Make nice-looking plots of dependency trees

[wlandau-lilly]

It should be as simple as DiagrammeR::from_igraph(build_graph(...)), but from_igraph() does not yet work for me.

With DiagrammeR, I should also be able to assign different node colors for objects, functions, files, targets, etc. Maybe I can color edges by whether a target is created with a command or imported.

[wlandau-lilly]

See rich-iannone/DiagrammeR#219.

[wlandau-lilly]

@enantz-lilly suggested visNetwork. (This need not replace DiagrammeR. Users could choose the backend.)

library(visNetwork)
igraph_object = drake:::dbug()$graph
network_data = toVisNetworkData(igraph_object)
network_data$edges$arrows = "to" # For directed edges
visNetwork(nodes = network_data$nodes, edges = network_data$edges, height = "500px")

It would also be nice to have different-looking nodes for imports vs targets or files vs functions vs non-function objects.

[wlandau-lilly]

CodeDepends uses Rgraphviz, but that one is on Bioconductor.

[wlandau-lilly]

I will release improved plotting functionality with the next version of drake. Here's what it looks like in the development version. Now is a great time for suggestions and requests.

attach(drake:::dbug()) # Workflow and environment I use for debugging
envir$f = function(x) g(x) + a + NOT_FOUND # Sometimes some symbols cannot be imported as objects. Drake just rolls with it, even in make().
plot_graph(plan = plan, envir = envir) # Build and render the graph

[wlandau-lilly]

Even better: let's sort the graph from left to right and show everyone how the parallel computing works. In the new graph below, each column is a separate parallelizable stage. When you call make(), you go from left to right, and the targets in the current column are distributed over the available parallel jobs. In other words, you have parallelism within columns, and you move sequentially from one column to the next. In the example below, we can do the first 7 targets in parallel, then j and myinput in parallel, and then there's the unavoidably sequential move to h, and so on.

For those who are curious, I am now happily committed to visNetwork. I set the "level" column in the "nodes" data frame in visNetwork() and then call visHierarchicalLayout(direction = "LR").

Before I officially mark this issue as solved, I need to say more about graphing in the quickstart vignette at least.