vimwiki/doc/design_notes.md

Design Notes

This file is meant to document design decisions and algorithms inside vimwiki which are too large for code comments, and not necessarily interesting to users. Please create a new section to document each behavior.

Formatting tables

In vimwiki, formatting tables occurs dynamically, when navigating between cells and adding new rows in a table in the Insert mode, or statically, when pressing gqq or gqw (which are mappings for commands VimwikiTableAlignQ and VimwikiTableAlignW respectively) in the Normal mode. It also triggers when leaving Insert mode, provided variable g:vimwiki_table_auto_fmt is set. In this section, the original and the newer optimized algorithms of table formatting will be described and compared.

The older table formatting algorithm and why this is not optimal

Let's consider a simple example. Open a new file, say tmp.wiki, and create a new table with command VimwikiTable. This should create a blank table.

|   |   |   |   |   |
|---|---|---|---|---|
|   |   |   |   |   |

Let's put the cursor in the first header column of the table, enter the Insert mode and type a name, say Col1. Then press Tab: the cursor will move to the second column of the header and the table will get aligned (in the context of the table formatting story, words aligned and formatted are considered as synonyms). Now the table looks as in the following snippet.

| Col1 |   |   |   |   |
|------|---|---|---|---|
|      |   |   |   |   |

Then, when moving cursor to the first data row (i.e. to the third line of the table below the separator line) and typing anything here and there while navigating using Tab or Enter (pressing this creates a new row below the current row), the table shall keep formatting. Below is a result of such a random edit.

| Col1 |       |   |       |          |
|------|-------|---|-------|----------|
|      | Data1 |   | Data2 |          |
|      |       |   |       | New data |

The lowest row gets aligned when leaving the Insert mode. Let's copy Data1 (using viwy or another keystroke) and paste it (using p) in the second data row of the first column. Now the table looks mis-aligned (as we did not enter the Insert mode).

| Col1 |       |   |       |          |
|------|-------|---|-------|----------|
|      | Data1 |   | Data2 |          |
| Data1     |       |   |       | New data |

This is not a big problem though, because we can put the cursor at any place in the table and press gqq: the table will get aligned.

| Col1  |       |   |       |          |
|-------|-------|---|-------|----------|
|       | Data1 |   | Data2 |          |
| Data1 |       |   |       | New data |

Now let's make real problems! Move the cursor to the lowest row and copy it with yy. Then 500-fold paste it with 500p. Now the table very long. Move the cursor to the lowest row (by pressing G), enter the Insert mode, and try a new random editing session by typing anything in cells with Tab and Enter navigation interleaves. The editing got painfully slow, did not?

The reason of the slowing down is the older table formatting algorithm. Every time Tab or Enter get pressed down, all rows in the table get visited to calculate a new alignment. Moreover, by design it may happen even more than once per one press!

function! s:kbd_create_new_row(cols, goto_first)
  let cmd = "\<ESC>o".s:create_empty_row(a:cols)
  let cmd .= "\<ESC>:call vimwiki#tbl#format(line('.'))\<CR>"
  let cmd .= "\<ESC>0"
  if a:goto_first
    let cmd .= ":call search('\\(".s:rxSep()."\\)\\zs', 'c', line('.'))\<CR>"
  else
    let cmd .= (col('.')-1)."l"
    let cmd .= ":call search('\\(".s:rxSep()."\\)\\zs', 'bc', line('.'))\<CR>"
  endif
  let cmd .= "a"

  return cmd
endfunction

Function s:kbd_create_new_row() is called when Tab or Enter get pressed. Formatting of the whole table happens in function vimwiki#tbl#format(). But remember that leaving the Insert mode triggers re-formatting of a table when variable g:vimwiki_table_auto_fmt is set. This means that formatting of the whole table is called on all those multiple interleaves between the Insert and the Normal mode in s:kbd_create_new_row (notice \<ESC>, o, etc.).

The newer table formating algorithm

The newer algorithm was introduced to struggle against performance issues when formatting large tables.

Let's take the table from the previous example in an intermediate state.

| Col1 |       |   |       |          |
|------|-------|---|-------|----------|
|      | Data1 |   | Data2 |          |
| Data1     |       |   |       | New data |

Then move the cursor to the first data row, copy it with yy, go down to the mis-aligned line, and press 5p. Now we have a slightly bigger mis-aligned table.

| Col1 |       |   |       |          |
|------|-------|---|-------|----------|
|      | Data1 |   | Data2 |          |
| Data1     |       |   |       | New data |
|      | Data1 |   | Data2 |          |
|      | Data1 |   | Data2 |          |
|      | Data1 |   | Data2 |          |
|      | Data1 |   | Data2 |          |
|      | Data1 |   | Data2 |          |

Go down to the lowest, the 7th, data row and press gq1. Nothing happened. Let's go to the second or the third data row and press gq1 once again. Now the table gets aligned. Let's undo formatting with u, go to the fourth row, and press gq1. Now the table should look like in the following snippet.

| Col1 |       |   |       |          |
|------|-------|---|-------|----------|
|      | Data1 |   | Data2 |          |
| Data1     |       |   |       | New data |
|           | Data1 |   | Data2 |          |
|           | Data1 |   | Data2 |          |
|      | Data1 |   | Data2 |          |
|      | Data1 |   | Data2 |          |
|      | Data1 |   | Data2 |          |

What a peculiar command! Does using it make any sense? Not much, honestly. Except it shows how the newer optimized table formatting algorithm works in the Insert mode.

Indeed, the newer table formatting algorithm introduces a viewport on a table. Now, when pressing Tab or Enter in the Insert mode, only a small part of rows are checked for possible formatting: two rows above the current line and the current line itself (the latter gets preliminary shrunk with function s:fmt_row()). If all three lines in the viewport are of the same length, then nothing happens (case 1 in the example). If the second or the shrunk current line is longer then the topmost line in the viewport, then the algorithm falls back to the older formatting algorithm and the whole table gets aligned (case 2). If the topmost line in the viewport is longer than the second and the shrunk current line, then the two lowest lines get aligned according to the topmost line (case 3).

Performance of the newer formatting algorithm should not depend on the height of the table. The newer algorithm should also be consistent with respect to user editing experience. Indeed, as soon as a table should normally be edited row by row from the top to the bottom, dynamic formatting should be both fast (watching only three rows in a table, re-formatting only when the shrunk current row gets longer than any of the two rows above) and eager (a table should look formatted on every press on Tab and Enter). However, the newer algorithm differs from the older algorithm when starting editing a mis-aligned table in an area where mis-aligned rows do not get into the viewport: in this case the newer algorithm will format the table partly (in the rows of the viewport) until one of the being edited cells grows in length to a value big enough to trigger the older algorithm and the whole table gets aligned. When partial formatting is not desirable, the whole table can be formatted by pressing gqq in the Normal mode.