Convert /proc/cpuinfo to a JSON in linux bash

Question

I'd like to convert the entire entry of a cat /proc/cpuinfo to a JSON of "key" : "value".

Is this possible? I tried to explode the string using \n in hopes that I could get each line and then I could explode each line using :.

But the newline character seems to produce inconsistent array.

Answer 1

Well, for a simple approach, you could just do:

sed 's/\(.*\)\t:\(.*\)/"\1" : "\2"/' /proc/cpuinfo 

That will match everything up to a tab followed by a colon and save it as \1, then everything after the colon and save it as \2. The replacement puts quotes around them.

That, however, results in cases like these:

"fpu    " : " yes"
power management:

Items with extra whitespace before the tab have their whitespace included and empty ones are ignored. This Perl version deals with those correctly:

perl -F: -alpe 's/.*/"$F[0]" : "$F[1]"/' /proc/cpuinfo 

This will split the line on : into the @F array (-F sets the character to split on and -a turns on automatic splitting into @F) and print each side quoted. It will break if you have more than one : on a line but I don't think that will ever happen in /proc/cpuinfo. However, it also prints any blank lines in the file. To avoid that, pipe it through grep first:

grep . /proc/cpuinfo | perl -F: -alpe 's/.*/"$F[0]" : "$F[1]"/' /proc/cpuinfo 

Or, only print if a line contains ::

perl -F: -alne 's/.*:.*/"$F[0]" : "$F[1]"/ && print' /proc/cpuinfo 

Answer 2

sed -n '/./s/ *\(\( *[^:[:blank:]]\)*\)[^:]*\(:*\)/"\1"\3/gp' /proc/cpuinfo

...works in sed. With GNU shortcuts you can write practically the same statement like:

sed -En 's/ *(( ?[^ :\t])+)\s*(:?)/"\1"\3/gp' /proc/cpuinfo

This is kind of lame (because I could do better with a few keystrokes and jw), but I was playing around w/ sed and...

set '   ' $'\\\n' $'\n' '      '
sed -En "\$c$2$1}$2]${3}1ccpus$1=$1[$2$1{$3/^$/c$2$1},$2$1{
        s/ *(( ?[^ :\t])+)(\s*:\s*)?\s*/\"\1\"\3/g
        s/\"(([yn]).s?|([0-9]+))\"$/\3false\2true/
        s/falsey|ntrue|([0-9])f\w+$/\1/
        /^.flags./{
                s/\s*:\s*/$1=$1[/;h
                s/.*\t.//;x;s/(.*\t.).*/\1/;x
                s/ /\",$2\"/g
                s/^/pr -to24 -a4 <<''$2/e;H;x
                s/$/$2$1$1]/
};      s/.*/$1$4&,/p" /proc/cpuinfo

I think the output is something close to valid. I took a picture because I can never seem to to get tabs to line up in the browser window as they do in the terminal window.

Most of the tabulations are cheated anyway - pr does the flags reformatting via the GNU sed e command, and for the rest I preserve as much of the originally formatted output as I might. For example, because I had to insert two chars - the double-quotes - around the name members, when I also insert leading tabulation I make it 1 actual tab followed by six spaces in a crude attempt at retaining the current tab-stop position.

Here's the first hash:

Answer 3

My version:

#!/usr/bin/env perl

use strict;
use warnings;

use JSON;
use Storable qw(dclone);

my ( %h, @cpu );

while (<>) {
    chomp;

    if (m/^$/o) {
        push @cpu, dclone( \%h );
        undef %h;
        next;
    }

    my ( $k, $v ) = split /\s*:\s*/, $_, 2;

    if ( !defined $v or $v eq '' ) { $h{$k} = undef }
    elsif ( $k eq 'flags' )        { $h{$k} = [ split /\s/, $v ] }
    elsif ( $v =~ /^\d+$/o )       { $h{$k} = int ($v) }
    elsif ( $v eq 'yes' )          { $h{$k} = \1 }
    elsif ( $v eq 'no' )           { $h{$k} = \0 }
    else                           { $h{$k} = $v }
}

print JSON->new->pretty->encode( \@cpu );

This assumes there's a blank line between each CPU definition.