sed matching and reformatting a string with a keyword, a random word and a known symbol

Question

After a couple of hours trying to get my sed query to work, I'm about to give up!

I have the following string, extracted from source code - the aim is to generate prototypes for a lot of old and undocumented code. For example:

function foo(bar=1);

I want to end up with something like :

function foo(
    bar=1)

I want to match all lines that start with function followed by a random alphanumerical (and - and _) word, and add line breaks and tabs with sed.

My problem is that sed is greedy by default, and I cannot get sed to add a line break after the first ( due to regex greedyness.

So something like this hardcoded works:

echo 'function foo(bar=true)' | sed 's:\(function foo(\)\(.*\):\1\n\t\2:g'

which gives me the expected output:

function foo(
    bar=true)

I can modify this to the following, adding the newline after the ( character:

echo 'function foo(bar=true)' | sed 's:\(function.*(\)\(.*\):\1\n\t\2:g'

Which gives the same expected result as before - until I get to a function in the code that contains an array as a default value for an argument - which is where the greedy regex trips me up:

echo 'function foo(bar=array())' | sed 's:\(function.*(\)\(.*\):\1\n\t\2:g'

Which actually gives:

function foo(bar=array(
    ))

The greedyness causes the newline and tab to be added after the last ( and not the first. Unfortunately sed does not support non greedy regex which would solve everything immediately...

So I tried trying to make a regex that would do something like below, but I'm not getting anywhere:

• 's:\(function [\w+]\)\(.*\):\1\n\2:g' trying to extract an alpha numericals which should match everything to the first (
• Doing the same trying using an :alnum: class to match the word
• Doing the same replacing with the more sed friendly [A-Za-z0-9_-] but having difficulties understanding how to get this to match more than 1 character in the pattern, so it takes the word up to the first ( - and then get the rest in the second return.

It seems that these character classes are just being ignored in the query, and I'm out of ideas.

As I cannot make sed non-greedy, how can I match a string which is of the format

KnownKeyword SomethingRandomAlphaNumerical-_(SomethingElse())

into a string that after line breaking after the first (, will look like:

KnownKeyword SomethingRandomAlphaNumerical-_(
SomethingElse())

Where am I going wrong? What pattern would accomplish this?

Answer 1

Do not use a dot . as the "any character", use a negative character match [^(]. So, your regex would be:

$ echo 'function foo(bar=array())' | sed 's:\(function[^(]*(\)\(.*\):\1\n\t\2:g'

function foo(
    bar=array())

A negative match will match any character except the ones inside the brackets after the initial ^ (and except newlines). What this means is that a [^(] could be understood as the match "not (". Then there is an * which means repeating as many times as possible, it is still greedy but won't match a (. In short: it will match every character up to the next (. This technique is to limit the greediness of .* by giving it a limiting character.

Answer 2

Match lines starting with function, sub the first opening parenthesis, add a new line and tab using GNU sed

$ sed '/^function/s/(/&\n\t/' input_file
function foo(
    bar=array())

Answer 3

To match your requirements precisely, the function identifier[1] is (alphanumeric or underscore or hyphen) characters:

sed -E 's:function[[:blank:]]+[[:alnum:]_-]+\(:&\n\t:' file`

there's no need to capture anything, just match "function" up to the open parentheses, and use & to refer to the matched text in the replacement string.

[1] Although many language restrict the identifier: the first character is limited to alpha or underscore.