Extract version string from filename

Question

We have a list of filenames in file a.txt with their version string embedded in them. Eg:

gson-2.1
xmlParserAPIs-2.4.0
acrobat-1.1
orai18n-mapping-12.1.0.2
jdbc-se2.0
eclipse-core-runtime-20070801
trove-2.0.1
antisamy-1.3
javax.annotation
dojo-4342
org.json-0.0.1
castor-1.2-jdo

We tried cat a.txt | tr -d "[:alpha:]-_" | less, but it doesn't look right.

eg
2.1
2.4.0
1.1
1812.1.0.2   <--- wrong
2.0
20070801
2.0.1
1.3
.
4342
.0.0.1
1.        <--- wrong

Any help is appreciated.

Answer 1

It isn't possible to perfectly match all of those strings as you've listed them as there's no way to tell the difference between something like "orai18n-" and "-se2.0". If you create a regex that looks for strings of digits and dots that begin with a dash, you'll match all but the "java-se2.0" string pretty well:

sed 's/.*-\([0-9\.][0-9\.]*\).*/\1/'

(Depending on your version, you can use sed -r to allow the use of [0-9.]+)

This produces an output where all recognised versions are extracted and the entire string displayed where nothing suitable is found:

2.1
2.4.0
1.1
12.1.0.2
jdbc-se2.0
20070801
2.0.1
1.3
javax.annotation
4342
0.0.1
1.2

Answer 2

You can use grep:

grep -oP '(?<=-)([0-9]+\.?)+' a.txt

That extracts all version numbers. If a line contains no version (for example javax.annotation) nothing is printed.

The regex:

• (?<=-): first look for a dash (-), but it should not be a part of the match
• [0-9]+: search for numbers, they should appear at least one or multiple times
• ([0-9]+\.?)+: at dot (.) can be present or not, and all that must occure at least once.

Answer 3

You can try the small script below:

cat a.txt | sed 's/[-/a-zA-Z]//g'