mjhjmtu

Using source-highlight in conjunction with less works really well, but I'm struggling to see how I can use it with git show because there's no file extension to clue source-highlight about the language used in the git output...

less somefile.rb # result is syntax colourised
git show master:somefile.rb | less # no colouring

Using calling git show master:somefile.rb | less is effectively the same as calling less somefile (ie no .rb); because there's no extension, there's no way for source-highlight to guess the syntax.

Is there a non-extension way for source-highlight to guess, or can I pass a --lang-def option into the LESSOPEN variable somehow?

Edit 1
Ah, so it looks like source-highlight can use other methods to infer the language but I don't have any of those in my source files.

source-highlight can guess the language without using the extension. See 6.1 How the input language is discovered.

Unfortunately, that feature requires that the input is a regular file. When I try to use it to recognize stdin, I get this error:

source-highlight: missing feature: language inference requires input file

So it's not available in the middle of a pipeline, which is how it would be invoked when formatting git output.

git does have some options for configuring its pager -- see the documentation for core.pager and pager.<cmd>. But there's no mechanism for passing on additional information (such as the filename) to the pager -- git-show only has formatting options when showing commits. The "pager command" you specify would have to be an additional wrapper script, that performs the following steps:

1. reads the entirety of the input and saves it in a temporary file


2. calls source-highlight ... on that file


3. pipes the output to your actual pager

And this would still fail on files that don't have an appropriate "sha-bang" to identify them.

I ended up modifying gnu source-highlight's src-hilite-lesspipe.sh so it could work with piped files: https://gist.github.com/jaygooby/9494858d3d481a64819d227a9318f6c7

This means that as well invoking in a regular fashion:

less code.py

You can also get source highlighting for extensionless files that don't have any of the language inference features that source-highlight uses:

less /tmp/mycode

And (the original motivation for doing this) piped files:

cat /tmp/file.rb | less
git show master:obfusicated.perl # implicit pipe to less via git's pager

If you know your code uses a syntax that source-highlight doesn't have any definition files for, you can override the guess, by setting a similar language. c is often a good fallback:

SRCLANG=c git show master:app/views/layouts/application.html.erb

Here's a raw copy of the gist that lives at: https://gist.github.com/jaygooby/9494858d3d481a64819d227a9318f6c7

#! /bin/bash
#
# Based on http://git.savannah.gnu.org/cgit/src-highlite.git/tree/src/src-hilite-lesspipe.sh.in
# by Lorenzo Bettini
#
# Modified by Jay Caines-Gooby to support piped files
# jay@gooby.org
# @jaygooby
#
# Typically called by setting:
#
# export LESSOPEN="|-/path/to/src-hilite-lesspipe.sh %s"
# export LESS=-R
#
# If we're less-ing a file, %s will be replaced by the name of the file. If
# there's no file and we're reading from a pipe, then %s is set to -
#
# This script differs from the original src-hilite-lesspipe.sh
# in that it can handle pipes and files with no extensions and will
# attempt to guess their language using the file command.
#
# So as well as invoking on regular files:
#
# less some.rb
# less some.py
#
# It will should be able to work on:
#
# less no-extension-but-contains-perl
#
# and even with more complex examples (my original motivation
# https://unix.stackexchange.com/questions/469982/how-can-i-use-source-highlight-with-git-show)
#
# git show master:some.rb
#
# It uses bashisms to do this, so is no longer a pure POSIX sh script.
set -eu

# Users can override the guessed language by setting SRCLANG:
# SRCLANG=c git show master:app/views/layouts/application.html.erb
SRCLANG=$SRCLANG:-

guess_language() file - 

# check if the language passed as $1 is known to source-highlight
# In an earlier version of this script I set a fallback (c.lang)
# but this causes issues with paging man pages etc
check_language_is_known() 
 fallback=""
 lang=$(source-highlight --lang-list 

for source in "$@"; do
 case $source in
 *ChangeLog|*changelog)
 source-highlight --failsafe -f esc --lang-def=changelog.lang --style-file=esc.style -i "$source" ;;
 *Makefile|*makefile)
 source-highlight --failsafe -f esc --lang-def=makefile.lang --style-file=esc.style -i "$source" ;;
 *.tar|*.tgz|*.gz|*.bz2|*.xz)
 lesspipe "$source" ;;
 *)

 # naive check for a file extension; let source-highlight infer language
 # but only when source isn't - (ie. from a piped file)
 if [[ "$source" != "-" && $(basename "$source") =~ . ]]; then
 source-highlight --failsafe --infer-lang -f esc --style-file=esc.style -i "$source"
 else
 # We're being piped to, or the filename doesn't have an extension
 # so guess the language.

 # When we're being piped to, we cat stdin, but when it's a file
 # without an extension, we cat the file instead.

 # unset IFS so line breaks are preserved and capture the file's contents
 # (will only work for files up to bash's available memory). There should
 # be a better way to replicate this with tee or process substitution...
 IFS= file=$([ "source" = "-" ] && cat || cat "$source")
 lang=$(guess_language $file)
 lang=$(check_language_is_known $lang)

 # Don't call if source-highlight doesn't know the language
 # BUT also let users override the guessed lang if the environment
 # variable SRCLANG is set. This can help where you know e.g. your
 # source code is c-like, but source-highlight has no specific syntax
 # definition for your code
 [ -n "$SRCLANG" ] && lang="$SRCLANG"

 if [ -n "$lang" ]; then
 echo $file | source-highlight --failsafe -f esc --src-lang=$lang --style-file=esc.style
 else
 echo $file
 fi
 fi

 ;;
 esac
done