GNU bug report logs - #49873
Replacing all \n with spaces doesn't work in GNU sed as expected

Previous Next

Full log

View this message in rfc822 format

From: Assaf Gordon <assafgordon <at> gmail.com>
To: AlvinSeville7cf <alvinseville7cf <at> gmail.com>, 49873 <at> debbugs.gnu.org
Subject: bug#49873: Replacing all \n with spaces doesn't work in GNU sed as expected
Date: Wed, 4 Aug 2021 14:07:22 -0600

tag 49873 notabug
close 49873
stop

Hello,

On 2021-08-04 4:27 a.m., AlvinSeville7cf wrote:
> Hello! I want to read entire file and then replace all *\n* with space. 

For that I would recommend using 'tr' - it'll be much faster:

    tr '\n' ' ' < input > output


> My sed script is (I know that it is not optimal but it demonstrates 
> problem):
> 
> |:a $! { N; ta } s/\n/ /g p |

The above script isn't valid as-is (perhaps line breaks were lost in the 
email?).

I'm going to assume you meant the following script, and used "sed -n":

   sed -n ':a $! { N; ta } ; s/\n/ /g ; p' < input > output

or with line breaks:

   sed -n ':a
           $! { N; ta }
           s/\n/ /g
           p' < input > output


> So why even with *g* flag *s* command replaces only first *\n* in 
> pattern space? For instance I have the following file:

You script is almost correct :)
I assume that with the "$!{N;ta}" command you meant to accumulate all
lines except the last in the pattern space, and then replace all
the new lines and print the patern space.

The only 'bug': "t" is "conditional jump".
It jumps once to label "a", accumulating one more line, but then
doesn't jump again - so the "s///" is executed and the two lines are 
printed (and one newline replaced with space). The "s///" command also 
resets the "t" conditional, so the next line (3rd line in the input 
file) then does causes a jump.

Observe:

  $ seq 10 | sed -n ':a $! { N; ta } ; s/\n/ /g ; p'
  1 2
  3 4
  5 6
  7 8
  9 10

If you replace the "t" with a "b" command (b = always jump),
it behaves as you expected:

  $ seq 10 | sed -n ':a $! { N; ba } ; s/\n/ /g ; p'
  1 2 3 4 5 6 7 8 9 10

Note that even with this script, the last newline is preserved and
printed.

As a work-around, you can instruct "sed" to use NUL as line-breaks,
causing "\n" characters to be treated like any other character:

   $ seq 10 | sed -z 's/\n/ /g'
   1 2 3 4 5 6 7 8 9 10

But this won't be as efficient as using 'tr'.


> |It was the best of times, it was the worst of times, it was the age of 
> wisdom, it was the age of foolishness, |
> 
> The result of script execution is:
> 
> |It was the best of times, it was the worst of times, it was the age of 
> wisdom, it was |

> I use GNU sed 4.8. It seems to be a bug.

Without line breaks it's a bit hard to reproduce your case,
but I hope the explanation above was sufficient.

As such I'm closing this as "not a bug",
but discussion can continue by replying to this thread.

regards,
 - assaf

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.