CallGraphs

A package for analyzing source-code callgraphs, particularly of Julia's src/ directory. The main motivation for this package was to aid in finding all functions that might trigger garbage collection by directly or indirectly calling jl_gc_collect; however, the package has broader uses.

Installation

Add with

Pkg.clone("https://github.com/timholy/CallGraphs.jl.git")

You'll also need to have clang++ installed, as well at the corresponding opt tool. On the author's machine, opt is called opt-3.4.

Analyzing a source repository

Extracting the callgraph

An example script is callgraph_jlsrc.bash, which is set to analyze julia's src directory. It should be called from within that directory. You may need to change the OPT variable to match your system. This script can be modified to analyze other code repositories.

This writes a series of *.ll and *.dot files. These *.dot files are then analyzed by the julia code in this repository.

Analyzing the callgraph

The most general approach is

using CallGraphs
cgs = parsedots()   # or supply the dirname
calls, calledby = combine(cgs...)

This will merge data from all the *.dot files in the directory into a single callgraph. parsedots and combine are both described in online help.

Garbage-collection analysis

If your main interest is analyzing the callgraph of julia's garbage collection, you will likely be more interested in

using CallGraphs
gcnames = findgc()
highlight(srcfilename, gcnames)

which produces output that looks like this:

Shown in red are all functions that might trigger a call to jl_gc_collect. The general principle is to look for cases where one line's allocation is not protected from a later garbage-collection.

You can save a (crude) emacs highlighting file with

emacs_highlighting(filename, gcnames)