Fix empty regular expression matches (fix #2565)

[itchyny]

As reported by #2565, "ab" | match(""; "g") should yield 3 matches at offset 0, 1, and 2. Fixing this behavior also fixes "a" | gsub(""; "a") to emit "aaa" not "aa". Fixes #2565.

[pkoppstein]

@itchyny - Congratulations!

You could remove the disclaimer at onig.test:91
("The following is a regression test ...")

[nicowilliams]

Thanks!

[emanuele6]

Fixing this behavior also fixes "a" | gsub(""; "a") to emit "aaa" not "aa".

I think that is a bug, it should emit "aa", no?

[itchyny]

Why? Empty regex matches every character boundaries. Easier example is "xyz" | gsub(""; "a") should be axayaza

[emanuele6]

@itchyny
Oh, you are right, I got confused.

[pkoppstein]

I think that is a bug, it should emit "aa", no?

No. @itchyny and (not surprisingly :-) gojq are correct.

In any case, gsub relies (and ought to rely) on match(_;"g"). In this case, the call to match produces two results, ergo two insertions.

[emanuele6]

I was used to the sed behaviour where sed 's/1*/2/g' <<< 11' outputs 2, but it looks like in javascript and other languages, if the pattern is able to match the empty string, it matches the empty string at the end of the input even if the previous match goes up to the last character of the input: node -p '"1111".replaceAll(/1*/g, "2")' outputs 22

[emanuele6]

jq's gsub used to behave like sed's s///g, but now it behaves more like javascript's replaceAll after this patch:

$ jq -n '"1111" | gsub("1*";"2")'
2
$ ./jq -n '"1111" | gsub("1*";"2")'
22

Fine, I guess