Identical wavelength for different transitions in the same file

Motivations:

In its current version, aurora load spectral lines from ADAS adf15 files into a dictionary where they are sorted by wavelength (i.e.: the wavelentgth, as a float, is the key of the dictionary)
This breaks down if, unfortunately, several different transitions share the same wavelength.
A long-term fix would be to set a unique key to each transition, using something else than the wavelength.
This issue was encountered while following tutorial #46

Current behaviour:

On python 3.7.11, the behaviour when fed several times the same key is that the dict will only retain the last provided value:

In [1]: {0: 1, 0: 2}
Out[1]: {0: 2}

It means that in its current version, aurora, when confronted with several transitions of identical wavelengths, will only provide the one it encounters last in the file.

Example:

Interestingly, it happens forAr16+, when ones looks for the k lines, it turns out a different transition with the same wavelength (3.9900 A) exists in the same file pec#ar15.dat:

Hi @fsciortino , thanks for the update on master.

Using it raised one error, that I corrected on my fork of aurora, and then a second error occured, that I also corrected on my fork.

Then it works, but I started thinking that when a line wavelength is shifted, the code should raise a warning so the user knows it, so I implemented the warning.

Using that version of the code revealed that there are actually many lines with redundant wavelengths.

So I implemented a more synthetic warning that is raised only once and gives the list of all redundant wavelengths (ordered according to the wavelength).

For file pec#ar15.dat here is the result: 183 lines have redundant wavelength.

In [11]: import aurora

In [12]: filepath_li = '/Home/DV226270/Diag_XICS/Atomic_data/Aurora/pec#ar15.dat'

In [13]: di = aurora.read_adf15(filepath_li)

I see at least 2 causes for concern:

you will notice that more than 2 transitions can share the same wavelength => shifting the wavelength by +1e-6 A is not enough, because when 3 lines have the same wavelength, 2 of them will remain redundant (both shifted by 1e-6 A, so still equal) => Incremental shifts may fix that
you will see that some different lines can, unfortunately,be separated by ...1e-6 A. => If lines a and b share the same wavelength lam, and line c has wavelength lam+1.e-6, if line b is shifted by 1.e-6 A, it will become redundant with line c, which originally, was not redundant with any other line.

Thoughts on this:

I think a solid fix would be to set index_lines=True by default, and to prefer that kind of indexing everywhere in the code, but obviously that's non-negligible work that may take some time
A temporary fix might be to:
- shift lines by an increment smaller than 1.e-6 A (maybe 1.e-10 A ?) to avoid case 2 (even though it theory, it can still happen)
- implement the possibility of multiple increments to avoid case 1.

Proposition:

I submit my code changes for merging into master in PR #49, feel free to merge it or just to keep it for back-up and select parts of the changes for your own changes.

fsciortino / Aurora