How to urlencode data for curl command?

Here is the pure BASH answer.

Update: Since many changes have been discussed, I have placed this on https://github.com/sfinktah/bash/blob/master/rawurlencode.inc.sh for anybody to issue a PR against.

Note: This solution is not intended to encode unicode or multi-byte characters - which are quite outside BASH's humble native capabilities. It's only intended to encode symbols that would otherwise ruin argument passing in POST or GET requests, e.g. '&', '=' and so forth.

Very Important Note: DO NOT ATTEMPT TO WRITE YOUR OWN UNICODE CONVERSION FUNCTION, IN ANY LANGUAGE, EVER. See end of answer.

rawurlencode() {
  local string="${1}"
  local strlen=${#string}
  local encoded=""
  local pos c o

  for (( pos=0 ; pos<strlen ; pos++ )); do
     c=${string:$pos:1}
     case "$c" in
        [-_.~a-zA-Z0-9] ) o="${c}" ;;
        * )               printf -v o '%%%02x' "'$c"
     esac
     encoded+="${o}"
  done
  echo "${encoded}"    # You can either set a return variable (FASTER) 
  REPLY="${encoded}"   #+or echo the result (EASIER)... or both... :p
}

You can use it in two ways:

easier:  echo http://url/q?=$( rawurlencode "$args" )
faster:  rawurlencode "$args"; echo http://url/q?${REPLY}

[edited]

Here's the matching rawurldecode() function, which - with all modesty - is awesome.

# Returns a string in which the sequences with percent (%) signs followed by
# two hex digits have been replaced with literal characters.
rawurldecode() {

  # This is perhaps a risky gambit, but since all escape characters must be
  # encoded, we can replace %NN with \xNN and pass the lot to printf -b, which
  # will decode hex for us

  printf -v REPLY '%b' "${1//%/\\x}" # You can either set a return variable (FASTER)

  echo "${REPLY}"  #+or echo the result (EASIER)... or both... :p
}

With the matching set, we can now perform some simple tests:

$ diff rawurlencode.inc.sh \
        <( rawurldecode "$( rawurlencode "$( cat rawurlencode.inc.sh )" )" ) \
        && echo Matched

Output: Matched

And if you really really feel that you need an external tool (well, it will go a lot faster, and might do binary files and such...) I found this on my OpenWRT router...

replace_value=$(echo $replace_value | sed -f /usr/lib/ddns/url_escape.sed)

Where url_escape.sed was a file that contained these rules:

# sed url escaping
s:%:%25:g
s: :%20:g
s:<:%3C:g
s:>:%3E:g
s:#:%23:g
s:{:%7B:g
s:}:%7D:g
s:|:%7C:g
s:\\:%5C:g
s:\^:%5E:g
s:~:%7E:g
s:\[:%5B:g
s:\]:%5D:g
s:`:%60:g
s:;:%3B:g
s:/:%2F:g
s:?:%3F:g
s^:^%3A^g
s:@:%40:g
s:=:%3D:g
s:&:%26:g
s:\$:%24:g
s:\!:%21:g
s:\*:%2A:g

While it is not impossible to write such a script in BASH (probably using xxd and a very lengthy ruleset) capable of handing UTF-8 input, there are faster and more reliable ways. Attempting to decode UTF-8 into UTF-32 is a non-trivial task to do with accuracy, though very easy to do inaccurately such that you think it works until the day it doesn't.

Even the Unicode Consortium removed their sample code after discovering it was no longer 100% compatible with the actual standard.

The Unicode standard is constantly evolving, and has become extremely nuanced. Any implementation you can whip together will not be properly compliant, and if by some extreme effort you managed it, it wouldn't stay compliant.

Use curl --data-urlencode; from man curl:

This posts data, similar to the other --data options with the exception that this performs URL-encoding. To be CGI-compliant, the <data> part should begin with a name followed by a separator and a content specification.

Example usage:

curl \
    --data-urlencode "paramName=value" \
    --data-urlencode "secondParam=value" \
    http://example.com

See the man page for more info.

This requires curl 7.18.0 or newer (released January 2008). Use curl -V to check which version you have.

You can as well encode the query string:

curl --get \
    --data-urlencode "p1=value 1" \
    --data-urlencode "p2=value 2" \
    http://example.com
    # http://example.com?p1=value%201&p2=value%202

Another option is to use jq:

$ printf %s 'encode this'|jq -sRr @uri
encode%20this
$ jq -rn --arg x 'encode this' '$x|@uri'
encode%20this

-r (--raw-output) outputs the raw contents of strings instead of JSON string literals. -n (--null-input) doesn't read input from STDIN.

-R (--raw-input) treats input lines as strings instead of parsing them as JSON, and -sR (--slurp --raw-input) reads the input into a single string. You can replace -sRr with -Rr if your input only contains a single line, or if you don't want to replace linefeeds with %0A:

$ printf %s\\n 'multiple lines' 'of text'|jq -Rr @uri
multiple%20lines
of%20text
$ printf %s\\n 'multiple lines' 'of text'|jq -sRr @uri
multiple%20lines%0Aof%20text%0A

Or this percent-encodes all bytes:

xxd -p|tr -d \\n|sed 's/../%&/g'

Use Perl's URI::Escape module and uri_escape function in the second line of your bash script:

...

value="$(perl -MURI::Escape -e 'print uri_escape($ARGV[0]);' "$2")"
...

Edit: Fix quoting problems, as suggested by Chris Johnsen in the comments. Thanks!