Complex sed search and replace (multiline, regexp, non-greedy)

I had a large chunk of text (output from another command), and needed to perform a search and replace on it. I usually do that either with zsh builting ${var:gs/x/y/} syntax or sed, but this time my pattern was spread on several lines.

Multi-line with --null-data

The trick here is to use sed ---null-data to make it operate on the full text instead of on invidual lines. Technically, it now considers a "line" to end with a NUL character, and not a \n anymore.

Improve readability with --regexp-extended

To improve readability of my regexps, I also used --regexp-extended which allows me to write capturing groups without having to escape them ((.*) instead of \(.*\)).

Non-greedy

sed does not have a non-greedy mode, which means it will always capture the largest group it can.

The way to work around that involves writing a slightly more complex regexp by specifically defining the one character we don't want to capture.

For example with the string foo_bar_baz, I might want to find the part before the first _ by doing ^(.*)_. This will actually return foo_bar because it's greedy.

The way to work around that is to use ^([^_]*)_ instead. This can be read as: capture everything that is not a _ from the start until you met a _.