repodeps

repodeps

This repository contains the code from an experiment to analyze, store, and update package-level dependencies from Git repositories. Currently it works only for Go, but we plan to try adding support for other languages.

Be warned that this code is not production ready and may change without notice.

Installation

git clone github.com/creachadair/repodeps
cd repodeps
./install.sh  # copies binaries to $GOBIN or $GOPATH/bin
go install github.com/creachadair/jrpc2/cmd/jcall

The rest of these instructions assume the installed binaries are somewhere in your $PATH. You will also need jq. On macOS you can get jq via brew install jq.

Set Up a Database

The depserver program maintains a database of dependency graph information, and a database of repository state. Most of the other programs work by sending JSON-RPC requests to depserver. To set up a server, choose an address (either a host:port pair or a Unix-domain socket), and run:

export DEPSERVER_ADDR=$TMPDIR/depserver.sock
depserver -address $DEPSERVER_ADDR -graph-db path/to/graphdb -repo-db path/to/repodb

This will create empty databases if they do not already exist, or re-open existing ones.

Initializing the Database

1. Generate an initial list of repositories. For our experiment, we did this using a GitHub search for repositories with a go.mod file at the root. You could also start with a listing of import paths from godoc.org:

   curl -L https://api.godoc.org/packages | jq -r .results[].path > paths.txt
   resolvedeps -filter-known -stdin < paths.txt > repos.txt
   

   Note that doing this for the complete list of godoc.org packages will take a long time, and the results will contain a lot of noise, as that corpus includes vendored packages, internal packages, code that doesn't build, and so forth. For our experiment we had better results from the search query. For local experimentation, you can use your own repositories:

   # If you get rate limited, set GITHUB_TOKEN.
   hub api --paginate users/creachadair/repos \
   | jq -r '.[]|select(.fork|not).html_url' > repos.txt
   

2. Extract dependency information into a database. Given a list of repository URLs as in step (1), use:

   xargs -I@ jcall -c "$DEPSERVER_ADDR" Update '{"repository":"@"}' < repos.txt
   

   This will be slow if the list is very long, but fine for a few dozen.

Finding Missing Dependencies

1. Find import paths mentioned as dependencies, but not having their own nodes in the graph. Resolve each of these to a repository URL:

   missingdeps -domain-only | resolvedeps -stdin | sort -u > missing.txt
   

2. Attempt to fetch and update the contents of the missing repositories into the database. This may not succeed for repositories that require authentication, or if there are other issues cloning the repository.

   xargs -I@ jcall -c "$DEPSERVER_ADDR" Update '{"repository":"@"}' < missing.txt
   

   Be warned, however, that programs in the wild often have very weird dependencies. For example, there are packages that depend on non-standard forks of the standard library, and you may wind up pulling those forks into your database.

This process may be iterated to convergence, in theory, but in practice it will never fully converge because there are a lot of custom build hacks, dead code, deleted repositories, and so forth. Call it, rather, iterated improvement.

Updating the Graph

To scan all the repositories currently mentioned by the graph to check for updates:

jcall -c "$DEPSERVER_ADDR" Scan '{"logUpdates":true, "logErrors":true}'

This may be rerun as often as desired; the repository database keeps track of which repository digests it has seen most recently, and will only update those that change. Use the -interval flag on the depserver tool to control how often this may occur. It doesn't make sense to choose an interval shorter than the time it takes to fully run the update (which depends on the current database size).

Indexing the Standard Library

The standard library packages follow different rules. To index them:

git clone https://github.com/golang/go
repodeps -store "$DEPSERVER_ADDR" -stdlib -trim-repo -import-comments=0 -sourcehash=0 ./go/src

Generally these only need to be reindexed when there is a new release.

Computing PageRank

To compute or re-compute ranking stats,

jcall -c "$DEPSERVER_ADDR" Rank '{"logProgress": true, "update": true}'

Converting to Other Formats

These tools work directly on the database, so you have to stop depserver if you want to use them. TODO(creachadair): Fix that.

• To convert to CSV for Gephi:

  # With everything inline.
  csvdeps -store path/to/graphdb > output.csv
  
  # With a separate vocabulary file.
  csvdeps -store path/to/graphdb -ids vocab.txt > output.csv
  

• To convert to RDF N-quads for a graph database like Cayley:

  # To write RDF quads to stdout.
  quaddeps -store path/to/graphdb
  
  # To write to a Bolt database for Cayley.
  # N.B. If your database is very large, Bolt may choke.
  quaddeps -store path/to/graphdb -output cayley.bolt