replace shell+perl canonicalise-translations.sh with a Python script

The Python script is faster, because it processes all the files with a single
process rather than spawning a new perl process for each translation JSON file.

Of course, it could have been written in perl and got the same speedup, but I
don't know perl so Python was easier for me.

scripts/canonicalise-translations.sh

@@ -1,9 +0,0 @@
-#!/bin/sh
-
-# XXX this assumes you have push access to github master and transifex, and
-# assume paths. you'll want to hack it before trying to use it yourself
-
-set -x
-set -e
-find data/lang -name \*.json -not -name en.json | while read f ; do cat $f | perl -MJSON -e 'undef $/;$j=JSON->new->pretty->utf8->indent->canonical;print $j->encode($j->decode(<>))' | sponge $f ; done
-exit 0

scripts/canonicalise_translations.py

@@ -0,0 +1,32 @@
+#!/usr/bin/env python3
+# vim: set ts=8 sts=4 sw=4 expandtab autoindent fileencoding=utf-8:
+
+import json
+import os
+import sys
+
+def read_translation_file(path):
+    """Reads a JSON translation file."""
+    with open(path, 'r', encoding='utf-8') as fl:
+        return json.load(fl)
+
+def write_translation_file(path, data):
+    """Writes a JSON translation file with canonical formatting."""
+    with open(path, 'w', encoding='utf-8') as fl:
+        json.dump(data, fl,
+                  ensure_ascii=False,
+                  indent=3,
+                  separators=(',',' : '),
+                  sort_keys=True)
+        fl.write('\n')
+
+def main():
+    for dirpath, _, filenames in os.walk('data/lang'):
+        for path in (os.path.join(dirpath, fname)
+                     for fname in filenames if fname.endswith('.json')):
+            print('Canonicalising', path, '...')
+            data = read_translation_file(path)
+            write_translation_file(path, data)
+
+if __name__ == '__main__' and not sys.flags.interactive:
+    main()