sed - Search and replace always puts capture group at the end - sed

I'm refactoring some code across lots of files, and am trying to do a search and replace using sed on a Mac.
The goal is to go from this:
fooActions: typeof fooActionCreators
to this:
fooActions: Partial<typeof fooActionCreators>
I've gotten it to mostly work using this sed expression:
ag -l "Actions: typeof " | xargs sed -i "" -e "s/Actions: typeof \(\w*\)/Actions: Partial<typeof \1>/g"
Basically using ag to find the list of file names containing a matching string, then use xargs to pass those as input to my sed expression.
The output I get though looks like this:
fooActions: Partial<typeof >fooActionCreators
I can't figure out why my capture group is being placed at then end instead of where I have it in the replace clause.

The sed that comes with mac doesn't support \w for word characters.
If you have gsed (gnu sed) this should work as expected.
$ echo 'Actions: typeof foo' | gsed -e "s/Actions: typeof \(\w*\)/Actions: Partial<typeof \1>/g"
Actions: Partial<typeof foo>
Otherwise you can use something like [a-zA-Z]* instead of \w*.
$ echo 'Actions: typeof foo' | sed -e "s/Actions: typeof \([a-zA-Z]*\)/Actions: Partial<typeof \1>/g"
Actions: Partial<typeof foo>
The seemingly odd behavior that you're seeing is because in the plain sed version, the \w* is matching the empty string in front of the thing you're hoping to match, as illustrated below:
Plain sed matches empty string in front of a
$ echo a | sed "s/\(\w*\)/\1x/"
xa
Gnu sed matches the a as intended
$ echo a | gsed "s/\(\w*\)/\1x/"
ax

if your data in 'd' file, by gnu sed,
sed -E 's/^(fooAction)s: (typeof \1Creators)/\1s: Partial<\2>/' d

Related

Using a single sed call to split and grep

This is mostly by curiosity, I am trying to have the same behavior as:
echo -e "test1:test2:test3"| sed 's/:/\n/g' | grep 1
in a single sed command.
I already tried
echo -e "test1:test2:test3"| sed -e "s/:/\n/g" -n "/1/p"
But I get the following error:
sed: can't read /1/p: No such file or directory
Any idea on how to fix this and combine different types of commands into a single sed call?
Of course this is overly simplified compared to the real usecase, and I know I can get around by using multiple calls, again this is just out of curiosity.
EDIT: I am mostly interested in the sed tool, I already know how to do it using other tools, or even combinations of those.
EDIT2: Here is a more realistic script, closer to what I am trying to achieve:
arch=linux64
base=https://chromedriver.storage.googleapis.com
split="<Contents>"
curl $base \
| sed -e 's/<Contents>/<Contents>\n/g' \
| grep $arch \
| sed -e 's/^<Key>\(.*\)\/chromedriver.*/\1/' \
| sort -V > out
What I would like to simplify is the curl line, turning it into something like:
curl $base \
| sed 's/<Contents>/<Contents>\n/g' -n '/1/p' -e 's/^<Key>\(.*\)\/chromedriver.*/\1/' \
| sort -V > out
Here are some alternatives, awk and sed based:
sed -E "s/(.*:)?([^:]*1[^:]*).*/\2/" <<< "test1:test2:test3"
awk -v RS=":" '/1/' <<< "test1:test2:test3"
# or also
awk 'BEGIN{RS=":"} /1/' <<< "test1:test2:test3"
Or, using your logic, you would need to pipe a second sed command:
sed "s/:/\n/g" <<< "test1:test2:test3" | sed -n "/1/p"
See this online demo. The awk solution looks cleanest.
Details
In sed solution, (.*:)?([^:]*1[^:]*).* pattern matches an optional sequence of any 0+ chars and a :, then captures into Group 2 any 0 or more chars other than :, 1, again 0 or more chars other than :, and then just matches the rest of the line. The replacement just keeps Group 2 contents.
In awk solution, the record separator is set to : and then /1/ regex is used to only return the record having 1 in it.
This might work for you (GNU sed):
sed 's/:/\n/;/^[^\n]*1/P;D' file
Replace each : and if the first line in the pattern space contains 1 print it.
Repeat.
An alternative:
sed -Ez 's/:/\n/g;s/^[^1]*$//mg;s/\n+/\n/;s/^\n//' file
This slurps the whole file into memory and replaces all colons by newlines. All lines that do not contain 1 are removed and surplus newlines deleted.
An alternative to the really ugly sed is: grep -o '\w*2\w*'
$ printf "test1:test2:test3\nbob3:bob2:fred2\n" | grep -o '\w*2\w*'
test2
bob2
fred2
grep -o: only matching
Or: grep -o '[^:]*2[^:]*'
echo -e "test1:test2:test3" | sed -En 's/:/\n/g;/^[^\n]*2[^\n]*(\n|$)/P;//!D'
sed -n doesn't print unless told to
sed -E allows using parens to match (\n|$) which is newline or the end of the pattern space
P prints the pattern buffer up to the first newline.
D trims the pattern buffer up to the first newline
[^\n] is a character class that matches anything except a newline
// is sed shorthand for repeating a match
//! is then matching everything that didn't match previously
So, after you split into newlines, you want to make sure the 2 character is between the start of the pattern buffer ^ and the first newline.
And, if there is not the character you are looking for, you want to D delete up to the first newline.
At that point, it works for one line of input, with one string containing the character you're looking for.
To expand to several matches within a line, you have to ta, conditionally branch back to label :a:
$ printf "test1:test2:test3\nbob3:bob2:fred2\n" | \
sed -En ':a s/:/\n/g;/^[^\n]*2[^\n]*(\n|$)/P;D;ta'
test2
bob2
fred2
This is simply NOT a job for sed. With GNU awk for multi-char RS:
$ echo "test1:test2:test3:test4:test5:test6"| awk -v RS='[:\n]' '/1/'
test1
$ echo "test1:test2:test3:test4:test5:test6"| awk -v RS='[:\n]' 'NR%2'
test1
test3
test5
$ echo "test1:test2:test3:test4:test5:test6"| awk -v RS='[:\n]' '!(NR%2)'
test2
test4
test6
$ echo "foo1:bar1:foo2:bar2:foo3:bar3" | awk -v RS='[:\n]' '/foo/ || /2/'
foo1
foo2
bar2
foo3
With any awk you'd just have to strip the \n from the final record before operating on it:
$ echo "test1:test2:test3:test4:test5:test6"| awk -v RS=':' '{sub(/\n$/,"")} /1/'
test1

