Keep matching pattern in shell parameter expansion

Question

I can remove a pattern in a bash variable using ${variable##pattern} (leading) or ${variable%%pattern} (trailing).

But I can't find a bash-only way to keep the pattern and throw the rest.

I know there are solutions using sed, awk, or grep, but I want to know if there is a reasonably efficient bash-only solution that I am overlooking?

PS: This isn't just an idle question. The initial problem is that I want to process files where the name contains a pattern (technically, '[A-Z]+([A-Z])-[0-9][0-9]+([0-9])': capital letters followed by a dash and digits) and I want to use the same pattern to list the files and extract the matching string for further processing.

Answer 1

${var%"${var##pattern}"}
${var#"${var%%pattern}"}

Example:

$ k='ab*10cd20ef*'
$ echo "${k%"${k##*[0-9]}"}"
ab*10cd20
$ echo "${k#"${k%%[0-9]*}"}"
10cd20ef*

Note the quotes are important to prevent the shell from interpreting the expansions as a pattern. Try echo "${k#${k%%[0-9]*}}" to see it outputs an incorrect result.

Answer 2

In Bash you could also use regexes:

#!/bin/bash
re='[A-Z][A-Z]+-[0-9][0-9][0-9]+'
file=foo-BAR-1234.txt
if [[ $file =~ $re ]]; then
    echo "filename '$file' matches, matching part is '${BASH_REMATCH[0]}'"
fi

With file=foo-BAR-1234.txt, that would match the part BAR-1234 and print accordingly. You could also use parenthesis in the regex to capture part of the pattern, they'd be available in ${BASH_REMATCH[1]} etc.

Of course, note that the format for regexes is different than that of Bash/Ksh extended globs: instead of +([abc]), you need [abc]+, or ([abc])+, the parenthesis being optional when the asterisk is applied to just that one bracket group. Same for * and ?. Also, you could write e.g. [0-9]{3,} for three or more digits.

Answer 3

Getting only the shell-pattern match by a single variable-expansion:

A bit late, but might be useful for someone else. Example: I constructed this to extract the Episode numbering from a TV-Series text. (which is a shell-pattern, with unknown numbers)

The following code has some explaining comments.

#!/bin/bash
    
#This is for extended globbing
shopt -s extglob

#Getting only the Shell-patternmatch by variable-expansion
Var="Text before match_S02E12_Text after match"

#We only want the pattern from TV-Series Episode
MatchPattern="S[0-9][0-9]E[0-9][0-9]"

OnlyTheMatch="${Var//@(${Var//@($MatchPattern)*/}|${Var//*@($MatchPattern)/})/}"

echo "${OnlyTheMatch}" > Output_OnlyTheMatchPattern.txt

#Unset extended globbing
shopt -u extglob

#This does de following:
#We want to 'replace' the text before and the text after with 'nothing'  ${Var//match/}
#(and by that remove them)

#But the  ${Var//match/} can remove only one match, 
#but we want to remove multiple (the text before AND after)
#To use multiple match-patterns we must put them in @(..|..) separated by vertical bars  '|'
#The @ in @(pattern) means: Match exactly one occurrence of the pattern
#OnlyTheMatch="${Var//@(TextBeforeMatch|TextAfterMatch)/}"

#The Text before the match can be found by Var Expansion 'replace':
# ${Var//@($MatchPattern)*/} 
#(Here the * matches everything after the MatchPattern, leaving only the text before the match)

#The Text after the match can also be found by Var Expansion 'replace':
# ${Var//*@($MatchPattern)/} 
#(Here the * matches everything before the MatchPattern, leaving only the text after the match)

#And those variables can both be written in the  @(..|..) separated by vertical bars  '|'
#Together making the single variable-expansion to only get the matching pattern.

So, in case of the OP to "..keep the pattern and throw (away) the rest." :

MatchPattern="[A-Z]+([A-Z])-[0-9][0-9]+([0-9])"

Maybe this comes close to what you are looking for.