lm_sensors: update broken json parser

[lasers]

This changes how we parse broken json blocks from upstream. Instead of splitting chunks based on }\n{ , this just run every line until we get a valid json chunk before we add it to a dictionary.

Ought to address #2226 (comment). /cc @MacGyverNL for testing.

[MacGyverNL]

Gonna be a few days before I can test this, I'm afraid. The system on which this is a problem is currently undergoing maintenance.

[MacGyverNL]

Can confirm this works, although with this splitting, my case never hits the codepath with JSONDecodeErrors. So I can't testify to the effectiveness / robustness of building the chunk across multiple lines.

[lasers]

@MacGyverNL Can you test this again? The first commit was simple, but will fail on several lines (repeatedly) until we get a valid JSON (again, repeatedly). I took second look at this and wondered if we can make do with this (second commit) instead.

[MacGyverNL]

Your most recent version does not chunk, I think you mishandled the caret and dollar placement. I don't know what your goal was with those, but if I fix those naively (r'}$\n^{'), then it crashes again on the first chunk, because the split still gobbles up one closing bracket, and they end up unbalanced. Don't you want something like a positive lookbehind / lookahead, along the lines of r'(?<=})$\n^(?={)'? But I don't know how expensive lookbehind / lookahead matching is when compared to the older failure type.

[lasers]

Noted! It's a better solution. Please try it again.

[MacGyverNL]

Yeah this should work (edit for clarity: the commit I tested, 773b9ff, doesn't), once you get the ^ and $ right. I think what's confusing you is where to match start- and end-of-line, especially in multiline mode. These aren't indicators of where the pattern starts, they're match points you'd normally use to match only at the start and end of a line. The way I read the documentation, in multiline mode they still match right before and after each newline. So in principle, because we're explicitly matching the newline, you don't actually need them at all. self.pattern = re.compile(r'(?<=})\n(?={)', re.MULTILINE) suffices (and works, already tested it).

(I know I could commit this but I don't currently have a fork, I'm just manual-editing the python code of my system package, and it's a single line.)

[lasers]

Thank you for your help / looking into this. We should be good here now.

[lasers]

@ultrabug \o/