Question

I have a shell variable:

all_apk_file="a 1 2.apk x.apk y m.apk"

I want to replace the a 1 2.apk with TEST, using the command:

echo $all_apk_file | sed 's/(.*apk ){1}/TEST/g'

The .*apk means end with apk, {1} means only match one time, but it doesn’t work; I only got the original variable as output: a 1 2.apk x.apk y m.apk

Can anyone tell me why?

Answer 1

One part of the problem is that in regular sed, the () and {} are ordinary characters in patterns until escaped with backslashes. Since there are no parentheses in the variable’s value, the regex never matches. With GNU sed, you can also enable extended regular expressions with the -r flag. If you fix that problem, you will then run into the problem that .* is greedy, and the g modifier actually doesn’t change anything:

$ echo $all_apk_file | sed 's/\(.*apk \)\{1\}/TEST/g'
TESTy m.apk
$ echo $all_apk_file | sed -r 's/(.*apk ){1}/TEST/g'
TESTy m.apk
$ echo $all_apk_file | sed -r 's/(.*apk ){1}/TEST/'
TESTy m.apk
$

It only stops there because there isn’t a space after m.apk in the echoed value of the variable.

The issue now is: what is it that you want replaced? It sounds like ‘everything up to and including the first occurrence of apk at the end of a word. This is probably most easily done with trailing context or non-greedy matching as found in Perl regular expressions. If switching to Perl is an option, do so. If not, it is not trivial in normal sed regular expressions.

$ echo $all_apk_file | sed 's/^[^.]* [^.][^.]*\.apk /TEST /'
TEST x.apk y m.apk
$

This looks for anything without dots in it, followed by a blank, followed by no dots again, and .apk; this means that the first dot allowed is the one in 2.apk. It works for the sample data; it would not work if the variable contained:

all_apk_file="a 1.2 2.apk m.apk y.apk 37"

You’ll need to tune this to meet your requirements.