Generalise julia-indent-hanging to cover more cases

[ajhalme]

Having looked at the indentation code for a bit, I decided to tackle issue #155 just to learn more about this stuff.

The reported issue is concerned with indentation in situations where an operator is the last token on the line.

In ESS, R code is formatted like so:

# ESS 18.10.2

sample1 <- left_join(df1, df2, by = c("longname1", "longname2", "longname3", "longname4")) %>%
    do_stuff

sample2 <-left_join(df1, df2, by = c("longname1", "longname2",
                                     "longname3", "longname4",
                                     "longname5", "longname6")) %>%
    do_stuff

sample3 <-
    left_join(df1, df2, by = c("longname1",
                               "longname2", "longname3", "longname4")) %>%
    do_stuff

sample4 <-
    left_join(df1,
              df2, by = c("longname1", "longname2",
                          "longname3", "longname4")) %>%
    do_stuff

Similar code in Julia is currently formatted like so:

sample1 = @pipe leftjoin(df1, df2, on = [:longname1, :longname2, :longname3, :longname4]) |>
    do_stuff

sample2 = @pipe leftjoin(df1, df2, on = [:longname1, :longname2,
                                         :longname3, :longname4,
                                         :longname5, :longname6]) |>
                                             do_stuff

sample3 =
    @pipe leftjoin(df1, df2, on = [:longname1,
                                   :longname2, :longname3, :longname4]) |>
                                       do_stuff

sample4 =
    @pipe leftjoin(df1,
                   df2, on = [:longname1, :longname2,
                              :longname3, :longname4]) |>
                                  do_stuff

In other words, in some cases featuring the hanging operator -- here the pipe, |> -- julia-mode is not indenting the subsequent line as neatly as ESS indents R.

This PR generalises the existing hanging operator indentation routine, julia-indent-hanging, by iterating over the hanging line scanner and by introducing an additional check for detecting split parentheses. (This is powered by the existing julia-parent-indent function.)

This fixes indentation for cases where:

1. two hanging operators are NOT on adjacent lines, OR

2. a multi-line parenthesis expression is followed by a hanging operator.

With this PR, the above examples indent as below, which is arguably nicer. Other indentation should not be affected.

sample1 = @pipe leftjoin(df1, df2, on = [:longname1, :longname2, :longname3, :longname4]) |>
    do_stuff

sample2 = @pipe leftjoin(df1, df2, on = [:longname1, :longname2,
                                         :longname3, :longname4,
                                         :longname5, :longname6]) |>
    do_stuff

sample3 =
    @pipe leftjoin(df1, df2, on = [:longname1,
                                   :longname2, :longname3, :longname4]) |>
    do_stuff

sample4 =
    @pipe leftjoin(df1,
                   df2, on = [:longname1, :longname2,
                              :longname3, :longname4]) |>
    do_stuff

[tpapp]

Thank you very much for doing this, I made a minor suggestion but this is already very well written.

[tpapp]

When (> (julia-prev-line-skip-blank-or-comment) 0) is nil, I think this just returns best-indent which is nil. Would it make sense to just bail early, eg

(when (> (julia-prev-line-skip-blank-or-comment) 0)
  (save-excursion
    (let* ((prev-indent (current-indentation))
           (prev-paren-indent (julia-paren-indent))
           (hanging-match (julia--hanging-operator-p))
           (double-hanging)
           (best-indent))
      ;; Base case: one preceding hanging operator => increase indent from preceding line
      (if hanging-match
          (setq best-indent (+ (current-indentation) julia-indent-offset)))
      ;; Loop rows back until we run out of indentation or hit a second hanging indent
      (while (and hanging-match
                  (not double-hanging)
                  (> (current-indentation) 0)
                  (> (julia-prev-line-skip-blank-or-comment) 0))
        (setq double-hanging (julia--hanging-operator-p))
        ;; Special case: two hanging operator indents already => use the first one
        (when double-hanging
          (setq best-indent prev-indent))
        ;; Special case: open paren followed by one hanging operator
        (when (and (not double-hanging) prev-paren-indent)
          (setq best-indent (+ (current-indentation) julia-indent-offset)))
        (setq prev-indent (current-indentation))
        (setq prev-paren-indent (julia-paren-indent)))
      best-indent)))

(note: did not test this, and this is just a style suggestion, feel free to ignore)

[ajhalme]

Yeah, makes sense.

(when (> (julia-prev-line-skip-blank-or-comment) 0)
  (save-excursion

has to be

  (save-excursion
     (when (> (julia-prev-line-skip-blank-or-comment) 0)

or calling the indentation moves (point) back by one row. Somehow it also seemed to get caught in an infinite loop or something.

Swapping the two rows works fine, tests pass. I've pushed a patch with these changes.

[tpapp]

I would appreciate if someone else could please review this, but I pinged quite a few people yesterday so I will wait a bit.

[tpapp]

@ajhalme: why did you close this?