Git-Heat-Map

Map showing the files in cpython that Guido van Rossum changed the most; full SVG image available in repo

Website now available

A version of this program is now available for use at heatmap.jonathanforsythe.co.uk

Basic use guide

• Generate database with python generate_db.py {path_to_repo_dir}
• Install flask from pip
• Run web server with python app.py or flask run (flask run --host=<ip> to run on that ip address, with 0.0.0.0 being used for all addresses on that machine)
• Connect on 127.0.0.1:5000
• Available repos will be displayed, select the one you want to view
• Add emails, commits, filenames, and date ranges you want to highlight by using the form on the right, with % acting as a wildcard
• Clicking on any of these entries will cause the query to exclude results matching that entry
• Choose the minimum size of box to draw, with smaller numbers resulting in greater detail at the cost of performance
• Choose the hue that you want the chart to use for highlighting
• Press submit query
• Click on directories to zoom in, and the back button in the sidebar to zoom out
• Update text rendering depth as desired

Project Structure

This project consists of two parts:

1. Git log -> database
2. Database -> treemap

Git log -> database

Scans through an entire git history using git log, and creates a database using three tables:

• Files, which just keeps track of filenames
• Commits, which stores commit hash, author, committer
• CommitFile, which stores an instance of a certain file being changed by a certain commit, and tracks how many lines were added/removed by that commit
• Author, which stores an author name and email
• CommitAuthor, which links commits and Author in order to support coauthors on commits

Using these we can keep track of which files/commits changed the repository the most, which in itself can provide useful insight

Database -> treemap

Taking the database above, uses an SQL query to generate a JSON object with the following structure:

directory:
  "name": <Directory name>
  "val": <Sum of sizes of children>
  "children": [<directory or file>, ...]

file:
  "name": <File name>
  "val": <Total number of line changes for this file over all commits>

then uses this to generate an inline svg image representing a treemap of the file system, with the size of each rectangle being the val described above.

Then generates a second JSON object in a similar manner to above, but filtering for the things we want (only certain emails, date ranges, etc), then uses this to highlight the rectangles in varying intensity based on the vals returned eg highlighting the files changed most by a certain author.

Performance

These speeds were attained on my personal computer.

Database generation

Time taken seems to scale linearly, going through approximately 300 commits/second, or requiring 0.0033 seconds/commit. Database size also scales linearly, with approximately 2600 commits/MB, or requiring 384 B/commit.

Querying database and displaying treemap

For this test I filtered each repo by its most prominent authors:

These times are with minimum size drawn = 0, on very large repositories, so the performance is not completely unreasonable. This does not include the time for the browser to actually render the svg, which can take longer.

Wanted features

Submodule tracking

Currently the only submodule changes that can be seen are the top level commit pointer changes. In the future would like to recursively explore submodules and add their files to the database.

Faster database generation

Currently done using git log which can take a very long time for large repos. Will look into any other ways of getting needed information on files.

Multiple filters per query

Currently the user can submit only a single query for the highlighting. Ideally they could have a separate filter dictating which boxes to draw in the first place, and possibly multiple filters that could result in multiple colour highlighting on the same image.