Learn R Programming

mvbutils (version 1.1.1)

foodweb: Shows which functions call what

Description

foodweb is applied to a group of functions (e.g. all those in a workspace); it produces a graphical display showing the hierarchy of which functions call which other ones. This is handy, for instance, when you have a great morass of functions in a workspace, and want to figure out which ones are meant to be called directly. callers.of(funs) and callees.of(funs) show which functions directly call, or are called directly by, funs.

Usage

foodweb( funs, where=1, charlim=80, prune=character(0), rprune, ancestors=TRUE,
 descendents=TRUE, plotting =TRUE, plotmath=FALSE,
 generics=c( "c","print","plot", "["), lwd=0.5, xblank=0.18,
 border="transparent", boxcolor="white", textcolor="black",
 color.lines=TRUE, highlight="red", ...)
 plot.foodweb(x, textcolor, boxcolor, xblank, border, textargs = list(),
 use.centres = TRUE, color.lines = TRUE, poly.args = list(),
 expand.xbox = 1.05, expand.ybox = expand.xbox * 1.2, plotmath = FALSE, ...)
 callers.of( funs, fw=foodweb( plotting=FALSE))
 callees.of( funs, fw=foodweb( plotting=FALSE))

Arguments

funs
character vector OR (in foodweb only) the result of a previous foodweb call
where
position(s) on search path
charlim
controls maximum number of characters per horizontal line of plot
prune
character vector. If omitted, all funs will be shown; otherwise, only ancestors and descendants of functions in prune will be shown. Augments funs if required.
rprune
regexpr version of prune; prune <- funs %matching% rprune. Does NOT augment funs. Overrides prune if set.
ancestors
show ancestors of prune functions?
descendents
show descendents of prune functions?
plotting
graphical display?
plotmath
leave alone
generics
calls TO functions in generics won't be shown
lwd
see par
xblank
leave alone
border
border around name of each object (TRUE/FALSE)
boxcolor
background colour of each object's text box
textcolor
of each object
color.lines
will linking lines be coloured according to the level they originate at?
highlight
seemingly not used
...
passed to plot.foodweb and thence to par
textargs
not currently used
use.centres
where to start/end linking lines. TRUE is more accurate but less tidy with big webs.
expand.xbox
how much horizontally bigger to make boxes relative to text?
expand.ybox
how much vertically bigger to ditto?
poly.args
other args to rect when boxes are drawn
fw
an object of class foodweb, or the funmat element thereof (see VALUE)
x
a foodweb (as an argument to plot.foodweb)

Value

  • foodweb returns an object of (S3) class foodweb. This has three components:
  • funmata matrix of 0's and 1's showing what (row) calls what (column). The dimnames are the function names.
  • xshows the x-axis location of the centre of each function's name in the display, in par("usr") units
  • levelshows the y-axis location of the centre of each function's name in the display, in par("usr") units. For small numbers of functions, this will be an integer; for larger numbers, there will some adjustment around the nearest integer
  • Apart from graphical annotation, the main useful thing is funmat, which can be used to work out the "pecking order" and e.g. which functions directly call a given function. callers.of and callees.of return a character vector of function names.

Details

The main value is in the graphical display. At the top ("level 0"), functions which don't call any others, and aren't called by any others, are shown without any linking lines. Functions which do call others, but aren't called themselves, appear on the next layer ("level 1"), with lines linking them to functions at other levels. Functions called only by level 1 functions appear next, at level 2, and so on. Functions which call each other will always appear on the same level, linked by a bent double arrow above them. The colour of a linking line shows what level of the hierarchy it came from. foodweb makes some effort to arrange the functions on the display to keep the number of crossing lines low, but this is a hard problem! Judicious use of prune will help keep the display manageable. Perhaps counterintuitively, any functions NOT linked to those in prune (which all will be, by default) will be pruned from the display. foodweb tries to catch names of functions that are stored as text, and it will pick up e.g. glm in "do.call( glm, glm.args)". There are limits to this, of course (?methods?). The argument list may be somewhat daunting, but the only ones normally used are funs, where, and prune. Also, to get a readable display, you may need to reduce cex and/or charlim. A number of the less-obvious arguments are set by other functions which rely on plot.foodweb to do their display work. Several may disappear in future versions. If the display from foodweb is unclear, try foodweb( .Last.value, cex=<>, charlim=<>). This works because foodweb will also accept a foodweb-class object as its argument. You can also assign the result of foodweb to a variable, which is useful if you expect to do a lot of tinkering with the display, or to inspect the who-calls-whom matrix by hand. callers.of and callees.of process the output of foodweb, looking for immediate dependencies only. The second argument will call foodweb by default, so it may be more efficient to call foodweb first and assign the result to a variable.

Examples

Run this code
foodweb( ) # functions in .GlobalEnv
foodweb( where="package:mvbutils", cex=0.4, charlim=60) # yikes!
foodweb( c( find.funs("package:mvbutils"), "paste"))
# functions in .GlobalEnv, and "paste"
foodweb( find.funs("package:mvbutils"), prune="paste")
# only those parts of the tree connected to "paste";
# NB that funs <- unique( c( funs, prune)) inside "foodweb"
foodweb( where='package:mvbutils', rprune="aste")
# doesn't include "paste" as it's not in "mvbutils", and rprune doesn't augment funs
foodweb( where='package:mvbutils', rprune="name") # does work
fw <- foodweb( where="package:mvbutils")
fw$funmat # a big matrix
callers.of( "mlocal", fw)
callees.of( find.funs() %matching% "name", fw)

Run the code above in your browser using DataLab