scivision

Thanks Volker. Line profiling shows that per-epoch time conversion is penalized almost as much (within factor of 2) by the Numpy text=>ndarray conversion as xarray.concat I suspect Numpy can be...

Do you have a public example of a large RINEX file you'd like to benchmark? I think from `TimeRinex.py`, which essentially is [this function](https://github.com/scivision/georinex/blob/master/georinex/obs3.py#L102) that it can be fast to:...

OK, checkout the latest release v1.6.1. It is more than 10x faster overall by preallocating Numpy array and only at the very end, after the big loop is done xarray.Dataset...

Yes I felt that doing a fully-allocated Numpy array was simpler to debug, and that lists probably would not be faster than this, although lists would be better for memory...

I had to read/discard using _skip because when using TextIO streams (for seamless streaming of ZIP/Z/GZ/Hatanaka) seeking doesn't work, you just have to read and discard it seems. I did...

preallocation without reading the file twice is enabled in v1.6.6. This is about a 15% speedup, and it may very well be faster than linked lists because the array is...

Great, v1.6.7 attempts to reduce RAM usage by reading a few dozen lines of data to better estimate needed RAM. If lots of measurements (L1, C1, etc.) are in file,...

@VOMAY just a lack of time to do so, waiting for someone to ask :+1:

I see anand is still working on the citrus branch. I may have time at year-end to look at this and at least make it an optional module e.g. `import...

Yes after refactoring obs2 previously, I also saw that sticking with lists and dicts until closer to the end also generally gave noticeably better performance. I think I originally came...