Question

I’m having trouble using sed to replace non-printable characters with other non-printable characters.
Specifically, I want sed to look for a line in a table starting with a TAB and ‘insert’ a BACKSPACE, essentially bringing the text from that line up to the previous line.

The reason for this is that I have a table split into columns, with some rows taking up multiple lines:

column1,row1    column2,row1,line1
                column2,row1,line2

I’d like it to end up like this (ish)

column1,row1    column2,row1,line1         column2,row1,line2

(the spacings don’t matter)

I can’t seem to manage this though. I’m unsure if the expressions \t sand \b aren’t recognized, but they don’t seem to be in the way I’ve tried. I also can’t get echo commands with octal representations of these

These are the kind of things I’ve tried:

sed 's/^\t/\b/' file.txt newfile.txt


sed 's/^(`echo "\011"`)/`echo "\010"`/' file.txt newfile.txt

Any help would be appreciated.

Answer 1

I wouldn’t do this with sed. Sure, it’s possible to have sed store a line in a hold buffer, then read the next line and if it starts with whitespace, pop and print the hold buffer then print the current line … but the code will look like your modem just lost carrier.

I’d use awk.

awk '/^[^ \t]/ && buf{print buf;buf=""} {buf=buf $0} END{print buf}' input.txt

What does this do?

• The second section, {buf=buf $0}, adds the current line to a buffer named buf.
• Then, the first section, /^[^ \t]/ && buf{print buf;buf=""} (which is ignored for the first line because buf has not yet been set), prints buf for any line that doesn’t start with whitespace (i.e. the start of a new output line). Then it resets buf.
• The last section, END{print buf}, prints any leftovers.