Basic support for dependencies

[richfitz]

As with the previous PRs, mostly fairly straightforward I hope. Major issues are:

• detection of a "safe" location to copy files into. We need this for global resources too.
• we can probably be more relaxed about what we pull from the arguments to the dependency function, particularly handling around the query parameter, but that can be done later I think
• not sold on orderly_depends; this could have been orderly_use_dependency perhaps. It might make sense to review all the names in one go?

[r-ash]

Looks good, one comment about an additional test but could also leave that for a separate PR

[r-ash]

Can we add a test case where the name of the dependency is different from the artefact? Ran into an issue with this in orderly2 which is what vimc/orderly2#8 is addressing. Could also do that in a separate PR.

[richfitz]

In this example the dependency is being renamed, but there's no "unexpected files" check yet here (it'll be a bit harder to add). Or is it something else that's missing in the orderly2 test?

[r-ash]

I mean the name of the dependency after being pulled into this report run is different from the name of the artefact. They are the same here as you say. Though actually if we have no unexpected files check here then comment above is not relevant!

[hillalex]

I guess the most consistent with the other function names would be orderly_dependency

[richfitz]

done in #9

[hillalex]

Feels like it would be nice if you could just do function(name, query, ...) where ... is the named file arguments.

[richfitz]

I agree, but here's the pair of issues issue with that:

First, In the case where you want to compute the filenames passing in dots is hard as R does not have a spread/splat operator. This will be an issue for Pete's case where we are pulling in things from different regions and renaming them as what we can do with an explicit use argument is:

for (r in regions) {
  query <- rlang::expr(parameter:region == !!r)
  use <- setNames("data.rds", file.path("data", paste0(r, ".rds")))
  orderly_depends("data", query, use)
}

Second, you can compromise by adding a get collisions with names, so we could write:

orderly_depends <- function(name, query, ..., use = c(...)) {
}

or by carefully processing ..., but this requires some care with any analysis of the contents in static_orderly_dependency I think.

I had a look at sorting this out but it looks complicated enough that I'd rather kick it down the road a bit

[richfitz]

https://mrc-ide.myjetbrains.com/youtrack/issue/mrc-4173