Bash – Elegant way to merge lines with multi-char-delimiter, ignoring blank lines, supporting \n, \r or \r\n

awkbashperlsedtext processing

I want to read a multi-line file in a bash script, using the file path from a variable, then merge the lines using a multi-char delimiter and save the result to another variable.

I want to skip blank lines and trailing new lines and do not want a trailing delimiter.

Additionally I want to support \r\n and – if at no further "cost" – why not also \r as line break (and of course \n).

The script should run on RHEL with GNU's bash 4.2.46, sed 4.2.2, awk 4.0.2, grep 2.20, coreutils 8.22 (tr, cat, paste, sort, cut, head, tail, tee, …), xargs 4.5.11 and libc 2.17 and with perl 5.16.3, python 2.7.5 and openjdk 11.0.8.

It should run about twice per day on files with ca. 10 lines on a decent machine/VM.
If readability, maintainability and brevity don't suffer too much I'm very open to more performant solutions though.

The files to be read from can be created and modified either on the same machine or other Win7 or Win10 systems.

My approach so far is

joined_string_var=$(sed 's/\r/\n/g' $filepathvar | grep . | sed ':a; N; $!ba; s/\n/; /g')

  • So first I replace \r with \n to cover all newline formats and make the output readable for grep.

  • Then I remove blank lines with grep .

  • And finally I use sed for the actual line merging.

I used sed instead of tr in the first step to avoid using cat, but I'm not quite sure if I prefer it like that:

joined_string_var=$(cat $filepathvar | tr '\r' '\n' | grep . | sed ':a; N; $!ba; s/\n/; /g')

UPDATE: I somehow completely missed simple redirection:

joined_string_var=$(tr '\r' '\n' <$filepathvar | grep . | sed ':a; N; $!ba; s/\n/; /g')

Any thoughts how this might be done more elegantly (less commands, better performance, not much worse brevity and readability)?

Best Answer

The elegance may come from the correct regex. Instead of changing every \r to a \n (s/\r/\n/g) you can convert every line terminator \r\n, \r, \n to the delimiter you want (in GNU sed, as few sed implementations will understand \r, and not all will understand -E):

sed -E 's/\r\n|\r|\n/; /g'

Or, if you want to remove empty lines, any run of such line terminators:

sed -E 's/[\r\n]+/; /g'

That will work if we are able to capture all line terminators in the pattern space. That means to slurp the whole file into memory to be able to edit them.

So, you can use the simpler (one command for GNU sed):

sed -zE 's/[\r\n]+/; /g; s/; $/\n/' "$filepathvar"

The -z takes null bytes as line terminators effectively getting all \r and \n in the pattern space.

The s/[\r\n]+/; /g converts all types of line delimiters to the string you want.

The s/; $/\n/ converts the (last) trailing delimiter to an actual newline.

Notes

The -z sed option means to use the zero delimiter (0x00). The use of that delimiter started as a need of find to be able to process filenames with newlines (-print0) which will match the xargs (-0) option. That meant that some tools were also modified to process zero delimited strings.

That is a non-posix option that breaks files at zeros instead of newlines.

Posix text files must have no zero (NIL) bytes, so the use of that option means, in practice, to capture the whole file in memory before processing it.

Breaking files on NILs means that newline characters end being editable on the pattern space of sed. If the file happens to have some NIL bytes, the idea still works correctly for newlines, as they still end being editable in each chunk of the file.

The -z option was added to GNU sed. The ATT sed (on which posix was based) did not have such option (and still doesn't), some BSD seds also still don't.

An alternative to the -z option is to capture the whole file in memory. That could be done Posixly in some ways:

sed 'H;1h;$!d'          # capture whole file in hold space.
sed ':a;N;$!ba'         # capture whole file in pattern space.

Having all newlines (except the last one) in the pattern space makes it possible to edit them:

sed -Ee 'H;1h;$!d;x'   -e 's/(\r\n|\r|\n)/; /g

With older sed's it is also required to use the longer and more explicit (\r\n|\r|\n)+ instead of [\r\n]+ because such sed's don't understand \r or \n inside bracket expressions [].

Line oriented

A solution that works one line at a time (a \r is also a valid line terminator in this solution), which means that there is no need to keep the whole file in memory (less memory used) is possible with GNU awk:

awk -vRS='[\r\n]+' 'NR>1{printf "; "}{printf $0}END{print ""}'  file

Must be GNU awk because of the regex record separator [\r\n]+. In other awk, the record separator must be a single byte.

Related Question