grep: unescaped ^ or $ not supported with -Pz - regex

I wonder why in the new version of grep (Ubuntu 16.04) my bash script stopped working:
...
COMMIT_REGEX='^\[[A-Z]+-[0-9]+\] \s*\S+(?:.|\n|\r)*\s* \(review: ([a-z]+\.[a-z]+|MYSELF)\)$'
if ! grep -Paz "$COMMIT_REGEX" "$1"; then
...
I get "grep: unescaped ^ or $ not supported with -Pz". I've tried to escape ^ and $ symbols, but it doesn't help.
In Ubuntu 15.10 script works perfectly.

It seems that the problem is the result of a bug with grep -Pz (credit to Lars Fischer for finding the relevant report).
I would suggest dropping the -P switch and using -E instead:
commit_re='^\[[A-Z]+-[0-9]+\] \s*\S+(.|\n|\r)*\s* \(review: ([a-z]+\.[a-z]+|MYSELF)\)$'
if ! grep -qEaz "$commit_re" "$1"; then
The only changes that I've made are to change -P to -E and add the -q (quiet) switch, since you're only interested in the return code. You don't really need a non-capturing group, so I changed it to a normal one.
I also don't like to see ALL_CAPS variable names as they should really be reserved for use by the shell.

Related

Case insensitive sed for pattern [duplicate]

I'm trying to use SED to extract text from a log file. I can do a search-and-replace without too much trouble:
sed 's/foo/bar/' mylog.txt
However, I want to make the search case-insensitive. From what I've googled, it looks like appending i to the end of the command should work:
sed 's/foo/bar/i' mylog.txt
However, this gives me an error message:
sed: 1: "s/foo/bar/i": bad flag in substitute command: 'i'
What's going wrong here, and how do I fix it?
Update: Starting with macOS Big Sur (11.0), sed now does support the I flag for case-insensitive matching, so the command in the question should now work (BSD sed doesn't reporting its version, but you can go by the date at the bottom of the man page, which should be March 27, 2017 or more recent); a simple example:
# BSD sed on macOS Big Sur and above (and GNU sed, the default on Linux)
$ sed 's/ö/#/I' <<<'FÖO'
F#O # `I` matched the uppercase Ö correctly against its lowercase counterpart
Note: I (uppercase) is the documented form of the flag, but i works as well.
Similarly, starting with macOS Big Sur (11.0) awk now is locale-aware (awk --version should report 20200816 or more recent):
# BSD awk on macOS Big Sur and above (and GNU awk, the default on Linux)
$ awk 'tolower($0)' <<<'FÖO'
föo # non-ASCII character Ö was properly lowercased
The following applies to macOS up to Catalina (10.15):
To be clear: On macOS, sed - which is the BSD implementation - does NOT support case-insensitive matching - hard to believe, but true. The formerly accepted answer, which itself shows a GNU sed command, gained that status because of the perl-based solution mentioned in the comments.
To make that Perl solution work with foreign characters as well, via UTF-8, use something like:
perl -C -Mutf8 -pe 's/öœ/oo/i' <<< "FÖŒ" # -> "Foo"
-C turns on UTF-8 support for streams and files, assuming the current locale is UTF-8-based.
-Mutf8 tells Perl to interpret the source code as UTF-8 (in this case, the string passed to -pe) - this is the shorter equivalent of the more verbose -e 'use utf8;'.Thanks, Mark Reed
(Note that using awk is not an option either, as awk on macOS (i.e., BWK awk and BSD awk) appears to be completely unaware of locales altogether - its tolower() and toupper() functions ignore foreign characters (and sub() / gsub() don't have case-insensitivity flags to begin with).)
A note on the relationship of sed and awk to the POSIX standard:
BSD sed and awk limit their functionality mostly to what the POSIX sed and
POSIX awk specs mandate, whereas their GNU counterparts implement many more extensions.
Editor's note: This solution doesn't work on macOS (out of the box), because it only applies to GNU sed, whereas macOS comes with BSD sed.
Capitalize the 'I'.
sed 's/foo/bar/I' file
Another work-around for sed on Mac OS X is to install gsedfrom MacPorts or HomeBrew and then create the alias sed='gsed'.
If you are doing pattern matching first, e.g.,
/pattern/s/xx/yy/g
then you want to put the I after the pattern:
/pattern/Is/xx/yy/g
Example:
echo Fred | sed '/fred/Is//willma/g'
returns willma; without the I, it returns the string untouched (Fred).
The sed FAQ addresses the closely related case-insensitive search. It points out that a) many versions of sed support a flag for it and b) it's awkward to do in sed, you should rather use awk or Perl.
But to do it in POSIX sed, they suggest three options (adapted for substitution here):
Convert to uppercase and store original line in hold space; this won't work for substitutions, though, as the original content will be restored before printing, so it's only good for insert or adding lines based on a case-insensitive match.
Maybe the possibilities are limited to FOO, Foo and foo. These can be covered by
s/FOO/bar/;s/[Ff]oo/bar/
To search for all possible matches, one can use bracket expressions for each character:
s/[Ff][Oo][Oo]/bar/
The Mac version of sed seems a bit limited. One way to work around this is to use a linux container (via Docker) which has a useable version of sed:
cat your_file.txt | docker run -i busybox /bin/sed -r 's/[0-9]{4}/****/Ig'
Use following to replace all occurrences:
sed 's/foo/bar/gI' mylog.txt
I had a similar need, and came up with this:
this command to simply find all the files:
grep -i -l -r foo ./*
this one to exclude this_shell.sh (in case you put the command in a script called this_shell.sh), tee the output to the console to see what happened, and then use sed on each file name found to replace the text foo with bar:
grep -i -l -r --exclude "this_shell.sh" foo ./* | tee /dev/fd/2 | while read -r x; do sed -b -i 's/foo/bar/gi' "$x"; done
I chose this method, as I didn't like having all the timestamps changed for files not modified. feeding the grep result allows only the files with target text to be looked at (thus likely may improve performance / speed as well)
be sure to backup your files & test before using. May not work in some environments for files with embedded spaces. (?)
Following should be fine:
sed -i 's/foo/bar/gi' mylog.txt

