Bash script: split word on each letter

Question

How can I split a word's letters, with each letter in a separate line?

For example, given "StackOver" I would like to see

S
t
a
c
k
O
v
e
r

I'm new to bash so I have no clue where to start.

Solution 1

I would use grep:

$ grep -o . <<<"StackOver"
S
t
a
c
k
O
v
e
r

or sed:

$ sed 's/./&\n/g' <<<"StackOver"
S
t
a
c
k
O
v
e
r

And if empty space at the end is an issue:

sed 's/\B/&\n/g' <<<"StackOver"

All of that assuming GNU/Linux.

Solution 2

You may want to break on grapheme clusters instead of characters if the intent is to print text vertically. For instance with a e with an acute accent:

• With grapheme clusters (e with its acute accent would be one grapheme cluster):

  $ perl -CLAS -le 'for (@ARGV) {print for /\X/g}' $'Ste\u301phane'
  S
  t
  é
  p
  h
  a
  n
  e
  

  (or grep -Po '\X' with GNU grep built with PCRE support)

• With characters (here with GNU grep):

  $ printf '%s\n' $'Ste\u301phane' | grep -o .
  S
  t
  e
  
  p
  h
  a
  n
  e
  

• fold is meant to break on characters, but GNU fold doesn't support multi-byte characters, so it breaks on bytes instead:

  $ printf '%s\n' $'Ste\u301phane' | fold -w 1
  S
  t
  e
  �
  �
  p
  h
  a
  n
  e
  

On StackOver which only consists of ASCII characters (so one byte per character, one character per grapheme cluster), all three would give the same result.

Solution 3

If you have perl6 in your box:

$ perl6 -e 'for @*ARGS -> $w { .say for $w.comb }' 'cường'       
c
ư
ờ
n
g

work regardless of your locale.

Solution 4

With many awk versions

awk -F '' -v OFS='\n' '{$1=$1};1' <<<'StackOver'

Solution 5

You can use the fold (1) command. It is more efficient than grep and sed.

$ time grep -o . <bigfile >/dev/null

real    0m3.868s
user    0m3.784s
sys     0m0.056s
$ time fold -b1 <bigfile >/dev/null

real    0m0.555s
user    0m0.528s
sys     0m0.016s
$

One significant difference is that fold will reproduce empty lines in the output:

$ grep -o . <(printf "A\nB\n\nC\n\n\nD\n")
A
B
C
D
$ fold -b1 <(printf "A\nB\n\nC\n\n\nD\n")
A
B

C


D
$ 

View more solutions

Comments

[Sijaan Hallak]

grep -o . <<< ¿¿¿ .. -o searches for the PATTERN provided right? and what it does here in your command?

[jimmij]

@SijaanHallak grep searches for pattern, an in this example it searches for every character . and prints it in the separate line. See also sed solution.

[Sijaan Hallak]

Thanks! so this "." dot means every character.. Can you please give me a link where I can read about things such as this dot? or what ar these things called?

[jimmij]

I'm surprised grep -Po doesn't do what one would expect (like grep -P does).

[Stéphane Chazelas]

Note that both -o and \n are a GNU extension. <<< is a zsh extension (also available in recent versions of ksh93 and the GNU shell (bash)).

[Stéphane Chazelas]

@jimmij, what do you mean? grep -Po . finds characters (and a combining acute accent following a newline character is invalid), and grep -Po '\X' finds graphem clusters for me. You may need a recent version of grep and/or PCRE for it to work properly (or try grep -Po '(*UTF8)\X')

[jimmij]

@SijaanHallak The best manual you have already on you computer, just run man grep and then just look for the chapter "REGULAR EXPRESSIONS" (if that is what you are interested in).

[Avinash Raj]

Second answer would produce a new line after last...

[cuonglm]

NP, should we add a note about the locale?

[Sijaan Hallak]

@jimmij I cant find any help on what <<< really does! any help?

[Sijaan Hallak]

This won't help as it prints a new line at the end

[jimmij]

@SijaanHallak This is so called Here string, grosso modo equivalent of echo foo | ... just less typing. See tldp.org/LDP/abs/html/x17837.html

[kay]

Does not work for combining characters like Stéphane Chazelas answer, but with proper normalization this should not matter.

[mikeserv]

@Kay - it's works for combining characters if you want it to - that's what sed scripts are for. i'm not likely to write one right about now - im pretty sleepy. it's really useful, though, when reading a terminal.

[mikeserv]

@cuonglm - if you like. it should just work for the locale, given a sane libc, though.

[Sijaan Hallak]

@jimmij the second solution here seems to have a problem. it prints a new line at the end! I changed it to this sed -e 's/./\n&/g' <<< "$1" But this prints a new line at the beggining.. any suggestion how to overcome this?

[jimmij]

@SijaanHallak change . to \B (doesn't match on word boundary).

[Sijaan Hallak]

@jimmij \B will not work as it prints "Stack Over" -> the "O" will be printed near the letter "k" at the same line and then it does \n

[jpmc26]

@SijaanHallak These might be helpful: joelonsoftware.com/articles/Unicode.html, eev.ee/blog/2015/09/12/dark-corners-of-unicode

[Stéphane Chazelas]

Note that dd will break multibyte characters, so the output will not be text anymore so the behaviour of sed will be unspecified as per POSIX.

[mikeserv]

@StéphaneChazelas - do you have a link to reference that statement? a NUL can't occur in a multibyte character, and a dot can only match a whole character which is not NUL, and it has worked with every sed i've tried. how could it not work?

[mikeserv]

oh wait - you mean because input isn't a text file. possibly, but sed is spec'd to handle conditions which exceed/break text file specs, too, such as 4k pattern spaces scripts which is well beyond line max. its also spec'd to evaluate chars bytewise w/ l - even when a single char is multiple bytes. i think the text file restriction for sed is probably based on the NUL prohibition - many seds replace delimiter in their scripts w/ NULs, and ive never managed to seek past a NUL in pattern space with heirloom sed except with D and G.

[mikeserv]

@SijaanHallak - you can drop the second sed like: sed -et -e's/./\n&/g;//D'

[Yunus]

since each byte have a width=1 the result will be the same !

[VocalFan]

So how is this not a duplicate of the earlier answer?

[Yunus]

because it shows tha same cmd with different argyment , and that is nice to know .

[eruve]

Great! But on my version of nAWK ("One True AWK") that doesn't work. However this does the trick: awk -v FS='' -v OFS='\n' '{$1=$1};1' (wondering if that's more portable since -F '' might yield the ERE: //)

[done]

This removes white space from the original string.

[done]

An eval could be a big risk, a double eval is even more risky. Specially with arbitrary input from $s. Just saying !!

[done]

Are you claiming that $'e\u301' is equivalent/equal to é ?

[Stéphane Chazelas]

@Isaac, no, I'm not claiming any such thing though there are some definitions of "equivalent" for which that would be true.

[done]

Your description seems to imply that because Perl is able to join together characters and accents (much like a text editor join them to select an specific glyph) other software should be able also. But no, not all programs are text editors, Nor all utilities understand the complex (specially in Hangul) set of rules to join some individual Unicode codepoints (unicode.org/reports/tr29 and search for Devanagari kshi). So, no, nor grep, sed or fold understand any of this issue (yet).