Simon Biggs

So, I might have messed something up with the benchmark. I am just using `timeit` + running the function once for preheating. And, gee wizz, it looks like XLA absolutely...

Also, @axeldavy, I tried to build and install your fork: https://colab.research.google.com/drive/1TNIUJN4W95V81VkWq0n2TksKisnfbZTR?usp=sharing#scrollTo=EdjOCYULwXv1 But I didn't have success. Keen to know what I did wrong there. It would be nice to be...

I'm going on leave for the next few weeks so won't be in a position to look at this again until the 3rd quarter of Jan. If someone else is...

So after taking three weeks of leave, it seems my plate is now exceptionally full. I'm going to have to put this in my backlog for the time being.

Yup, the key part of the code to adjust would be here: https://github.com/pymedphys/pymedphys/blob/9ebe4ad4261709c5a1c3bcef120be0660e715e84/lib/pymedphys/_trf/decode/header.py#L61-L85

One option here is to read the file in reverse and verify that the file aligns to the right number of columns?

Historically iirc it was two bytes per item.

Ahh it seems like it changed between version 3 and version 4. Version 1-3 I decode with 16 bit integers, then version 4 I decoded with 32 bit integers. https://github.com/pymedphys/pymedphys/blob/d8bb83c8e80f5701bc342e5184f70321fad2cc90/lib/pymedphys/_trf/decode/config.json#L21-L42

Also my decoding approach was overly complicated :( sorry about that

@mfizyczka, might you be open to uploading your reference TRF files to Zenodo? Within pymedphys, in order to verify that future code changes don't break features like being able to...