Machine-readable output file

[markdewing]

There are several areas that would benefit from machine-readable output that is separate from the human-readable output.

• metadata for estimator postprocessing
• metadata for restart
• better for analysis tools (eg, twist averaging or multiple timesteps in the same run)

Any other requirements?

[prckent]

Has a spec

[ye-luo]

Currently there is a *.qmc.xml generated for each section but I have no idea about how it was used. Probably it is the supposed machine-readable output + metadata.

[prckent]

• Some thinking on how the output should relate to our machine readable input
• A default name that clearly indicates it is output (*.qmc.xml is ambiguous)

[lshulen]

I would prefer to have a single file per run. I think that would be cleanest in thinking about a restart or post processing. Also, the per series files don't currently document the estimators used and I think it would be redundant to add it to each one.

This is not to say that this global file couldn't point to individual files for each series if that makes the most sense.

One comment about Paul's statement about how this relates to our input. If we use XML and have a continuous file for the entire run of qmcpack, there is an issue for calculations that don't complete. In this case, there will be no closing tag and the xml file will technically be invalid. We could simply have all post processing tools check for this and maybe even close the tag, but it's a bit ugly.