shotgun

A minimal screenshot utility for X11. Shotgun was written to replace maim in my workflow.

Features:

• Exports PNG screenshots to file or stdout
• Masks off-screen areas on multi-head setups
• Supports selections by window ID and geometry
• On average, shotgun is more than twice as fast as maim

Installation

From source:

• install a recent Rust toolchain
• clone this repository and run cargo install --path .
• or install from crates.io: cargo install shotgun

Usage

Usage: shotgun [options] [file]

Options:
    -i, --id ID         Window to capture
    -g, --geometry WxH+X+Y
                        Area to capture
    -f, --format png/pam
                        Output format
    -s, --screen        Capture the current screen
    -h, --help          Print help and exit
    -v, --version       Print version and exit

Examples

To use with hacksaw: take a screenshot and copy to clipboard

#!/bin/sh -e

selection=$(hacksaw -f "-i %i -g %g")
shotgun $selection - | xclip -t 'image/png' -selection clipboard

To use with slop (as a replacement for maim -s):

#!/bin/sh -e

sel=$(slop -f "-i %i -g %g")
shotgun $sel "$1"

shotgun vs maim

• Only PNG and PAM are supported
• Does not attempt to wrap slop
• No cursor blending
• Defaults to a time-stamped file instead of dumping raw PNG data into your terminal (use - as the file name if you want to pipe output to something else)
• Most command-line flags were omitted
• The XShape extension is not supported
• shotgun is written in Rust, maim in C++
• The code base is kept as small and simple as possible (as much as Rust permits)

There are several reasons for omitting these features:

• Features that can be replaced trivially by external programs and wrapper scripts:
  • Use ImageMagick's convert and shotgun's -f pam for JPEG output
  • slop output is easy to process in a shell script
  • Use sleep instead of -d, since slop has to be called separately, this flag is not necessary
  • -x shouldn't even exist in the first place, set $DISPLAY instead
• I never use cursor blending, and I know that most users do not actually care for it
• -w (geometry relative to another window) is difficult to use and hardly useful, instead, shotgun always interprets the input geometry relative to the root window (maim's default is the captured window itself)
• There is rarely a reason to take a screenshot of an XShape window, most of them are special like slop's selection window or keynav's crosshair. Supporting XShape properly could add a significant amount of overhead, both in code length and performance, which are not desirable.

Performance

I've claimed that shotgun is twice as fast as maim, here's some supporting evidence (using hyperfine):

$ xrandr --fb 3840x2160
$ hyperfine --warmup 15 --min-runs 50 \
>     'maim > /dev/null' \
>     'shotgun - > /dev/null'
Benchmark #1: maim > /dev/null
  Time (mean ± σ):     629.3 ms ±   3.7 ms    [User: 570.7 ms, System: 52.1 ms]
  Range (min … max):   624.8 ms … 646.7 ms    50 runs

Benchmark #2: shotgun - > /dev/null
  Time (mean ± σ):     293.0 ms ±   3.2 ms    [User: 239.6 ms, System: 52.9 ms]
  Range (min … max):   287.8 ms … 298.2 ms    50 runs

Summary
  'shotgun - > /dev/null' ran
    2.15 ± 0.03 times faster than 'maim > /dev/null'

Further profiling has shown that the bottleneck in shotgun lies fully within the PNG encoder.

Going faster

The PNG encoder bottleneck can be avoided by using -f pam. This sets the output format to Netpbm PAM - an uncompressed binary image format.

By using an uncompressed format both encoding and decoding performance is improved:

Encoding

$ hyperfine --warmup 15 --min-runs 50 \
>     'shotgun -f png - > /dev/null' \
>     'shotgun -f pam - > /dev/null'
Benchmark #1: shotgun -f png - > /dev/null
  Time (mean ± σ):     294.5 ms ±   3.3 ms    [User: 240.0 ms, System: 54.1 ms]
  Range (min … max):   289.2 ms … 301.4 ms    50 runs

Benchmark #2: shotgun -f pam - > /dev/null
  Time (mean ± σ):     116.8 ms ±   2.8 ms    [User: 62.5 ms, System: 53.7 ms]
  Range (min … max):   113.8 ms … 122.7 ms    50 runs

Summary
  'shotgun -f pam - > /dev/null' ran
    2.52 ± 0.07 times faster than 'shotgun -f png - > /dev/null'

Decoding (using ImageMagick to convert to jpg)

$ hyperfine --warmup 15 --min-runs 50 \
>     'shotgun -f png - | convert - jpg:- > /dev/null' \
>     'shotgun -f pam - | convert - jpg:- > /dev/null'
Benchmark #1: shotgun -f png - | convert - jpg:- > /dev/null
  Time (mean ± σ):     600.7 ms ±   5.8 ms    [User: 506.4 ms, System: 96.5 ms]
  Range (min … max):   594.9 ms … 620.7 ms    50 runs

Benchmark #2: shotgun -f pam - | convert - jpg:- > /dev/null
  Time (mean ± σ):     350.4 ms ±   3.9 ms    [User: 217.0 ms, System: 139.3 ms]
  Range (min … max):   345.5 ms … 367.8 ms    50 runs

Summary
  'shotgun -f pam - | convert - jpg:- > /dev/null' ran
    1.71 ± 0.03 times faster than 'shotgun -f png - | convert - jpg:- > /dev/null'