Build dependency graphs between Coq objects, where Coq is the famous formal proof management system (see http://coq.inria.fr/).
Travis CI status on master branch:
First of all, it is a small tool (a Coq plug-in) that extracts the dependencies between Coq objects, and produces a file (we suggest using the suffix .dpd) with this information.
The idea is that other small tools can be then developed to process the .dpd files. At the moment, there is:
dpd2dot
that reads these .dpd files and produces a graph file using .dot format (cf. http://www.graphviz.org/) that makes possible to view them;dpdusage
: to find unused definitions.
Hope other tools later on to do more things. Feel free to contribute!
You can:
- either clone it from GitHub at: https://github.com/Karmaki/coq-dpdgraph
- or get the opam package named
coq-dpdgraph
from the opam-coq archive (repository "released") - or get the latest distributed version
- The latest version runs with Coq 8.7.0
- it has been tested with a version of Coq installed using opam and with Ocaml version 4.04.2
- ocamlgraph (for dpd2dot tool) Any version should work since only the basic feature are used.
First download the archive, unpack it, and change directory to the coq-dpdgraph
directory.
Depending on how you got hold of the archive, you may be in one of three situations:
1/ Makefile is present
You should type the following command.
$ make && make install
2/ configure is present, but no Makefile
You should type the following command.
$ ./configure && make && make install
3/ configure is not present, Makefile is not present
You should type the following command.
$ autoconf
$ configure && make && make install
If you use opam, you can install coq-dpdgraph
and ocamlgraph
$ opam repo add coq-released https://coq.inria.fr/opam/released
$ opam install coq-dpdgraph
If you install the archive by cloning the git repository, you have a sub-directory containing test files. These can be tested using the following command.
$ make -s test
to check if everything is ok.
- to have compiled the tools (see above)
- a compiled Coq file.
You can for instance use
tests/Test.v
(a modified clone of CoqList.v
) and compile it doing :
$ coqc tests/Test.v
The available commands are :
-
Generate dependencies between a list of objects:
Print FileDependGraph <module name list>.
A module can be a file, or a sub-module in a file. Example :
Print FileDependGraph M M2.A.B.
Take all the objects of the specified modules and build the dependencies between them.
-
Generate the dependencies of one objects:
Print DependGraph my_lemma.
Analyse recursively the dependencies of
my_lemma
. -
Change the name of the generated file (default is
graph.dpd
):Set DependGraph File "f.dpd".
Useful when one needs to build several files in one session.
Advice:
you need to use Require
to load the module that you want to explore,
but don't use any Import
/Export
command because the tool is then unable
to properly group the nodes by modules.
Example:
$ ledit coqtop -R . dpdgraph -I tests/
Welcome to Coq 8.5 (April 2016)
Coq < Require dpdgraph.dpdgraph.
[Loading ML file dpdgraph.cmxs ... done]
Coq < Require List.
Coq < Print FileDependGraph List.
Print FileDependGraph List.
Fetching opaque proofs from disk for Coq.Lists.List
Info: output dependencies in file graph.dpd
Coq < Set DependGraph File "graph2.dpd".
^D
$ ./dpd2dot graph.dpd
Info: read file graph.dpd
Info: Graph output in graph.dot
There are some options :
$ ./dpd2dot -help
Usage : ./dpd2dot [options]
-o : name of output file (default: name of input file .dot)
-with-defs : show everything (default)
-without-defs : show only Prop objects
-rm-trans : remove transitive dependencies (default)
-keep-trans : keep transitive dependencies
-debug : set debug mode
-help Display this list of options
--help Display this list of options
If the name of the output file finishes with .dot
, then the name before
the .dot
suffix is used as the graph name in the dot syntax. There
are two exceptions: graph
and digraph
will be replaced with
escaped_graph
and escaped_digraph
respectively.
The only useful option at the moment is -without-defs
that export only
Prop
objects to the graph (Axiom
, Theorem
, Lemma
, etc).
You need :
- graphviz (ie. dot tool)
- a visualizer: we tested zgrviewer, xdot, kgraphviewer, but there are others.
You can also convert .dot file to .svg file using :
$ dot -Tsvg file.dot > file.svg
You can then use firefox
or inskape
to view the .svg
graph.
The main advantage of using firefox
is that the nodes are linked to
the coqdoc
pages if they have been generated in the same directory.
But the navigation (zoom, moves) is very poor and slow.
The graph can be interpreted like this :
- edge n1 --> n2 : n1 uses n2
- node :
- green : proved lemma
- orange : axiom/admitted lemma
- dark pink : Definition, etc
- light pink : Parameter, etc
- violet : inductive,
- blue : constructor,
- multi-circled : not used (no predecessor in the graph)
- yellow box : module
- objects that are not in a yellow box are Coq objects.
You can use dpdusage
command to find unused definitions.
Example:
$ ./dpdusage tests/graph2.dpd
Info: read file tests/graph2.dpd
Permutation_app_swap (0)
In the example above it reports that Permutation_app_swap
was
references 0 times. You can specify max number of references allowed
(default 0) via -threshold
command line option.
graph : obj_list
obj : node | edge
node : "N: " node_id node_name node_attributes ';'
node_id : [0-9]+
node_name : '"' string '"'
node_attributes :
| empty
| '[' node_attribute_list ']'
node_attribute_list :
| empty
| node_attribute ',' node_attribute_list
node_attribute :
| kind=[cnst|inductive|construct]
| prop=[yes|no]
| path="m0.m1.m2"
| body=[yes|no]
edge : "E: " node_id node_id edge_attributes ';'
edge_attributes :
| empty
| '[' edge_attribute_list ']'
edge_attribute_list :
| empty
| edge_attribute ',' edge_attribute_list
edge_attribute :
| weight=NUM
The parser accept .dpd files as described above,
but also any attribute for nodes and edges having the form :
prop=val
or prop="string..."
or prop=NUM
so that the generated .dpd
can have new attributes without having to change
the other tools.
Each tool can then pick the attributes that it is able to handle;
they are not supposed to raise an error whenever there is
an unknown attribute.
Also see: