haxe-checkstyle
haxe-checkstyle copied to clipboard
HaxeCheckstyle
Silent crash on Windows with very long line
Checkstyle dies on this line:
https://github.com/HaxeFlixel/flixel/blob/4.2.0/flixel/graphics/frames/FlxBitmapFont.hx#L31
And apparently only on Windows.
Easiest way to reproduce is to clone the flixel repo and run haxelib run checkstyle -s . -progress - FlxBitmapFont.hx will be the last file displayed before it exits.
I know that line is the one causing the problem because it works fine after it's removed.
I took a quick look at the issue on a Windows 10 machine, here is what I found:
neko.exe crashes in regexp.ndll with a stackoverflow
I have narrowed it down to line based checks (tested with IndentationCharacter, TabForAligning, TrailingWhitespace), so I guess it has to do with the multiline / string detection stuff.
Hm, I guess it must be a Neko bug then. :/ Perhaps it's trying to run a regex over that long line and the regex engine can't handle strings that long on Windows for some reason?
I think it might be more of an "evil" regex problem ( https://www.owasp.org/index.php/Regular_expression_Denial_of_Service_-_ReDoS ) Maybe it works on Linux / OSX, because neko uses larger stacksizes on those systems. Just wild guesses, I haven't looked any further.
A solution might be to change our regexes, so that there is less need for backtracking or recursion. Though I'm not sure if that's possible.
We should definitely at least find a workaround for crashes like this... Though that might be tricky. Maybe some heuristic like "if the line is longer than x characters, don't risk it". :/
As a workaround we could make lines above 1000 (or 2000, needs testing) spit out a checkstyle message for all line based checks. So that users get some kind of feedback on those files, and can fix line length or exclude checks.
A more stable solution might be to go hybrid as proposed in #278, where comment and string detection comes though token tree. But that will come with some performance impact.
I doubt that's necessary, this seems like a pretty rare edge case considering it hasn't come up until now.
Just ran into what seems like this issue as well. In particular on the TabForAligning, EmptyLines, and IndentationCharacter checks.
We don't use the TrailingWhitespace check. All these checks extend LineCheckBase.
Neko returning 3221225725, which is apparently a stack overflow?
Happy to investigate a fix and do a pull request. What version of haxe should I be using to build?
To add it was also due to an extremely long line in a test.
I understand the priority is pretty low due to it being an edge case, but its taken me a good couple hours to track down. So again, let me know what compiler version I should be using and I will attempt a fix.
I would recommend Haxe 4rc5 for compiling git checkstyle.
I haven't done extensive tests, but it seems running a nodejs version of checkstyle might be bit more reliable.
Maybe we need to do the same nodejs switcheroo as we do with formatter. And maybe I should make that checkstyle release, that I keep postponing...