With the internet being full of silly but interesting things I would have thought this would be an easy find but I couldn’t find it anywhere – a couple of people asking about it with some vague bits of code here and there but most of the time people would say something dumb like “use (awk|perl|uudecode)”<\/p>\n
Not because I think it’s particularly useful or sensible (it certainly isn’t that, since it takes around 200 times as long as the C version!) but just to see if it were possible, here’s my bash uudecode implementation.<\/p>\n
#!\/bin\/bash\r\nbs=0\r\nwhile read -rs t ; do\r\n if [ $bs -eq 1 ] ; then\r\n if [ \"a$t\" = \"aend\" ] ; then\r\n bs=2\r\n else\r\n x=1\r\n i=($(printf \"%d \" \"'${t:0:1}\" \"'${t:1:1}\" \"'${t:2:1}\" \"'${t:3:1}\" \"'${t:4:1}\" \"'${t:5:1}\" \"'${t:6:1}\" \"'${t:7:1}\" \"'${t:8:1}\" \"'${t:9:1}\" \"'${t:10:1}\" \"'${t:11:1}\" \"'${t:12:1}\" \"'${t:13:1}\" \"'${t:14:1}\" \"'${t:15:1}\" \"'${t:16:1}\" \"'${t:17:1}\" \"'${t:18:1}\" \"'${t:19:1}\" \"'${t:20:1}\" \"'${t:21:1}\" \"'${t:22:1}\" \"'${t:23:1}\" \"'${t:24:1}\" \"'${t:25:1}\" \"'${t:26:1}\" \"'${t:27:1}\" \"'${t:28:1}\" \"'${t:29:1}\" \"'${t:30:1}\" \"'${t:31:1}\" \"'${t:32:1}\" \"'${t:33:1}\" \"'${t:34:1}\" \"'${t:35:1}\" \"'${t:36:1}\" \"'${t:37:1}\" \"'${t:38:1}\" \"'${t:39:1}\" \"'${t:40:1}\" \"'${t:41:1}\" \"'${t:42:1}\" \"'${t:43:1}\" \"'${t:44:1}\" \"'${t:45:1}\" \"'${t:46:1}\" \"'${t:47:1}\" \"'${t:48:1}\" \"'${t:49:1}\" \"'${t:50:1}\" \"'${t:51:1}\" \"'${t:52:1}\" \"'${t:53:1}\" \"'${t:54:1}\" \"'${t:55:1}\" \"'${t:56:1}\" \"'${t:57:1}\" \"'${t:58:1}\" \"'${t:59:1}\" \"'${t:60:1}\"))\r\n l=$[${i[0]} -32 & 63 ]\r\n while [ $l -gt 0 ] ; do\r\n i0=$[${i[$[x++]]} -32 & 63]\r\n i1=$[${i[$[x++]]} -32 & 63]\r\n i2=$[${i[$[x++]]} -32 & 63]\r\n i3=$[${i[$[x++]]} -32 & 63]\r\n if [ $l -gt 2 ] ; then\r\n echo -ne \"\\0$[$i0 >> 4]$[$i0 >> 1 & 7]$[$i0 << 2 & 4 | $i1 >> 4]\\0$[$i1 >> 2 & 3]$[$i1 << 1 & 6 | $i2 >> 5]$[$i2 >> 2 & 7]\\0$[$i2 & 3]$[$i3 >> 3 & 7]$[$i3 & 7]\"\r\n elif [ $l -eq 2 ] ; then\r\n echo -ne \"\\0$[$i0 >> 4]$[$i0 >> 1 & 7]$[$i0 << 2 & 4 | $i1 >> 4]\\0$[$i1 >> 2 & 3]$[$i1 << 1 & 6 | $i2 >> 5]$[$i2 >> 2 & 7]\"\r\n else\r\n echo -ne \"\\0$[$i0 >> 4]$[$i0 >> 1 & 7]$[$i0 << 2 & 4 | $i1 >> 4]\"\r\n fi\r\n l=$[l-3]\r\n done\r\n fi\r\n elif [ \"${t:0:5}\" = \"begin\" ]; then\r\n bs=1\r\n fi\r\ndone\r\n<\/pre>\nNote that I’ve used as few subprocesses as possible – I’ve got it down to one per line of input (the stupidly long “printf” line – more about that later) because the thing that bash does really<\/em> badly is spawn off a subprocess: you can do a huge<\/strong> amount of fairly complex string and number manipulation in bash in place of a single subprocess spawn and it will still cut the time used massively. So for example I was using (as recommended across the web)<\/p>\n
\nh=$(printf \"%X\" $d)
\n<\/code><\/p>\nto convert decimal-to-hex: it’s about 10 times quicker (and actually not much less obvious) to create an array (0 1 2 3 4 5 6 7 8 9 a b c d e f) and build the hex string yourself using ${arr[d>>4]}${arr[d&15]} (I suppose using a 256 entry array would actually be quicker still but I gave up on the hex thing anyway to use octal)<\/p>\n
The most interesting thing (fairly obvious, when you think about it) is the speed increase when you change from dripping through converting character by character<\/p>\n
\nc = `printf \"%d\" \"'$c\"`
\n<\/code><\/p>\nto the massive and horrible one-line-at-a-time printf above. You’re talking about a 15x speedup for the entire operation just by doing that.<\/p>\n
Anyway, I found it all very challenging; I hope you find it useful\/interesting :-).<\/p>\n
Feel free to tell me what an idiot I am or (if you’re feeling more constructive) suggesting optimisations. If you can figure out a way of getting a character into a charcode integer without using a subprocess then obviously that would be really<\/em> useful…<\/p>\n