Strip off word pattern at the end of string using sed

Question

I have following string:

ksh -x SCRIPT2.ksh $BUSSINESSDATE & PID=$!; ksh -x SCRIPT1.ksh $BUSSINESSDATE & PID=$!;

I want to strip off the last word & PID=$!;. Desired result is:

ksh -x SCRIPT2.ksh $BUSSINESSDATE & PID=$!; ksh -x SCRIPT1.ksh $BUSSINESSDATE

Note: that I do have & PID=$!; in the middle of the string, I only need to get rid of the last word which contains & PID=$!;

I am using sed achieve this. But it doesn't seems to produce the expected output.

sed 's/\(\& PID=\$\!\).*$//'

Solution 1

sed -e 's/& PID=\$\!;$//'

The $ toward the end anchors it to the end of the string.

Solution 2

All answers are good so far, but for this concrete example you can save your time and write a few characters less:

sed 's/ & [^ ]*$//'

or even better

sed 's/ & \S*$//'

Solution 3

To remove the last occurrence of a pattern on a line, you can just squeeze it.

sed 's/\(.*\)& PID=$!;/\1/'

That will match as much as possible in \1 before matching the string so it will always strip only the last occurrence - regardless of whether or not the string you strip is actually at the end of the line.

Your attempt:

 sed 's/\(\& PID=\$\!\).*$//'

...does something like the opposite. It strips all characters on a line beginning with the first occurrence of your string and following on to the end of the line.

Remember the leftmost-longest rule - a regexp will always begin its match as far to the left in the string as it might and continue to match for as long as it might.

---

Note that if you want to remove just n last characters then with gnu sed

sed -E 's/.{9}$//'

Here -E option is necessary to use extended regular expressions.

Comments

[Costas]

Try sed 's/& PID=$!;\s*$//'

[Sas]

@Costas It works. But wondering what does "\s" in the middle does in your command?

[janos]

\s* is zero or more white-space characters. You might not need it. It seems Costas was being extra careful.

[jimmij]

@mikeserv .{9}$ is only 5, so definitely the best one! :)

[mikeserv]

@jimmij - you should probably give up the -r habit. The -E syntax is compatible with GNU and BSD seds and should be POSIX as well if/when this ever takes effect.

[jimmij]

Yes, probably. But -E is not documented neither in man nor in info, at least for my GNU sed version 4.2.1.

[mikeserv]

@jimmij - yup, that remains an undocumented feature*(?)*. And it actually is slightly less capable - as near as I can tell the only thing you can do with -r that you cannot do w/ -E is switch from a BRE context to an ERE context. With -r you can do sed -e 's/BREhere\(.*\)/w \1 portable backrefs/' -er 's/(nowERE).*/and \1 backrefs only as a non-POSIX extension/'. While the backrefs w/ ERE still work w/ -E you can't switch from BRE to ERE like that. And even w/ -r you can't switch back to a BRE context anyway.