By default, it does one pass with command 1, one pass with command 2, then one pass with command 3. But some commands alter what happens after them, like /..../d. So that preprocessor lines are (1) printed unaltered, then (2) deleted, preventing them getting being mangled by anything happening in step (3). That sounds like a bad combination... again, those tools are line-based and this junk is NOT. If you slice out bits and pieces instead of operating on the entire line, how're you going to put it back together?

regular expressions and backreferences. You use brackets to match the parts you need.

Code:

# matches and replaces the entire string with nothing.
$ echo "int a(int b, int c)" | sed 's/\([a-z]* [a-z]*\)(\([a-z]* [a-z]*\)\(, [a-z]* [a-z]*\))//g'

# prints first match.
$ echo "int a(int b, int c)" | sed 's/\([a-z]* [a-z]*\)(\([a-z]* [a-z]*\)\(, [a-z]* [a-z]*\)*)/\1/g'
int a
# prints second match
$ echo "int a(int b, int c, int d)" | sed 's/\([a-z]* [a-z]*\)(\([a-z]* [a-z]*\)\(, [a-z]* [a-z]*\)*)/\2/g'
int b
# prints third match
$ echo "int a(int b, int c)" | sed 's/\([a-z]* [a-z]*\)(\([a-z]* [a-z]*\)\(, [a-z]* [a-z]*\)*)/\3/g'
, int d
$

Note how the third match only matches the last thing in (...)* when there's more than one. I think you'd have to nest it like ( (...)* ) to get the entire block.

You can use that in awk too,

Code:

match("string", /regex/, array)

and the resulting matches fill array from 1 to n. I'm not sure whether you need to escape () like \( \) in awk or not.