How to replace a character between a pattern using sed - sed

Question

I am trying to replace a hyphen character - in the key name of a JSON file with _ without impacting the value side of the key-value pair.
Example input:
{
"outcome": "failed",
"failure-description": "ra ra ra - and more",
"rolled-back": true
}
Is there any way to do this using sed? It could be a match pattern where sed would only replace between "(.*[^"])": but I have not been able to work out how to replace the unwanted character in the matched substring.
The expected result would look like this:
{
"outcome": "failed",
"failure_description": "ra ra ra - and more",
"rolled_back": true
}

Answer 1

This would work:
$ sed 's/-\([^:]*\):/_\1:/' infile
{
"outcome": "failed",
"failure_description": "ra ra ra - and more",
"rolled_back": true
}
This looks for a - followed by a captured series of characters other than a colon, and then a colon; it replaces the hyphen by an underscore, and puts the captured group and the colon back.
A limitation of this is that it only replaces the first hyphen. Assume our input looks like this:
{
"outcome": "failed",
"failure-description": "ra ra ra - and more",
"two-hyphens-here": "ra ra ra - and more",
"rolled-back": true
}
To replace all hyphens before the colon, we can use conditional branching:
$ sed ':a;s/-\([^:]*\):/_\1:/;ta' infile
{
"outcome": "failed",
"failure_description": "ra ra ra - and more",
"two_hyphens_here": "ra ra ra - and more",
"rolled_back": true
}
This sets a label (:a) and uses the t command (branch to label if a the pattern space was changed).
For BSD sed as found in Mac OS, the label has to be in a separate command:
sed -e ':a' -e 's/-\([^:]*\):/_\1:/;ta' infile
Notice that an inherent limitation of all this is that there must not be any colons between the quotes and it's generally advisable to use a proper JSON parser such as jq to do this kind of manipulation.

Answer 2

Use extended regular expressions and parenthesis structures.
-r, --regexp-extended
use extended regular expressions in the script.
This produces the correct results, but some adjustment may be needed to harden the regular expression against false matches:
sed -re 's/([:alpha:]*)[-]([:alpha:]*)/\1_\2/'
Result:
{
"outcome": "failed",
"failure_description": "ra ra ra - and more",
"rolled_back": true
}
Note that the simple expression given above is inadequate if the value side contains the pattern. Examine your data set, added more parenthesized expressions and references to them as needed to anchor the match more tightly. It is possible to nest parenthesized expressions, though that does complicate determining the back reference to it.
$ sed --version
GNU sed version 4.1.5

Answer 3

Just use awk:
$ awk 'BEGIN{FS=OFS="\": \""} {gsub(/-/,"_",$1)} 1' file
{
"outcome": "failed",
"failure_description": "ra ra ra - and more",
"rolled_back": true
}