aboutsummaryrefslogtreecommitdiff
path: root/json-format.pl
diff options
context:
space:
mode:
Diffstat (limited to 'json-format.pl')
-rwxr-xr-xjson-format.pl154
1 files changed, 154 insertions, 0 deletions
diff --git a/json-format.pl b/json-format.pl
new file mode 100755
index 0000000..571e89e
--- /dev/null
+++ b/json-format.pl
@@ -0,0 +1,154 @@
+#!/usr/bin/env perl
+use strict;
+use warnings;
+use Term::ANSIColor;
+use Getopt::Long;
+
+my $reset = color('reset');
+my %format = (
+ FIELD => color('bold'),
+ NULL => color('cyan'),
+ BOOL => color('red'),
+ NUMBER => color('magenta'),
+ STRING => color('blue'),
+ ERROR => color('bold white on_red'),
+);
+
+my ($color, $keep_ws, $help) = 'auto';
+if (!GetOptions('color=s' => \$color, 'keep-ws' => \$keep_ws, 'help' => \$help)
+ || $help) {
+ print STDERR
+ "Usage: $0 [OPTION...] [FILE...]\n" .
+ "Pretty-print and colorify JSON\n" .
+ "\n" .
+ " --help print this help\n" .
+ " --keep-ws retain all original whitespace\n" .
+ " --color=COLOR 'always', 'never' or 'auto' (the default)\n";
+ exit 2;
+}
+
+%format = ()
+ if $color eq 'never' || $color eq 'auto' && !-t STDOUT;
+
+# Hash lookup is the fastest way to qualify tokens, however it cannot be used
+# for everything and we need to fall back to regular expressions
+my %lookup = (
+ '[' => 'LBRACKET', '{' => 'LBRACE',
+ ']' => 'RBRACKET', '}' => 'RBRACE',
+ ':' => 'COLON', ',' => 'COMMA',
+ 'true' => 'BOOL', 'false' => 'BOOL', 'null' => 'NULL',
+);
+my @pats = (
+ ['"(?:[^\\\\"]*|\\\\(?:u[\da-f]{4}|["\\\\/bfnrt]))*"' => 'STRING'],
+ ['-?\d+(?:\.\d+)?(?:[eE][-+]?\d+)?' => 'NUMBER'],
+ ['[ \t\r\n]+' => 'WS'],
+);
+my @tokens = map {[qr/^$_->[0]$/s, $_->[1]]} @pats;
+
+# m//g is the fastest way to explode text into tokens in the first place
+# and we need to construct an all-encompassing regular expression for it
+my @all_pats = map {$_->[0]} @pats;
+push @all_pats, quotemeta for keys %lookup;
+my $any_token = qr/\G(${\join '|', @all_pats})/;
+
+# FIXME: this probably shouldn't be a global variable
+my $indent = 0;
+
+sub nexttoken ($) {
+ my $json = shift;
+ if (!@$json) {
+ return unless defined (my $line = <>);
+ push @$json, $line =~ /$any_token/gsc;
+ push @$json, substr $line, pos $line
+ if pos $line != length $line;
+ }
+
+ my $text = shift @$json;
+ if (my $s = $lookup{$text}) {
+ return $s, $text;
+ }
+ for my $s (@tokens) {
+ return $s->[1], $text if $text =~ $s->[0];
+ }
+ return 'ERROR', $text;
+}
+
+sub gettoken ($) {
+ my $json = shift;
+ while (my ($token, $text) = nexttoken $json) {
+ next if !$keep_ws && $token eq 'WS';
+ return $token, $text;
+ }
+ return;
+}
+
+sub printindent () {
+ print "\n", ' ' x $indent;
+}
+
+sub do_value ($$$);
+sub do_object ($) {
+ my $json = shift;
+ my $in_field_name = 1;
+ my $first = 1;
+ while (my ($token, $text) = gettoken $json) {
+ if ($token eq 'COLON') {
+ $in_field_name = 0;
+ } elsif ($token eq 'COMMA') {
+ $in_field_name = 1;
+ } elsif ($token eq 'STRING') {
+ $token = 'FIELD' if $in_field_name;
+ }
+ if ($token eq 'RBRACE') {
+ $indent--;
+ printindent unless $keep_ws;
+ } elsif ($first) {
+ printindent unless $keep_ws;
+ $first = 0;
+ }
+ do_value $token, $text, $json;
+ return if $token eq 'RBRACE';
+ }
+}
+
+sub do_array ($) {
+ my $json = shift;
+ my $first = 1;
+ while (my ($token, $text) = gettoken $json) {
+ if ($token eq 'RBRACKET') {
+ $indent--;
+ printindent unless $keep_ws;
+ } elsif ($first) {
+ printindent unless $keep_ws;
+ $first = 0;
+ }
+ do_value $token, $text, $json;
+ return if $token eq 'RBRACKET';
+ }
+}
+
+sub do_value ($$$) {
+ my ($token, $text, $json) = @_;
+ if (my $format = $format{$token}) {
+ print $format, $text, $reset;
+ } else {
+ print $text;
+ }
+ if ($token eq 'LBRACE') {
+ $indent++;
+ do_object $json;
+ } elsif ($token eq 'LBRACKET') {
+ $indent++;
+ do_array $json;
+ } elsif ($token eq 'COMMA') {
+ printindent unless $keep_ws;
+ } elsif ($token eq 'COLON') {
+ print ' ' unless $keep_ws;
+ }
+}
+
+my @buffer;
+while (my ($token, $text) = gettoken \@buffer) {
+ do_value $token, $text, \@buffer;
+ print "\n" unless $keep_ws;
+}