How to replace only last match in a line with sed?

With sed, I can replace the first match in a line using
sed 's/pattern/replacement/'
And all matches using
sed 's/pattern/replacement/g'
How do I replace only the last match, regardless of how many matches there are before it?
Copy pasting from something I've posted elsewhere:
$ # replacing last occurrence
$ # can also use sed -E 's/:([^:]*)$/-\1/'
$ echo 'foo:123:bar:baz' | sed -E 's/(.*):/\1-/'
foo:123:bar-baz
$ echo '456:foo:123:bar:789:baz' | sed -E 's/(.*):/\1-/'
456:foo:123:bar:789-baz
$ echo 'foo and bar and baz land good' | sed -E 's/(.*)and/\1XYZ/'
foo and bar and baz lXYZ good
$ # use word boundaries as necessary - GNU sed
$ echo 'foo and bar and baz land good' | sed -E 's/(.*)\band\b/\1XYZ/'
foo and bar XYZ baz land good
$ # replacing last but one
$ echo 'foo:123:bar:baz' | sed -E 's/(.*):(.*:)/\1-\2/'
foo:123-bar:baz
$ echo '456:foo:123:bar:789:baz' | sed -E 's/(.*):(.*:)/\1-\2/'
456:foo:123:bar-789:baz
$ # replacing last but two
$ echo '456:foo:123:bar:789:baz' | sed -E 's/(.*):((.*:){2})/\1-\2/'
456:foo:123-bar:789:baz
$ # replacing last but three
$ echo '456:foo:123:bar:789:baz' | sed -E 's/(.*):((.*:){3})/\1-\2/'
456:foo-123:bar:789:baz
Further Reading:
Buggy behavior if word boundaries is used inside a group with quanitifiers - for example: echo 'it line with it here sit too' | sed -E 's/with(.*\bit\b){2}/XYZ/' fails
Greedy vs. Reluctant vs. Possessive Quantifiers
Reference - What does this regex mean?
sed manual: Back-references and Subexpressions
This might work for you (GNU sed):
sed 's/\(.*\)pattern/\1replacement/' file
Use greed to swallow up the pattern space and then regexp engine will step back through the line and find the first match i.e. the last match.
A fun way to do this, is to use rev to reverse the characters of each line and write your sed replacement backwards.
rev input_file | sed 's/nrettap/tnemecalper/' | rev

