Here’s some statistics on commits I’ve authored collected from git-cstat over a few work and open-source repositories.

Min, max, average commit length in words

Note the logarithmic x-axis (words).

With my Dotfiles included:

Without my Dotfiles included:

Commentary

• Short commits are mostly init, chore: upgrade dependencies, things like that. Occasionally my shortest contribution to a project is long, though! I think this happens most when I contribute to open source or innersource projects where I need to spell out my reasoning more clearly regardless of the commit.
• Some extremely long commits are the result of squash-and-merge defaults at work. A few projects I’ve worked on required that as the only merge strategy and mostly kept the default “concatenate all commit messages” as the final message. So the reality is that several reasonably-long commits were squashed into a monster.

• My average average drops from a little over 40 to a little over 20 when including my Dotfiles. That makes sense: my Dotfiles have a long life, going back to before I use Git in the way I do today. Even now there are plenty of ~10 word commit messages and only a few long ones: for example, using my fields and bucket10¹ scripts:

    λ g cstat len | fields 1 | bucket10 | sort -n
    1     21 ██
    2    191 █████████████████████
    3    648 ███████████████████████████████████████████████████████████████████████
    4    380 ██████████████████████████████████████████
    5    346 ██████████████████████████████████████
    6    263 █████████████████████████████
    7    172 ███████████████████
    8    111 ████████████
    9     68 ███████
    10   178 ████████████████████
    20   209 ███████████████████████
    30   101 ███████████
    40    73 ████████
    50    16 ██
    60    18 ██
    70    14 ██
    80    15 ██
    90     7 █
    100   13 █
    200    4
    700    1
  

Raw Data

A comment marks the Dotfiles data

# work
# Short commits: "init", "chore: upgrade dependencies", etc.
# Extremely long commits: squashed several reasonable lengths
min: 1
max: 55
avg: 11.8571
min: 3
max: 662
avg: 50.4
min: 4
max: 179
avg: 41.8889
min: 208
max: 356
avg: 282
min: 1
max: 202
avg: 43.5833
min: 23
max: 3995
avg: 1141.25
min: 2
max: 8046
avg: 193.755
min: 2
max: 1145
avg: 62.1368
min: 1
max: 27
avg: 6.42222
min: 2
max: 251
avg: 55.1667
min: 7
max: 68
avg: 20.2222
min: 93
max: 598
avg: 299
min: 4
max: 9
avg: 7
min: 2
max: 279
avg: 104.364
min: 4
max: 98
avg: 33.3636
min: 3
max: 120
avg: 21.1
min: 1
max: 105
avg: 13.8854
min: 1
max: 215
avg: 10.2741
min: 1
max: 160
avg: 15.6222
min: 4
max: 464
avg: 115.13

# OSS
min: 1
max: 730
avg: 17,1961
min: 41
max: 343
avg: 103,4
min: 2
max: 102
avg: 23,9048
# Dotfiles
min: 1
max: 705
avg: 10,3589

Pict script

This is also in my bin directory.

It would arguably have been more efficient to read the data into a tidy dataframe from the outset, but I was more confident in this version of the parsing code, and it works.

The empty racket/gui require is to make it easy to replace (save-pict out) with show-pict when you want just see the results. I experimented with an interface to either show or save the picture, and I didn’t come up with anything I particularly liked.

PS Apparently the syntax highlighter here (rouge) doesn’t know how to parse Racket’s reader syntax for (Perl-ish) regular expressions.

#! /usr/bin/env racket
#lang racket
; vim: ft=racket

(require (only-in racket/gui)
         pict
         threading
         data-frame
         sawzall
         graphite)

(define-values (data-file out)
  (command-line
   #:args (data-file out)
   (values data-file out)))

(define-values (min max avg)
  (for/lists (min? max? avg?
              #:result (values (filter values min?)
                               (filter values max?)
                               (filter values avg?)))
             ([line (in-lines (open-input-file data-file))])
    (match line
      [(pregexp #px"^min: ([[:digit:].]+)" (list _ (app string->number min))) (values min #f #f)]
      [(pregexp #px"^max: ([[:digit:].]+)" (list _ (app string->number max))) (values #f max #f)]
      [(pregexp #px"^avg: ([[:digit:].]+)" (list _ (app string->number avg))) (values #f #f avg)]
      [_ (values #f #f #f)])))

(define df
  (make-data-frame #:series (list (make-series "min" #:data (list->vector min))
                                  (make-series "max" #:data (list->vector max))
                                  (make-series "avg" #:data (list->vector avg)))))

(~> df
    (pivot-longer everything #:names-to "stat" #:values-to "words")
    (graph #:data _
            #:mapping (aes #:y "stat" #:x "words")
            #:y-label "Stat" #:x-label "Number of words"
            #:x-transform logarithmic-transform
            #:title "Commit message length (words)"
            (boxplot #:invert? #t #:show-outliers? #t)
            (points))
    (save-pict out))

Notes

Categories: Blog