Command substitution within sed expression

问题

I'm having little problem with bash/sed. I need to be able to use command substitution within sed expression. I have two big text files:

first is logfile.txt which sometimes* shows error messages by ID (0xdeadbeef is common example) in format ERRORID:0xdeadbeef
second errors.txt has error messages stored in pairs LONG_ERROR_DESCRIPTION, 0xdeadbeef

I was trying to use sed with bash command substitution to do the task:

cat logfile.txt | sed "s/ERRORID:\(0x[0-9a-f]*\)/ERROR:$(cat errors.txt |
    grep \1 | grep -o '^[A-Z_]*' )/g"

(^^^ this should be in one line of course)

If it would work then I could get little nicer version of logfile with better error info.

   Lot's of meaningless stuff ERRORID:0xdeadbeef and something else =>
=> Lot's of meaningless stuff ERROR:LONG_ERROR_DESCRIPTION and something else

But it doesn't. The problem is that sed is unable to "inject" regex section (\1) into command substitution. What are my other options? I know that it's possible to build sed expression first or do it other way but I would like to avoid parsing those files several times (they can be huge).

As always big thanks for any help.

*there is no real formatting inside logfile. No sections, columns, tab/coma-separation are used inconsistently

PS. Just to explain. Following expression works, but of course there is no argument passing within it:

echo "my cute cat" | sed "s/cat/$(echo dog)/g"

回答1:

You can create a sed script from the error message catalog, then apply that sed script to the log file.

Basically, something along these lines:

sed 's/\(.*\), 0x\([0-9A-F]*\)$/s%ERRORID:0x\2%ERROR:\1%g/' errors.txt |
sed -f - logfile.txt

The output from the first sed script should be something like this:

s%ERRORID:0x00000001%ERROR:Out of memory%
s%ERRORID:0x00000002%ERROR:Stack overflow%
s%ERRORID:0x00000031%ERROR:values of beta may cause dom%

That is, a new sed script which specifies a substitution for each error code in the catalog.

There are different dialects of sed so this may require minor tweaking. The sed on Linux I believe should use backslash before grouping parentheses in regular expressions, and gladly tolerate standard input as the argument to the -f option. This is not portable to other Unices, though (but you could substitute Perl for sed if you need portability).

*Edit: If the error messages are fairly static, and/or you want to read the log from standard input, save the generated script in a file;

# Do this once
sed 's/\(.*\), 0x\([0-9A-F]*\)$/s%ERRORID:0x\2%ERROR:\1%g/' errors.txt >errors.sed
# Use it many times
sed -f errors.sed logfile.txt

You could also add #!/usr/bin/sed -f at the top of errors.sed and chmod +x it to make it into a self-contained command script.

回答2:

I don't know if this would work, since I can't get an answer on whether or not capture groups persist, but there is a lot more to sed than just the s command. I was thinking you could use a capture group in a regex line selector, then use that for the command substitution. Something like this:

/ERRORID:\(0x[0-9a-f]*\)/  s/ERRORID:0x[0-9a-f]*/ERROR:$(grep \1 errors.txt | grep -o '^[A-Z_]*' )/

Anyway, if that doesn't work I would change gears and point out that this is really a good job for Perl. Here's how I would do it, which I think is much cleaner / easier to understand:

#!/usr/bin/perl

while(<>) {
  while( /ERRORID:(0x[0-9a-f]*)/ ) {
    $name = system("grep $1 errors.txt | grep -o '^[A-Z_]*'");
    s/ERRORID:$1/ERROR:$name/g;
  }
  print;
}

Then execute:

./thatScript.pl logfile.txt

回答3:

With GNU awk for gensub() and the 3rg arg to match():

$ awk '
    NR==FNR {
        map[$NF] = gensub(/,[^,]+$/,"",1)
        next
    }
    match($0,/(.*ERRORID:)(0x[[:xdigit:]]+)(.*)/,a) {
        $0 = a[1] (a[2] in map ? map[a[2]] : a[2]) a[3]
    }
1' errors.txt logfile.txt
Lot's of meaningless stuff ERRORID:LONG_ERROR_DESCRIPTION and something else =>

The above will run much faster than the sed scripts in the currently accepted answer and won't fail given various possible contents of LONG_ERROR_DESCRIPTION such as % or & or \1, nor will it fail when a given ERRORID is a subset of another, e.g. if 0xdead and 0xdeadbeef are 2 separate error codes then the sed scripts can fail depending on the order they appear in errors.txt, e.g. they could convert ERRORS:0xdeadbeef to ERRORS:LONG_ERROR_DESCRIPTIONbeef. by mapping 0xdead first.

回答4:

Just to let people looking for solution with bare shell and sed. Not perfect but working:

cat logfile.txt | while read line ; do id=$(echo -E "$line" | 
    grep "ERRORID:0x[0-9a-f]*" | grep -o "0x[0-9a-f]*" ) ; 
    if [ ! -z "$id" ] ; then echo -E "$line" | sed "s/$id/$(grep $id errors.txt | 
    grep -o '^[A-Z_]*' )/g" ;else echo -E "$line" ; fi ; done

If you see some fixing options then please share.

来源：https://stackoverflow.com/questions/7666149/command-substitution-within-sed-expression

标签

Linux

bash

sed

substitution