change diff update to whole buffer

[glepnir]

diff update has bug .back to update whole buffer

[xiaoshihou514]

Eliminate update bugs caused by calculating the buffer area

Is apply_text_edit thread safe or sth?

[glepnir]

nope. just a try. diff update has bug. now back to update whole buffer if has result .

[xiaoshihou514]

• 1. Any idea why🤔?
• 2. lsp.util.apply_text_edit doesn't fix this?
• 3. How can I repro this? I would like to help

[glepnir]

diff return data like 240, 0, 241, 1 of start_a, start_a, start_b, count_b Didn't understand the rules. Causes buffer confusion after update. So first go back and update the entire buffer. I think a benchmark is needed here. Because we need to know whether using diff is worth it. The time to update the entire buffer and the time to run the diff algorithm. It should be possible to benchmark large files

[glepnir]

not sure. I remember this code will cause a format bug.

local t = {}

function t.a()
end

function t.b()
end

return t

[xiaoshihou514]

Some limited benchmarking:
https://github.com/xiaoshihou514/guard-bench
Tested by manually toggling update methods and nvim -u init.lua

• All files have the spaces changed to 8 spaces, in this case diffing is about 25% percent slower on average
  (diff 8ms, no diff 6ms with big.js)

• All files have " = " changed to "=", in this case diffing does better, but still does worse than not diffing
  (diff 7ms, no diff 6ms with big.js)

Someone may be interested to do more rigorous and automated benchmarking to find the overall balance, but at this point I think there is really no point perfing the update method. big.js is 10k LOC, which is about what you can get when writing serious code, but updating buffer only takes less than 10ms, and the difference made by diffing is even smaller, I argue this is negligible compared to the heavy lifting done by the formatter itself. Under all circumstances, diffing or not does not really change the user experience in terms of formatting speed. And if diffing the buffer contents causes bug from time to time, I would call for its removal.

edit: not sure about mem usage tho

[xiaoshihou514]

@glepnir mergable yet?