Alias doesn't handle # character in string

[hraytilakhealthcare]

How are you using the lua-language-server?

NeoVim

Which OS are you using?

Linux

What is the issue affecting?

Diagnostics/Syntax Checking

Expected Behaviour

Enum members can contain # character

Actual Behaviour

The LSP diagnostic reports an error on using it

Reproduction steps

1. Create an enum alias with members containing #
   Given the following alias

---| "/vfx#impact_fx_creator"
---| "/sfx"
---| "/player#player_upgrader"
---| "/cauli"
---| "/game_over_ui"
---| "/in_game_ui"

2. Notice that the type check considers the right-hand of # as a comment even though it's in the string

---@type component_url
local d =  "/vfx#impact_fx_creator"

4. See error

Diagnostics:
1. Cannot assign `string` to parameter `"/cauli"|"/game_over_ui"|"/in_game_ui"|"/player#"|"/sfx"...(+1)`.
   - `string` cannot match `"/cauli"|"/game_over_ui"|"/in_game_ui"|"/player#"|"/sfx"...(+1)`

Additional Notes

No response

Log File

No response

[tomlau10]

Root cause

After some debugging, this is related to how a multi-line alias annotation is parsed:

lua-language-server/script/parser/luadoc.lua

Lines 941 to 953 in 3a39d06

	local resumeHead = nextComm.text:match '^%-%s*%|'
	if resumeHead then
	NextComment(i)
	row = row + i + 1
	local finishPos = nextComm.text:find('#', #resumeHead + 1) or #nextComm.text
	parseTokens(nextComm.text:sub(#resumeHead + 1, finishPos), nextComm.start + #resumeHead + 1)
	local resume = parseResume(result)
	if resume then
	if comments then
	resume.comment = table.concat(comments, '\n')
	else
	resume.comment = nextComm.text:match('%s*#?%s*(.+)', resume.finish - nextComm.start)
	end

• in a multi-line alias, there is a practice that user can specify a description for an enum separated by a #, just like the example from official doc:
  https://luals.github.io/wiki/annotations/#alias

---@alias DeviceSide
---| '"left"' # The left side of the device
---| '"right"' # The right side of the device
---| '"top"' # The top side of the device
---| '"bottom"' # The bottom side of the device
---| '"front"' # The front side of the device
---| '"back"' # The back side of the device

• in the above parsing logic, it simply finds the position of a # to determine the start of the description part, which is the root cause

Observation

But actually I don't quite understand the need of this logic (finding the position of #) 😕
Because a description doesn't need to be preceded by a #.
The following will work too:

---@alias DeviceSide
---| '"left"' The left side of the device
---| '"right"' The right side of the device
---| '"top"' The top side of the device
---| '"bottom"' The bottom side of the device
---| '"front"' The front side of the device
---| '"back"' The back side of the device

Possible fix

I just remove the find('#') logic in line 945, and let parseTokens() to always be parse the whole line, seems can fix this issue 🤔

-    local finishPos = nextComm.text:find('#', #resumeHead + 1) or #nextComm.text
-    parseTokens(nextComm.text:sub(#resumeHead + 1, finishPos), nextComm.start + #resumeHead + 1)
+    parseTokens(nextComm.text:sub(#resumeHead + 1), nextComm.start + #resumeHead + 1)

• let the parser always parse the whole line (without the --| head of course)
• then the later logic at line 952 can still find a possible description starting from any unparsed resume.finish offset
• works for both descriptions with or without preceding #
  

---

I didn't dive deep to test it. Maybe you or others would like to finish it from there and open PR 😄