Substitute text between pattern with string using sed

Johng · 07-11-2020, 10:22 PM

I am trying to create a script to edit files using sed. I want to substitute some text between two patterns with a string generated in the script. The text to be replaced varies from file to file.

Change:

Code:

<name>some unknown text</name>

To:

Code:

<name>$newstring</name>

Any help?

NevemTeve · 07-11-2020, 10:51 PM

Untested:

Code:

FromName='some unknown text'
ToName='$newstring'
sed -i 's/<name>'"$FromName"'<\/name>/<name>'"$ToName"'<\/name>/' filename.xml

Or perhaps you mean this:

Code:

ToName='$newstring'
sed -i 's/<name>[^<]*<\/name>/<name>'"$ToName"'<\/name>/' filename.xml

syg00 · 07-11-2020, 10:51 PM

So what have you tried, and why didn't it work ?.

Johng · 07-12-2020, 12:59 AM

Thank you NevemTeve
The second version was just what I was looking for:

Quote:

sed -i 's/<name>[^<]*<\/name>/<name>'"$ToName$GPXname"'<\/name>/' xxxx.gpx

MadeInGermany · 07-12-2020, 02:20 AM

The / delimiter conflicts here. You can chose another delimiter:

Code:

sed -i 's#<name>[^<]*</name>#<name>'"$ToName$GPXname"'</name>#' xxxx.gpx

Note that the shell variables may not contain the delimiter, or sed will get a syntax error.

Further, PCRE lets you turn the greedy * match into a non-greedy *? match.
PCRE is not (yet) supported in sed, so here is the original - perl:

Code:

perl -pe -i 's#<name>.*?</name>#<name>'"$ToName$GPXname"'</name>#' xxxx.gpx

By using perl's own variables you can even avoid any delimiter conflict...

syg00 · 07-12-2020, 02:47 AM

Negated char classes are non-greedy by definition. No need for perl or perlre.

MadeInGermany · 07-12-2020, 03:34 AM

Quote:

Originally Posted by syg00

Negated char classes are non-greedy by definition. No need for perl or perlre.

Not true: by definition the [^<]* expands to the maximum. The trick is to except the < character so it cannot go beyond the < character.

shruggy · 07-12-2020, 07:55 AM

Hmm, and what <name> tag exactly is this supposed to be? Most probably /gpx/trk/name, but you see, a GPX file also may have /gpx/name and/or /gpx/wpt/name.

Anyway, this will change the value of /gpx/trk/name:

Code:

#!/bin/sh
file="xxxx.gpx"
grep -q 'xmlns=' "$file" && ns=_: || ns=
xmlstarlet ed -u //${ns}trk/${ns}name -v "$ToName$GPXname" "$file"

//${ns}name will instead edit any <name> tag present in the GPX file.

boughtonp · 07-13-2020, 06:12 AM

Quote:

Originally Posted by MadeInGermany

Quote:

Originally Posted by syg00

Negated char classes are non-greedy by definition. No need for perl or perlre.

Not true: by definition the [^<]* expands to the maximum. The trick is to except the < character so it cannot go beyond the < character.

Saying "not true" is missing the point - character classes have no notion of greedy or non-greedy, they are a definition of which characters should be included/excluded.

It is the quantifier which may or not be greedy, and the quantifier doesn't care whether it is acting on a regular character class or negative character class.

The point syg00 was probably trying to make (albeit with poorly chosen words) is that, when viable, using a boundary character inside a negative character class with a greedy quantifier is almost always a better choice than using an overly permissive class with a lazy quantifier - i.e. using "[^<]*<" is better than using ".*?<" so it makes no sense to suggest the latter as an "improvement" to a working pattern (especially at the expense of clouding the important bit of the post highlighting the potential benefit of using Perl variables over Bash ones).

Of course, the ideal solution to changing markup is generally to use a markup-aware parser, such as Shruggy posted.