Vim / sed regex backreference in search pattern

问题:

Vim help says that:

\1      Matches the same string that was matched by     */\1* *E65*
        the first sub-expression in \( and \). {not in Vi}
        Example: "\([a-z]\).\1" matches "ata", "ehe", "tot", etc. 

It looks like the backreference can be used in search pattern. I started playing with it and I noticed behavior that I can't explain. This is my file:

<paper-input label="Input label"> Some text </paper-input>
<paper-input label="Input label"> Some text </paper-inputa>
<aza> Some text </az>
<az> Some text </az>
<az> Some text </aza>

I wanted to match the lines where the opening and closing tags are matching i.e.:

<paper-input label="Input label"> Some text </paper-input>
<az> Some text </az>

And my test regex is:

%s,<\([^ >]\+\).*<\/\1>,,gn

But this matches lines: 1, 3 and 4. Same thing with sed:

$ sed -ne 's,<\([^ >]\+\).*<\/\1>,\0,p' file
<paper-input label="Input label"> Some text </paper-input>
<aza> Some text </az>
<az> Some text </az>

This: <\([^ >]\+\) should be greedy and when trying to match it without \1 at the end then all the groups are correct. But when I add \1 it seems that <\([^ >]\+\) becomes not greedy and it tries to force the match in 3rd line. Can someone explain why it matches 3rd line:

<aza> Some text </az>

This is also a regex101 demo

NOTE This is not about the regex itself (probably there is other way to do it) but about the behavior of that regex.

回答1:

To understand why your regex behaves the way it does you need to understand what a backtracking regex engine does.

The engine will greedily match and consume as many characters as it can. But if it doesn't find a match it goes back and tries to find a different match that still satisfies the pattern.

%s,<\([^ >]\+\).*<\/\1>,,gn

For line three <aza> Some text </az>,

The regex engine looks at \1 = aza. and sees if .*</aza> matches the rest of the string. It doesn't so it chooses something else for \1. The next time it chooses \1 = az and sees if .*</az> matches the rest of the string and it does. So the string matches

(This is a simplified version. I skipped over the fact that .* can potentially do a lot of backtracking itself)

---

Solving it is as easy as adding an anchor in the regex stops the regex from searching for other values that could satisfy \1. In this case matching a space or > is sufficient.

回答2:

You need to add \> to indicate end of word. There may be other solutions with 0-width patterns, but it'll complicates things.

Also, your separator is ,, not /

Which gives:

%s,<\([^ >]\+\)\>.*</\1>,,gn

回答3:

Currently the reason why line 3 (<aza>) is showing up as a match is that the .* term in your regex can match across multiple lines. So line 3 matches because line 5 has the closing tag. To correct this, force the regex to find a matching closing tag on the same line only:

%s,<\([^ >]\+\)[^\n]*?<\/\1>,,gn
               ^^^^^ use [^\n]* instead of .*