I'm using sed for updating my JSON configuration file in the runtime.
Sometimes, when the pattern doesn't match in the JSON file, sed still exits with return code 0.
Returning 0 means successful completion, but why does sed return 0 if it doesn't find the proper pattern and update the file? Is there a workaround for that?
as @cnicutar commented, the return code of a command means if the command was executed successfully. has nothing to do with the logic you implemented in the codes/scripts.
so if you have:
echo "foo"|sed '/bar/ s/a/b/'
sed will return 0 but if you write some syntax/expression errors, or the input/file doesn't exist, sed cannot execute your request, sed will return 1.
workaround
this is actually not workaround. sed has q command: (from man page):
q [exit-code]
here you can define exit-code as you want. For example '/foo/!{q100}; {s/f/b/}' will exit with code 100 if foo isn't present, and otherwise perform the substitution f->b and exit with code 0.
Matched case:
kent$ echo "foo" | sed '/foo/!{q100}; {s/f/b/}'
boo
kent$ echo $?
0
Unmatched case:
kent$ echo "trash" | sed '/foo/!{q100}; {s/f/b/}'
trash
kent$ echo $?
100
I hope this answers your question.
edit
I must add that, the above example is just for one-line processing. I don't know your exact requirement. when you want to get exit 1. one-line unmatched or the whole file. If whole file unmatching case, you may consider awk, or even do a grep before your text processing...
This might work for you (GNU sed):
sed '/search-string/{s//replacement-string/;h};${x;/./{x;q0};x;q1}' file
If the search-string is found it will be replaced with replacement-string and at end-of-file sed will exit with 0 return code. If no substitution takes place the return code will be 1.
A more detailed explanation:
In sed the user has two registers at his disposal: the pattern space (PS) in which the current line is loaded into (minus the linefeed) and a spare register called the hold space (HS) which is initially empty.
The general idea is to use the HS as a flag to indicate if a substitution has taken place. If the HS is still empty at the end of the file, then no changes have been made, otherwise changes have occurred.
The command /search-string/ matches search-string with whatever is in the PS and if it is found to contain the search-string the commands between the following curly braces are executed.
Firstly the substitution s//replacement-string/ (sed uses the last regexp i.e. the search-string, if the lefthand-side is empty, so s//replacement-string is the same as s/search-string/replacement-string/) and following this the h command makes a copy of the PS and puts it in the HS.
The sed command $ is used to recognise the last line of a file and the following then occurs.
First the x command swaps the two registers, so the HS becomes the PS and the PS becomes the HS.
Then the PS is searched for any character /./ (. means match any character) remember the HS (now the PS) was initially empty until a substitution took place. If the condition is true the x is again executed followed by q0 command which ends all sed processing and sets the return code to 0. Otherwise the x command is executed and the return code is set to 1.
N.B. although the q quits sed processing it does not prevent the PS from being reassembled by sed and printed as per normal.
Another alternative:
sed '/search-string/!ba;s//replacement-string/;h;:a;$!b;p;x;/./Q;Q1' file
or:
sed '/search-string/,${s//replacement-string/;b};$q1' file
These answers are all too complicated. What is wrong with writing a bit of shell script that uses grep to figure out if the thing you want to replace is there then using sed to replace it?
grep -q $TARGET_STRING $file
if [ $? -eq 0 ]
then
echo "$file contains the old site"
sed -e "s|${TARGET_STRING}|${NEW_STRING}|g" ....
fi
For 1 line of input. To avoid repeating the /pattern/:
When s succeeds to substitute, use t to jump conditionally to a label, e.g. x. Otherwise use q to quit with an exit code, e.g. 100:
's/pattern/replacement/;tx;q100;:x'
Example:
$ echo 1 > one
$ < one sed 's/1/replaced-it/;tx;q1;:x'
replaced-it
$ echo $?
0
$ < one sed 's/999/replaced-it/;tx;q100;:x'
1
$ echo $?
100
https://www.gnu.org/software/sed/manual/html_node/Branching-and-flow-control.html
Below is the pattern we use with sed -rn or sed -r.
The entire search and replace command ("s/.../.../...") is optional. If the search and replace is used, for speed and having already matched $matchRe, we use as fast a $searchRe value as possible, using . where the character does not need to be re-verified and .{$len} for fixed length sections of the pattern.
The return value for none found is $notFoundExit.
/$matchRe/{s/$searchRe/$replacement/$options; Q}; q$notFoundExit
For the following reasons:
Varying the case of Q commands will vary the behavior depending on when the exit should occur. Behaviors involving the application of Boolean logic to a multiple line input requires more complexity in the solution.
We have the answer above but it took some time for me work out what is happening. I am trying to provide a simple explanation for basic user of sed like me.
Lets consider the example:
echo "foo" | sed '/foo/!{q100}; {s/f/b/}'
Here we have two sed commands. First one is '/foo/!{q100}' This command actually check the pattern matching and return exist code 100 if no match. Consider following examples, -n is used to silent the output so we only get exist code.
This example foo matches so exit code return is 0
echo "foo" | sed -n '/foo/!{q100}'; echo $?
0
This example input is foo and we try match boo so no match and exit code 100 is returned
echo "foo" | sed -n '/boo/!{q100}'; echo $?
100
So if my requirement is only to check a pattern match or not I can use
echo "<input string>" | sed -n '/<pattern to match>/!{q<exit-code>}'
More examples:
echo "20200206" | sed -n '/[0-9]*/!{q100}' && echo "Matched" || echo "No Match"
Matched
echo "20200206" | sed -n '/[0-9]{2}/!{q100}' && echo "Matched" || echo "No Match"
No Match
Second command is '{s/f/b/}' is to replace the f in foo with b which I used many times.
For any number of input lines:
sed --quiet 's/hello/HELLO/;t1;b2;:1;h;:2;p;${g;s/..*//;tok;q1;:ok}'
Fills hold space on match, and checks it after the last line.
Returns status 1 if no match in file.
s/hello/HELLO - substitution to check fort1 - jump to label 1 if substitution succeededb2 - jump to label 2 unconditionally:1 - label 1h - copy pattern to hold space (when substitution succeeded):2 - label 2p - print pattern space, unconditionally${ ... } - match last line, evaluate block insideg - copy hold space into pattern space (non-empty if first substitution succeded before)s/..*// - dummy substitution, to set branch-flagtok - jump to label ok (if dummy substitution succeeded on non-empty hold space)q1 - exit with error status 1:ok - label ok
As we already know, when sed fails to match then it simply returns its input string - no error has occurred. It is true that a difference between the input and output strings implies a match, but a match does not imply a difference in the strings; after all sed could have simply matched all of the input characters. The flaw is created in the following example
h=$(echo "$g" | sed 's/.*\(abc[[:digit:]]\).*/\1/g')
if [ ! "$h" = "$g" ]; then
echo "1"
else
echo "2"
fi
where g=Xabc1 gives 1, while setting g=abc1 gives 2; yet both of these input strings are matched by sed! So, it can be hard to determine whether sed has matched or not. A solution:
h=$(echo "fix${g}ed" | sed 's/.*\(abc[[:digit:]]\).*/\1/g')
if [ ! "$h" = "fix${g}ed" ]; then
echo "1"
else
echo "2"
fi
in which case the 1 is printed if-and-only-if sed has matched.
I had wanted to truncate a file by quitting when the match was found (and exclude the matching line). This is handy when a process that adds lines at the end of the file may be re-run. "Q;Q1" didn't work but simply "Q1" did, as follows:
if sed -i '/text I wanted to find/Q1' file.txt then insert blank line at end of file + new lines fi insert just the new lines without the blank line