grep command with a lookahead pattern does not select anything

I was trying to use the following grep command:
grep '(.*)(?=(png|html|jpg|js|css)(?:\s*))(png|html|jpg|js|css.*\s)' file
File contains the following:
http://manage.bostonglobe.com/GiftTheGlobe/LandingPage.html
https://manage.bostonglobe.com/cs/mc/login.aspx?p1=BGFooter
https://www.bostonglobe.com/bgcs
/newsletters?p1=BGFooter_Newsletters
https://bostonglobe.custhelp.com/app/home?p1=BGFooter
https://bostonglobe.custhelp.com/app/answers/list?p1=BGFooter
/tools/help/stafflist?p1=BGFooter
https://www.bostonglobemedia.com/
https://manage.bostonglobe.com/Order/newspaper/Newspaper.aspx
https://www.facebook.com/globe
https://twitter.com/#!/BostonGlobe
https://plus.google.com/108227564341535363126/about
https://epaper.bostonglobe.com/launch.aspx?pbid=2c60291d-c20c-4780-9829- b3d9a12687cf
http://nieonline.com/bostonglobe/
https://secure.pqarchiver.com/boston-sub/no_default.html?ss=1&url=%2Fboston-sub%2Fadvancedsearch.html
/tools/help/privacy?p1=BGFooter
/tools/help/terms-service?p1=BGFooter
/termsofpurchase?p1=BGFooter
https://www.bostonglobemedia.com/careers
/css/globe-print.css?v=19256I1935
//meter.bostonglobe.com/css/style.css
/css/globe-print.css?v=19256I1935
//cdn.blueconic.net/bostonglobemedia.js
/js/lib/rwd-images.js,lib/respond.min.js,lib/modernizr.custom.min.js,globe- define.js,globe-controller.js?v=19256I1935
data:image/gif;base64,R0lGODlhAQABAIAAAAAAAP///ywAAAAAAQABAAACAUwAOw==
/js/lib/jquery.js,lib/lo-dash-custom-2.4.1.js,lib/a9.js,lib/pb.js,dist/ad- init.js,globe-newsletter.js,globe-profile-page.js,dist/globe-topic-nav.js,dist/rakuten.js?v=19256I1935
//dc8xl0ndzn2cb.cloudfront.net/js/bostonglobe/v0/keywee.min.js
For some reason it doesn't select anything from that file, Ive tried different flags but cant seem to figure out whats wrong
You are using a PCRE regex with the POSIX BRE engine that is default grep engine.
To make those patterns work, you should use -P option (available in GNU grep):
grep -P 'YOUR_PCRE_PATTERN'
^^
To develop and test PCRE patterns, a well-known regex101.com is usually recommended.
Note that on Mac OS, you may install GNU grep via brew.

