Boolean OR in sed regex

Question

I'm trying to replace all references of a package named boots in a configuration file.

The line format is add fast (package OR pkg) boots-(any-other-text), e.g.:

add fast package boots-2.3
add fast pkg boots-4.5

I want to replace it with:

add yinst pkg boots-5.0

I've tried the following sed commands:

sed -e 's/add fast (pkg\|package) boots-.*/add yinst pkg boots-5.0/g'
sed -e 's/add fast [pkg\|package] boots-.*/add yinst pkg boots-5.0/g'

What's the right regex? I think I'm missing something in the boolean or (package or pkg) part.

Solution 1

sed -e 's/add fast \(pkg\|package\) boots-.*/add yinst pkg boots-5.0/g'

You could always avoid the OR by doing it twice

sed 's/add fast pkg boots-.*/add yinst pkg boots-5.0/g
s/add fast package boots-.*/add yinst pkg boots-5.0/g'

Solution 2

Use extended regex mode, and don't escape the |.

sed -E -e 's/add fast (pkg|package) boots-.*/add yinst pkg boots-5.0/g'

Solution 3

You're mixing BRE's and ERE's either escape both () and | or none.

sed uses basic regular expressions by default, enabling use of extended regular expressions is implementation dependent, e.g. with BSD sed you use the -E switch, GNU sed has it documented as -r, but -E works as well.

Solution 4

GNU (Linux):

1) Make following random strings

   cidr="192.168.1.12"
   cidr="192.168.1.12/32"
   cidr="192.168.1.12,8.8.8.8"

to blank

2) sed with -r for using the logical operators in GNU like @Thor mentioned, and -i to on the fly edit a file on match found

$ echo '<user id="1000" cidr="192.168.1.12">' > /tmp/1000.xml
$ sed -r -i \ 
  s/'cidr="192.168.1.12\/32"|cidr="192.168.1.12"|192.168.1.12,'/''/ /tmp/1000.xml

-r = GNU sed
-i = search / match/ edit the changes to the file on the fly

View more solutions

Comments

[Adam Matan]

sed: invalid option -- E. Doesn't seem to work with GNU sed version 4.1.5 on Red Hat.

[Spencer Rathbun]

@AdamMatan Ah, use the -r flag for GNU sed, as Thor indicates.

[Evgenii Iablokov]

It's better to use semi colon in "twice" command. Like this: sed 's/\.ism/\.ism\.md5/g;s/\.usp/\.usp\.md5/g' /tmp/verify

[bobbogo]

@EvgeniyYablokov Any particular reason? Perhaps in short throwaway scripts (like this one) you may have a point, but I still think the two-line version is easier to read.

[dgo]

I'm using the GNU-utils version on Windows @ had to do as @EvgeniyYablokov suggested with the semi-colon for it to work.

[Walery Strauch]

On mac os I got sed: 1: "-E ": invalid command code - using sed -e -E. Switching parameter it worked fine: sed -E -e

[Ray Foss]

brew install gnu-sed && gsed --version... but don't replace it for the whole system as poorly programs that depend on MacOS having a "stable" shell environment will break... also your changes are usually destroyed with ever minor update.