sed expressions: line shows twice

I'm parsing a csv file with a sed command like this:
sed -n -e 's/abc/&/p' -e 's/xyz/&/p' <input >output
Now if there is both in one line (abc and xyz) I'll have the line twice in the output. I'd should have it just once.
Can I do that with sed?
If you only want to print a line with "abc" or "xyz":
sed -n '/abc\|xyz/p'
Other tools:
grep -F -e abc -e xyz
awk '/abc/ || /xyz/'
I believe you are mis-using the s///p just to print the lines. This is not necessary in sed - you can get them printed using //p.
Both expressions will evaluate, though, so you are still at risk of duplication. Your best bet (and fastest, for large datasets) will be to build the 'or' behavior into the matching regexp:
sed -Ene '/abc|xyz/p' input >output
sed -n -r -e 's/(abc|xyz)/&/p' <input >output
-r flag is for enabling extended regular expressions (is in GNU sed)
sed -n '/abc/{p;b
}
/xyz/p' Input > Output
for non GNU sed (where | is not allowed as OR)

Replace string with substring in lowercase using sed / awk / tr / perl?

I have a plaintext file containing multiple instances of the pattern $$DATABASE_*$$ and the asterisk could be any string of characters. I'd like to replace the entire instance with whatever is in the asterisk portion, but lowercase.
Here is a test file:
$$DATABASE_GIBSON$$
test me $$DATABASE_GIBSON$$ test me
$$DATABASE_GIBSON$$ test $$DATABASE_GIBSON$$ test
$$DATABASE_GIBSON$$ $$DATABASE_GIBSON$$$$DATABASE_GIBSON$$
Here is the desired output:
gibson
test me gibson test me
gibson test gibson test
gibson gibsongibson
How do I do this with sed/awk/tr/perl?
Here's the perl version I ended up using.
perl -p -i.bak -e 's/\$\$DATABASE_(.*?)\$\$/lc($1)/eg' inputFile
Unfortunately there's no easy, foolproof way with awk, but here's one approach:
$ cat tst.awk
{
gsub(/[$][$]/,"\n")
head = ""
tail = $0
while ( match(tail, "\nDATABASE_[^\n]+\n") ) {
head = head substr(tail,1,RSTART-1)
trgt = substr(tail,RSTART,RLENGTH)
tail = substr(tail,RSTART+RLENGTH)
gsub(/\n(DATABASE_)?/,"",trgt)
head = head tolower(trgt)
}
$0 = head tail
gsub("\n","$$")
print
}
$ cat file
The quick brown $$DATABASE_FOX$$ jumped over the lazy $$DATABASE_DOG$$s back.
The grey $$DATABASE_SQUIRREL$$ ate $$DATABASE_NUT$$s under a $$DATABASE_TREE$$.
Put a dollar $$DATABASE_DOL$LAR$$ in the $$ string.
$ awk -f tst.awk file
The quick brown fox jumped over the lazy dogs back.
The grey squirrel ate nuts under a tree.
Put a dollar dol$lar in the $$ string.
Note the trick of converting $$ to a newline char so we can negate that char in the match(RE), without that (i.e. if we used ".+" instead of "[^\n]+") then due to greedy RE matching if the same pattern appeared twice on one input line the matching string would extend from the start of the first pattern to the end of the second pattern.
This one works with complicated examples.
perl -ple 's/\$\$DATABASE_(.*?)\$\$/lc($1)/eg' filename.txt
And for simpler examples :
echo '$$DATABASE_GIBSON$$' | sed 's#$$DATABASE_\(.*\)\$\$#\L\1#'
in sed, \L means lower case (\E to stop if needed)
Using awk alone:
> echo '$$DATABASE_AWESOME$$' | awk '{sub(/.*_/,"");sub(/\$\$$/,"");print tolower($0);}'
awesome
Note that I'm in FreeBSD, so this is not GNU awk.
But this can be done using bash alone:
[ghoti#pc ~]$ foo='$$DATABASE_AWESOME$$'
[ghoti#pc ~]$ foo=${foo##*_}
[ghoti#pc ~]$ foo=${foo%\$\$}
[ghoti#pc ~]$ foo=${foo,,}
[ghoti#pc ~]$ echo $foo
awesome
Of the above substitutions, all except the last one (${foo,,}) will work in standard Bourne shell. If you don't have bash, you can instead do use tr for this step:
$ echo $foo
AWESOME
$ foo=$(echo "$foo" | tr '[:upper:]' '[:lower:]')
$ echo $foo
awesome
$
UPDATE:
Per comments, it seems that what the OP really wants is to strip the substring out of any text in which it is included -- that is, our solutions need to account for the possibility of leading or trailing spaces, before or after the string he provided in his question.
> echo 'foo $$DATABASE_KITTENS$$ bar' | sed -nE '/\$\$[^$]+\$\$/{;s/.*\$\$DATABASE_//;s/\$\$.*//;p;}' | tr '[:upper:]' '[:lower:]'
kittens
And if you happen to have pcregrep on your path (from the devel/pcre FreeBSD port), you can use that instead, with lookaheads:
> echo 'foo $$DATABASE_KITTENS$$ bar' | pcregrep -o '(?!\$\$DATABASE_)[A-Z]+(?=\$\$)' | tr '[:upper:]' '[:lower:]'
kittens
(For Linux users reading this: this is equivalent to using grep -P.)
And in pure bash:
$ shopt -s extglob
$ foo='foo $$DATABASE_KITTENS$$ bar'
$ foo=${foo##*(?)\$\$DATABASE_}
$ foo=${foo%%\$\$*(?)}
$ foo=${foo,,}
$ echo $foo
kittens
Note that NONE of these three updated solutions will handle situations where multiple tagged database names exist in the same line of input. That's not stated as a requirement in the question either, but I'm just sayin'....
You can do this in a pretty foolproof way with the supercool command cut :)
echo '$$DATABASE_AWESOME$$' | cut -d'$' -f3 | cut -d_ -f2 | tr 'A-Z' 'a-z'
This might work for you (GNU sed):
sed 's/$\$/\n/g;s/\nDATABASE_\([^\n]*\)\n/\L\1/g;s/\n/$$/g' file
Here is the shortest (GNU) awk solution I could come up with that does everything requested by the OP:
awk -vRS='[$][$]DATABASE_([^$]+[$])+[$]' '{ORS=tolower(substr(RT,12,length(RT)-13))}1'
Even if the string indicated with the asterix (*) contained one or more single Dollar signs ($) and/or linebreaks this soultion should still work.
awk '{gsub(/\$\$DATABASE_GIBSON\$\$/,"gibson")}1' file
gibson
test me gibson test me
gibson test gibson test
gibson gibsongibson
echo $$DATABASE_WOOLY$$ | awk '{print tolower($0)}'
awk will take what ever input, in this case the first agurment, and use the tolower function and return the results.
For your bash script you can do something like this and use the variable DBLOWER
DBLOWER=$(echo $$DATABASE_WOOLY$$ | awk '{print tolower($0)}');