Are negative lookbehind in regex searches possible in Geany?

Geany's documentation on negative assertions makes it look like they're possible.
For reference, this works and gives me results:
pcregrep -r "(?<!= )function\(" src/main-js/
But the same regex, or any regex with a negative lookbehind, gives me no result when launched from Geany (v 1.24.1)
Where is the problem? Is the documentation wrong?
Precision: the topic isn't about how to avoid doing a negative look behind, but about how to do any standard PCRE negative look behind.
I got support from the Geany developers on Freenode, and it was very helpful. Here is what they told me:
The documented RE syntax only applies to the RE engine directly used by
Geany (e.g. in Find), but the Find in Files features calls the grep tool
(as configured in preferences->tools->grep), which has its own syntax.
For GNU grep, you can add "-P" to the "Extra options" field in the
dialog
However, after you tried it, you had this error:
/bin/grep: conflicting matchers specified
... to which I was told this was a Geany bug. Geany calls grep -E, and -P is not compatible with it.
Your only workaround is to have a shell script calling grep with -P instead of -E, and use this script. You should be able to configure the grep tool to call in Geany preferences.
An example of said shell script:
#!/bin/sh
matchopts=$(echo "$1" | tr E P)
shift
exec grep $matchopts "$#"
Geany uses either -F or -E (these are the only available engines in POSIX grep) for grep, hence why you can't pass -P.
I've reported the bug to the Geany developers.
Another workaround is to avoid the negative lookbehind assertion… but it's a lot uglier:
(^.?|[^=] |=[^ ]|[^=][^ ])function

grep with regexp: whitespace doesn't match unless I add an assertion

GNU grep 2.5.4 on bash 4.1.5(1) on Ubuntu 10.04
This matches
$ echo "this is a line" | grep 'a[[:space:]]\+line'
this is a line
But this doesn't
$ echo "this is a line" | grep 'a\s\+line'
But this matches too
$ echo "this is a line" | grep 'a\s\+\bline'
this is a line
I don't understand why #2 does not match (whereas # 1 does) and #3 also shows a match. Whats the difference here?
Take a look at your grep manpage. Perl added a lot of regular expression extensions that weren't in the original specification. However, because they proved so useful, many programs adopted them.
Unfortunately, grep is sometimes stuck in the past because you want to make sure your grep command remains compatible with older versions of grep.
Some systems have egrep with some extensions. Others allow you to use grep -E to get them. Still others have a grep -P that allows you to use Perl extensions. I believe Linux systems' grep command can use the -P extension which is not available in most Unix systems unless someone has replaced the grep with the GNU version. Newer versions of Mac OS X also support the -P switch, but not older versions.
grep doesn't support the complete set of regular expressions, so try using -P to enable perl regular expressions. You don't need to escape the + i.e.
echo "this is a line" | grep -P 'a\s+line'

Why does this `grep -o` fail, and how should I work around it?

Given the input
echo abc123def | grep -o '[0-9]*'
On one computer (with GNU grep 2.5.4), this returns 123, and on another (with GNU grep 2.5.1) it returns the empty string. Is there some explanation for why grep 2.5.1 fails here, or is it just a bug? I'm using grep -o in this way in a bash script that I'd like to be able to run on different computers (which may have different versions of grep). Is there a "right way" to get consistent behavior?
Yes, 2.5.1's -o handling was buggy:
http://www.mail-archive.com/bug-grep#gnu.org/msg00993.html
Grep is probably not the right tool for this; sed or tr or even perl might be better depending on what the actual task is.
you can use the shell. its faster
$ str=abc123def
$ echo ${str//[a-z]/}
123
I had the same issue and found that egrep was installed on that machine. A quick solution was using
echo abc123def | egrep -o '[0-9]*'
This will give similar results:
echo abc123def | sed -n 's/[^0-9]*\([0-9]\+\).*/\1/p'
Your question is a near-duplicate of this one.
Because you are using a regex so you must use either:
grep -E
egrep (like Sebastian posted).
Good luck!