Fix CoreScalarResolver.INT regex to work with JS

[krzema12]

Found when converting the corresponding test to KMP: #397

What helped me here is this tool that renders the regex as a graph: https://www.debuggex.com/

Before the fix - I noticed that ^ and $ are attached to the groups nearest to them:

After the fix - creating a new group that contains the rest fixes the problem:

[Vampire]

Can you explain further, please?
Because that does not really make any sense at all.
The previous regex should match fine.
And (unless I miscopied on mobile) debuggex is also happy with the previous state.

The change you made does not change any functionality as long as whole strings are matched and even then the test would not fail as even a substring match would hold, besides that the change is part of my readability PR, just with non-capturing groups.

[krzema12]

@Vampire the regex before this change led to test failures in #397. See this run: https://github.com/krzema12/snakeyaml-engine-kmp/actions/runs/13174353963/job/36770339573 and the build scan: https://scans.gradle.com/s/k4ayl4yetgdve/tests/overview?outcome=Failed. The change merged in this PR fixes the problem, that's it. For me, this proof is enough. We'll convert the group into a non-capturing one in your regex-related PR.

[krzema12]

@Vampire if course if you see a other reason why the tests failed, let me know! This case also intrigued me, so I want to understand it well.

[Vampire]

I had a closer look and it is like I said, it doesn't make any sense.
The reason is, that we found a bug in K/JS.
The docs of Regex#matches expectedly say

Indicates whether the regular expression matches the entire input.

The problem is, that the K/JS Implementation at https://github.com/JetBrains/kotlin/blob/de5284d7e4af5b94d9d4144119134eb0285a5fa9/libraries/stdlib/js/src/kotlin/text/regex.kt#L97 is bullshit. It executes the regex against the input, gets a result that might only match a part (in this case just 0 matched my the first alternative, then compares whether that match is starting at first and ending at last character. That it could have matched the entire input too is missed and thus the result of the method wrong.

Have a look at https://pl.kotl.in/sfmoOkwcB. With JVM all four cases are correctly true, with either of the JS targets the first case is wrongly false, as only the first character is matched, even though all could match.

[Vampire]

Actually due to that bug, you should probably go through all regexes and check whether they are intended to match the entire input and add the anchors and non-matching group of they should. In my PR I only fixed the cases you fixed here, where the group was missing.

[krzema12]

Nice find! Indeed nativePattern.lastIndex == input.length looks shady. FWIW, the JS-specific unit tests don't cover it (ref), and I haven't found common tests for it.

So yeah, this PR turns to be a workaround for the issue you found. Are you planning to report it?

[Vampire]

No, I'm planning to fix it, I don't have a commit in Kotlin yet. :-D

[Vampire]

Actually it is even worse.
They don't only not handle entire match in matches properly,
but also try to be overly clever in matchEntire, using the original pattern if it starts with ^ and ends with $ which is wrong of course and if not, strips any ^ from start and $ from end and places the same there without a group which of course is also wrong.
🙈

[Vampire]

JetBrains/kotlin#5402