40

For example:

sed 's/\u0091//g' file1

Right now, I have to do hexdump to get hex number and put into sed as follows:

$ echo -ne '\u9991' | hexdump -C
00000000  e9 a6 91                                          |...|
00000003

And then:

$ sed 's/\xe9\xa6\x91//g' file1
34

Just use that syntax:

sed 's/馑//g' file1

Or in the escaped form:

sed "s/$(echo -ne '\u9991')//g" file1

(Note that older versions of Bash and some shells do not understand echo -e '\u9991', so check first.)

8
  • 1
    Does sed count 馑 as one character or 3? That is, does echo 馑 | sed s/...// print anything?
    – user253751
    Apr 17 '15 at 11:22
  • @immibis Since sed has the g modifier it replaces all occurence also when they follow each other. Also sed should count it as one character, see: echo -ne "馑" | wc -m gives 1. If you count the bytes (wc -c) it would return 3. Did I understand your question correctly?
    – chaos
    Apr 17 '15 at 11:28
  • I meant: does . mean "one character" or "one byte"?
    – user253751
    Apr 17 '15 at 11:30
  • @immibis I matches one character hence echo 馑 | sed s/...// gives me (nothing is replaced)
    – chaos
    Apr 17 '15 at 11:33
  • 4
    @chaos: It works under en_US.UTF-8, but doesn't under C.
    – choroba
    Apr 17 '15 at 12:28
18

Perl can do that:

echo 汉典“馑”字的基本解释 | perl -CS -pe 's/\N{U+9991}/Jin/g'

-CS turns on UTF-8 for standard input, output and error.

2
  • 9
    Perl can do almost anything..... Apr 17 '15 at 10:49
  • @wobbily_col Maybe. But it's written in C. So..if Perl can do almost anything C as the foundation of Perl can do anything. As it should be.
    – Pryftan
    Jan 16 '20 at 15:51
7

A number of versions of sed support Unicode:

  • Heirloom sed, which is based on "original Unix material".
  • GNU sed, which is its own codebase.
  • Plan 9 sed, which has been ported to Unix-like operating systems.

I couldn't find information on BSD sed, which I thought was strange, but I think the odds are good that it supports Unicode too. Unfortunately, there is no standard way to tell sed which encoding to use, so each one does this in its own ways.

3
  • Do they support UTF-16 with and without BOM ?
    – Bon Ami
    Apr 17 '15 at 17:12
  • 12
    UTF-16 is pretty unusable in Unix-based OSes. It's also an abomination that should have never seen the light of day.
    – Brian Bi
    Apr 17 '15 at 19:11
  • Whether or not they support UTF-16 depends on the implementation, and I'm afraid I don't have that data. I doubt that Plan 9 sed does (the original OS is UTF-8 everywhere), but I can't be sure, and even if it doesn't, the others might. Apr 17 '15 at 19:30
6

With recent versions of BASH, just omit the quotes around the sed expression and you can use BASH's escaped strings. Spaces within the sed expression or parts of the sed expression that might be interpreted by BASH as wildcards can be individually quoted.

$ echo "饥馑荐臻" | sed s/$'\u9991'//g
饥荐臻
2
  • 1
    This should be the new accepted answer, simple and clean!
    – Allen Wang
    Nov 6 '19 at 22:28
  • 1
    @AllenWang For reference, the $'...' type of quotes comes from ksh93 in 1993 while the \uxxxx within them comes from zsh in 2003 (inspired from GNU printf). Added in bash in 4.2 in 2010. So unless you're on macos which still comes with 3.2, that answer would have also been valid in 2015 when that question was asked. Jun 22 at 5:54
4

This works for me:

$ vim -nEs +'%s/\%u9991//g' +wq file1

It’s a drop more verbose than I’d like; here’s a full explanation:

  • -n disable vim swap file
  • -E Ex improved mode
  • -s silent mode
  • +'%s/\%u9991//g' execute the substitution command
  • +wq save and exit
2
  • I suppose this modifies file1 in-place, is that correct?
    – gerrit
    Jan 10 '19 at 10:32
  • @gerrit that’s correct, and thanks for pointing it out. Jan 10 '19 at 19:21
1

Works for me with GNU sed (version 4.2.1):

$ echo -ne $'\u9991' | sed 's/\xe9\xa6\x91//g' | hexdump -C
$ echo -ne $'\u9991' | hexdump -C
00000000  e9 a6 91

(As another replacement for sed you could also use GNU awk; but it don't seem necessary.)

1

Using Raku (formerly known as Perl_6)

~$ echo 汉典“馑”字的基本解释 | raku -pe 's:g/\x9991/Jin/;'
汉典“Jin”字的基本解释
~$ echo "饥馑荐臻" | raku -pe s:g/'\x9991'//;
饥荐臻

~$ raku -e 'print "e", "e\x301", "\x000e9";'
eéé
~$ raku -e 'say "e\x301" eq "\x000e9";'
True
~$ echo "Stephane" | raku -pe 's/e/e\x301/;'
Stéphane
~$ echo "Stephane" | raku -pe 's/e/\x000e9/;'
Stéphane

[Rakudo 2020.10; code tested on GNU bash, version 3.2.57(1)-release (x86_64-apple-darwin14]

https://raku.org/

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.