Generating a Call Graph in R
I've been given a big chunk of poorly formatted monolithic R code with plenty of functions, and I'd like to work out what functions call what functions.
I thought I could use roxygen's @callGraph stuff, but a) the code needs to be in a package, which will be a headache with this code, and b) it doesn't even seem to work when I do run it on a simple package of mine. I see a posting from one of the Roxygen authors saying call graph generation is disabled because of the Rgr开发者_如何学Caphviz dependency, but the code is there. Anyway.
Anyone got a better way of quickly working out that foo calls bar, baz, and qux, and that qux calls quux?
Edit: Solutions based on R's profiling system are great, assuming you can actually run the code... Half the stuff in the files doesn't get run, and I don't know what it does... Static analysis is too much to hope for, I guess.
Edit 2: Roxygen's call graph stuff seems to do a static analysis, based on recursive descent of the expression of the function and checking for is.callable. It would be lovely to be able to run this on any function... I may play with this tomorrow...
Would profr help you out? From the documentation:
> ?profr
> glm_ex <- profr(example(glm))
Read 17 items
> head(glm_ex)
f level time start end leaf source
8 example 1 0.32 0.00 0.32 FALSE utils
9 <Anonymous> 2 0.04 0.00 0.04 FALSE <NA>
10 source 2 0.28 0.04 0.32 FALSE base
11 prepare_Rd 3 0.02 0.00 0.02 FALSE <NA>
12 render 3 0.02 0.02 0.04 FALSE <NA>
13 getSrcLines 3 0.02 0.04 0.06 FALSE base
> summary(glm_ex)
f level time start
eval.with.vis :10 Min. : 1.000 Min. :0.02 Min. :0.0000
<Anonymous> : 3 1st Qu.: 4.000 1st Qu.:0.02 1st Qu.:0.1200
lazyLoadDBfetch: 3 Median : 6.000 Median :0.02 Median :0.2000
%in% : 3 Mean : 7.211 Mean :0.03 Mean :0.1769
inherits : 3 3rd Qu.: 9.000 3rd Qu.:0.02 3rd Qu.:0.2600
is.factor : 3 Max. :22.000 Max. :0.32 Max. :0.3000
(Other) :65
end leaf source
Min. :0.0200 Mode :logical Length:90
1st Qu.:0.1500 FALSE:75 Class :character
Median :0.2400 TRUE :15 Mode :character
Mean :0.2069 NA's :0
3rd Qu.:0.2800
Max. :0.3200
> plot(glm_ex)
Not quite what you are after, but you may be able to adapt it to your needs.
The CodeDepends
package (CRAN, website, GitHub) looks interesting, I haven't looked into it though. Among others, it promises
- creating call graphs between sets of functions
presumably through the makeCallGraph()
function.
I have not used it, but a quick look at the proftools package indicates that it can do this. You'll have to run Rprof first and then use the proftools to analyze the output. I think you want the plotProfileCallGraph()
function.
Besides proftools and profr, there is the Perl script by Romain on the R Wiki. Combined with graphviz, it does graphs (with edges weighted) -- see here for more.
精彩评论