Improve analysis of files with consistent indentation where offset is guessed wrongly due to merging about dtrt-indent HOT 2 OPEN

This is dtrt-indent working as intended (and the diagnosis tells you what's going on). It can't distinguish between the case where there are a few odd 2-indents somewhere in the middle of a long file (where an indent of 4 is what you really want), and the case where there are a few symmetric ones at each end (as here).

I thought this issue seemed familiar, and indeed, see #29. I'd rather not make dtrt-indent cleverer. On the other hand, decreasing the value of dtrt-indent-max-merge-deviation only works up to a certain point: for arbitrarily long JSON files you'd have to set it to zero.

I can't think of a simple analytical solution to this problem in general that doesn't involve understanding the syntax of the file.

One way to do it by trial and error: run indent-buffer with the first-guessed setting. If the buffer doesn't change, then keep the setting. This could be done in dtrt-indent-analyze when considering whether to merge two settings.

from dtrt-indent.

Thanks for the explanation, @rrthomas . That makes sense, and it seemed like a maybe-expected result of the design, but it also surprised me, hence my filing the issue.

I didn't find dtrt-indent-max-merge-deviation when I skimmed for something I could tweak that might help, so thanks, that looks useful.

I don't see a robust, correct analytical answer to the problem either.

Maybe an (opt-in?) dumb heuristic could be good enough. Before merging a lower indent level into a compatible higher indent level, check whether the file's first and last indents match the lower level?

If they do, on average, I bet the lower indent level would be correct.

Your trial-and-error suggestion also makes sense.

from dtrt-indent.