regex - SED script not matching single in Multiline pattern after line breaks -

i trying produce sed script converts

&&a_x* &&b_x;cx &&d_x* 

into

a_x ax b_x  cx  d_x dx 

a * should trigger duplication removed _, ; simple linebreak.

i have sed script first inserts linebreaks (including operation ;) , executes multiple line pattern duplication without _.

the multiple line pattern works if move seperate script file , pipe output of instructions executing linebreaks.

for strange reason single script file won't - want maintenance reasons.

here's combined version:

#!/bin/sed -f # remove whitespaces  s/\ //g  # linebreak on && s/\&\&/\ \&\&/g ### linebreak on ;  s/\;/\ /g # remove new line s/\n//  :extendvars /^..*\*$/ {     l                         //debug switch     h      s/\(\&\|\*\)\(\&\|\*\)*//g     p     g     s/\(\&\|_\|-\|\*\)\(\&\|_\|-\|\*\)*//g     p     d     bextendvars; } 

the debug switch 'l' in first line of multiline pattern should match lines ending * matches lines , outputs

&&a_x*\n&&b_x\bx\n&&c_x*$ 

in faulty, combined version. when piping it, sed correctly recognizes pattern:

&&a_x*$ ... &&c_x*$ 

faulty output (combined version):

&&a_x*\n&&b_x\nbx\n&&c_x*$  a_x  b_x  bx  c_x ax  bx  bx  cx 

correct ouput (piped version):

&&a_x*$ a_x  ax  &&b_x  bx  &&c_x*$ c_x cx 

i run script with

sed -f [scriptname] <old >new 

in version haved removed && &&b_x jet.

how can sed recognize correct pattern if execute statements in 1 script? why sed fail match single line ending *?

thanks help!

the reason code never loops again doesn't have loop condition; it's inside loop, run

    d 

...which aborts processing of current input line. constructed several lines in pattern space input line of no consequence; d tells sed stop doing, read next line of input (if there one) , start on that.

anyway, approach seems overly complicated me. i'd suggest (in gnu parlance, because mechanism more obvious in gnu sed code)

#!/bin/sed -rf  s/\s*(^|&&|;)\s*/\n/g      # split tokens onto several lines, make sure                            # there's newline in front of each (so next                            # regex matches all) s/(\n[^\n])_x\*/\1_x\1x/g  # match lines end _x*, expand                            # \nfoo_x\nfoox s/^\n*//                   # remove leading newlines (we put @ least 1                            # there in beginning) 

you seem have taken great pains make code work non-gnu sed, here's posix version same thing:

#!/bin/sed -f  s/[[:space:]]*&&[[:space:]]*/\ /g s/[[:space:]]*;[[:space:]]*/\ /g s/^/\ / s/\(\n[^\n]\)_x\*/\1_x\1x/g s/^\ *// 

this removes whitespaces around tokens. seemed sensible thing do. if don't want happen, space-matching parts have removed code, , provisions have made whitespace @ end of token line.

#!/bin/sed -rf  s/^|&&|;/\n/g s/(\n[^\n])_x\*([[:blank:]]*)/\1_x\1x\2/g s/^\n// 

is possible adaptation of gnu sed code.