unsorted themselves).

[codecov]

Codecov Report

Merging #723 (b7d4c94) into main (5f13650) will decrease coverage by 0.06%.
The diff coverage is 83.33%.

@@            Coverage Diff             @@
##             main     #723      +/-   ##
==========================================
- Coverage   97.22%   97.16%   -0.06%     
==========================================
  Files          42       42              
  Lines        7848     7869      +21     
==========================================
+ Hits         7630     7646      +16     
- Misses        218      223       +5     

Impacted Files	Coverage Δ
stingray/utils.py	98.29% <78.94%> (-0.73%)	⬇️
stingray/io.py	98.49% <87.50%> (-0.30%)	⬇️
stingray/gti.py	99.27% <100.00%> (+<0.01%)	⬆️
stingray/lightcurve.py	97.83% <100.00%> (-0.01%)	⬇️

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[dhuppenkothen]

One nitpicky question, but otherwise it looks good! 👍

[dhuppenkothen]

I'm a bit confused why you're running _is_sorted_numba twice, once for the first couple of elements, and then for everything? What's happening here?

[dhuppenkothen]

Oh, is this to check whether the array is extended precision? I feel like there should be a more explicit way to check that rather than just figuring out if Numba fails?

[matteobachetti]

It's kind of difficult to test whether an array is extended precision, because it can be called in different ways (np.longdouble, np.float128...), different architecture define these with different precision, and I'm not sure I have full control of all possible ways. Also, I wanted this test to be robust even for other kinds of data (not necessarily extended precision floats) that Numba does not accept and I haven't thought about.

[dhuppenkothen]

👍

[dhuppenkothen]

Codecov seems to be complaining?

[matteobachetti]

@dhuppenkothen there is just one line not tested, the one about detector_id which should be trivial ( and needs to change the data files). The numba function is executed (I'm sure of it, because the line that calls that function is executed) but codecov is probably not able to go inside it.