How do I push `sed` matches to the shell call in the replacement pattern?

I need to replace several URLs in a text file with some content dependent on the URL itself. Let's say for simplicity it's the first line of the document at the URL.
What I'm trying is this:
sed "s/^URL=\(.*\)/TITLE=$(curl -s \1 | head -n 1)/" file.txt
This doesn't work, since \1 is not set. However, the shell is getting called. Can I somehow push the sed match variables to that subprocess?
The accept answer is just plain wrong. Proof:
Make an executable script foo.sh:
#! /bin/bash
echo $* 1>&2
Now run it:
$ echo foo | sed -e "s/\\(foo\\)/$(./foo.sh \\1)/"
\1
$
The $(...) is expanded before sed is run.
So you are trying to call an external command from inside the replacement pattern of a sed substitution. I dont' think it can be done, the $... inside a pattern just allows you to use an already existent (constant) shell variable.
I'd go with Perl, see the /e option in the search-replace operator (s/.../.../e).
UPDATE: I was wrong, sed plays nicely with the shell, and it allows you do to that. But, then, the backlash in \1 should be escaped. Try instead:
sed "s/^URL=\(.*\)/TITLE=$(curl -s \\1 | head -n 1)/" file.txt
Try this:
sed "s/^URL=\(.*\)/\1/" file.txt | while read url; do sed "s#URL=\($url\)#TITLE=$(curl -s $url | head -n 1)#" file.txt; done
If there are duplicate URLs in the original file, then there will be n^2 of them in the output. The # as a delimiter depends on the URLs not including that character.
Late reply, but making sure people don't get thrown off by the answers here -- this can be done in gnu sed using the e command. The following, for example, decrements a number at the beginning of a line:
echo "444 foo" | sed "s/\([0-9]*\)\(.*\)/expr \1 - 1 | tr -d '\n'; echo \"\2\";/e"
will produce:
443 foo