"SortedSeek" option?

[RMGiroux]

Interesting tool!

I see your log files don't have timestamps, but if they did, it'd be useful to be able to focus in on a time, rather than a percentage of the file.

It might be nice to have a File::SortedSeek-like option to specify a key substring to bisect a sorted file on to focus down on the window when the problem occurred.

[matklad]

Ah, that's smart! Indeed, we can binary-search the file and still be within our time budget! I don't intend to add this myself, but can accept a PR if it comes with the tests (there's already some testing infra)

For unsorted files, the anchor is useful:

• you can use anchor to skip to the first occurence within the window
• the position of that occurence is printed as pos: field in the output
• you can paste that position back into window.toml, and remove anchor

This is how I find, eg, the first line where the view becomes equal to 6 or something.

🤔 I wonder if binary search idea works for views? Not every log line has a view, and view numbers are not monotonic. But I think "the first line where view=x is printed" sequence is actually monotonic?

[matklad]

UI-wise, I'd expect this to look like

position = "bisect view=6"

position = "bisect 2024-02-12T13:32:54"

where bisect is a keyword, and to the right is a pattern, where the tool should automatically extract either a single numeric field, or a single date-time filed.

[RMGiroux]

Not every log line has a view, and view numbers are not monotonic. But I think "the first line where view=x is printed" sequence is actually monotonic?

I think it's hard to do a sorted seek on a sequence that isn't monotonic. Of course, if you KNOW you're going to be keying on something like view, you could run a "tail -f" like thread where you read the lines as they're logged and then log offsets as key fields first appear, then do a binary search on that if there's more than a trivial number of results, but that sounds like a major complication.