150GiB JSON

People do actual JSON that large without resorting to JSONL or similar? Yikes, that is the kind of size where a streaming parser would be nice to have (like SAX parsers for XML. Dunno what is possible for JSON).

JSON::PullParser is a streaming JSON parser, the file was an array of 9 million items and I was categorising each entry as it came in and splitting it out into smaller more manageable files.

Is there a reason not to use unsigned here?

Unsigned types are more susceptible to errors on math operations. Technically, the domain range would be covered by unsigned values. But they propagate into other calculations and cause issues there.
We won't need the full range of UInt64 here for parsing JSON files over 8 exabytes.

I've marked the line/column accessors experimental, and the i32 methods now raise on overflow instead of saturate. I think this PR can probably be reviewed in more detail unless there's still concerns about the overall approach.

@RX14 We're in the middle of the freeze period for 1.18.0. This is a feature addition and shouldn't have been merged into master until after the release.

It seems I forgot adding it to the milestone after I approved it. It should've been added to 1.19.0 though.
But we should never merge anything before adding it to a milestone (see https://github.com/crystal-lang/crystal/blob/master/CONTRIBUTING.md#accepting-a-pull-request).

The commit has been reverted.
Please resubmit this PR so we can merge it for 1.19.

Sorry, that's my bad for not reading the processes. I've re-made the PR in #16211.