avoid repeated package dependency processing #615
Conversation
|
This PR looks good to me, although I think we might need to narrow the scope a bit more since (if I understand the test failures correctly) some APIs do rely on us building the full dependency tree for packages. What if we instead hid this behavior behind an option, e.g. something like |
|
Updated the approach to prune the search while still producing the full dependency tree. |
R/pkg.R
Outdated
| priorPkgRecords <- c() | ||
| for (pkgName in pkgNames) { | ||
| if (exists(pkgName, envir = .visited.packages)) { | ||
| append(priorPkgRecords, get(pkgName, envir = .visited.packages)) |
There was a problem hiding this comment.
| append(priorPkgRecords, get(pkgName, envir = .visited.packages)) | |
| priorPkgRecords <- append(priorPkgRecords, get(pkgName, envir = .visited.packages)) |
As is, this statement won't update priorPkgRecords -- is that intentional?
|
LGTM! (Remaining test failures are due to dependent R packages not being available for older R.) |
There was a problem hiding this comment.
LGTM -- thanks for putting this together! (And thanks to @andrie for getting this work started as well.)
an alternate tree-pruning approach from #614; part of an ongoing discussion with @andrie