Hmm. Section 4 of https://dmtn-181.lsst.io/v/u-ktl-initial-draft/index.html has the requirement – Kian-Tat Lim what's stopping us merging this to main? Maybe we decided that it superseded by https://rtn-023.lsst.io ?

And also I have no idea what @file syntax is.

I was envisaging a CSV with dataId values (or even serialized JSON DatasetRef) in it or else a text file with UUIDs in.

In Gen2 the way one specified a file with a list of dataIds on the command line was with the @ symbol and the path to the file, that's all.

What was the format of that file? How did you tell it which datasetTypes were relevant?

Well, in Gen2 land it was just a matter of passing dataIds to your task, as that would do a runDataRef() calling get() with the types it needed, using the supplied ids, so an example for processCcd just looked like

--id dayObs=2020-03-12 seqNum=375

--id dayObs=2020-03-12 seqNum=376

--id dayObs=2020-03-12 seqNum=377

Linking DM-31725, where I had planned to do this (among many other things), and will at least do some prototyping around the problems involved with implementing it.

Jim Bosch The linked ticket is closed but I don't think we've implemented the ability to read in a file of dataIds/UUIDs in pipetask qgraph.

Yes, this can stick around (as can DM-30438).  DM-31725 took a big step in the right direction but didn't get us all the way there.