Breaking changes in 3.X

[sumneko]

Introduce

I'm working on the 3.X version, the purpose is to add a more complete type system, and use this to implement features such as type checking. In previous versions I tried to keep the type system compatible with EmmyLua, but I will make some breaking changes after version 3.X, so the type system may not be suitable to continue to be called EmmyLua in the future, but I haven't thought about it yet, ~I will call it LuaDoc for now~.

Breaking changes

Always trace local set

local x
x = 1
x = true
print(x) --> x is `integer|boolean`

---@type string
local x
x = 1
x = true
print(x) --> x is `string`

local x = 'str'
x = 1
x = true
print(x) --> x is `string`

Dose not trace field inject

local x = {}
local y = x
y.field = 1

print(x.field) --> undefined field `field`

use ---@class instead

---@class MyClass
local x = {}
---@class MyClass
local y = x
y.field = 1

print(x.field) --> `x.field` is `1`

Finding definition does not recurse

local t = {}
t.x = 1
t.y = t.x

print(t.y) --> t.y is `1`

Find definition of t.y will not find out t.x

---@class C
local mt = {}

---@type C
local o

Find definition of o will not find out mt

Type unknown

local t = {}
print(t.x) --> t.x is `unknown` instead of `any`

Types can be used in doc-table keys

---@type { [number]: string }
local t

print(t[1]) --> t[1] is `string`
print(t.x) --> t.x is `unknown`

boolean type true and false

You can use ---@type { [string]: true } now.

class and alias support generic inheritance

---@class table<K, V>: { [K]: V } --> this is builtin

---@alias mark<K> { [K]: true } 

~table<integer, string> can not convert to string[]~

Default value of unrary and binary

---@type unknown
local x

local y = - x --> y is `number` instead of `unknown`

unknown can form union types with other types

function f()
    if x then
        return y
    else
        return nil
    end
end

local n = f() --> n is `unknown|nil`

Integer type

---@type 1|2|3|4|5
local v

doc-enum must be a string

---@param x 'aaa'|'bbb'
function f(x)
    print(x) --> x is `"aaa"` or `"bbb"`
end

This means you can no longer use ---@param x '1024'|'true'|'function () end' to represent a specific number, boolean or function. Use ---@param x 1024|true|fun() instead. For compatibility, I will continue to recognize '"xxx"' as the string xxx instead of "xxx".

integer is a class in Lua 5.1 and Lua 5.2

---@type integer
local v

print(v) --> v is `integer` instead of `number`

Although Lua 5.1 and Lua 5.2 runtimes do not have integer, many API parameters require a number that can be converted to an integer, so adding integer to the type system can help you handle this part.

Not supported for now

search metatable in this case

setmetatable(object, { __index = {
    default = 1
} })
print(object.default) --> undefined field `default`

This turns the reference into a definition, and checking the reference in order to search for a definition is too expensive.

Please use ---@class or use this:

object = setmetatable(object, { __index = {
    default = 1
} })
print(object.default) --> this is OK

Maybe can be resolved by indexing setmetable

[XeroOl]

I think LuaDoc name is taken by https://keplerproject.github.io/luadoc/

[sumneko]

I think LuaDoc name is taken by https://keplerproject.github.io/luadoc/

Maybe I can continue to call it Emmylua, since the original author is not active, then I can grab it 🙃.

[sumneko]

I have merged a 3.0 branch to Master, if you find a major problem, you can return to the label 2.6.8

[faimin]

The CPU usage of the latest plugin version (v3.0.0) in VSCode is full.

新版本插件（v3.0.0）在VSCode 中CPU爆满

ENV： 
MacOS 12.3.1
VSCode  1.66.1
Lua 5.4

[Silent-zzz]

多个文件声明类的方法该怎么写呢

[Silent-zzz]

例如

[sumneko]

The CPU usage of the latest plugin version (v3.0.0) in VSCode is full.

新版本插件（v3.0.0）在VSCode 中CPU爆满

ENV： 
MacOS 12.3.1
VSCode  1.66.1
Lua 5.4

请单独开个issue并提供日志

[sumneko]

多个文件声明类的方法该怎么写呢

你可以多次使用 ---@class unit

[firas-assaad]

Thanks for the great server/extension!

With the latest version of the VS Code extension (3.0.0) I get a lot of undefined field errors in classes. e.g.

I actually use middleclass for classes, but that used to work fine in previous versions. Is this the same unsupported case as "find definition in methods"?

As I use classes heavily, this is raising a lot of warnings. Is there a workaround or do you recommend that I revert to a previous 2.X version?

[sumneko]

Thanks for the great server/extension!

With the latest version of the VS Code extension (3.0.0) I get a lot of undefined field errors in classes. e.g.

I actually use middleclass for classes, but that used to work fine in previous versions. Is this the same unsupported case as "find definition in methods"?

As I use classes heavily, this is raising a lot of warnings. Is there a workaround or do you recommend that I revert to a previous 2.X version?

You can choose:

1. Use ---@field field boolean under ---@class x
2. Revert to 2.X
3. Waiting for supporting this case, maybe next week

[sumneko]

function mt:init()
    self.xxx = 1
end

Supported finding definition of self.xxx in version 3.1.0

[Arxareon]

I think this issue we found all links to here: #1158

I'd very much like to have the old functionality from 2.6.8 back. There are fringe cases in WoW addon development (addon namespace tables as an example) where field references being pulled through tables and subtables are very much needed. I tried adding @class and @field definitions to rectify the issue, but I wasn't able to fix annotations and types in all cases. Also, I wouldn't like adding more unnecessary lines to the already long files just to have annotations working between all files of a project.

See the above mentioned issue and further quoted issue from Ketho's WoW API extension repo (also mentioned within the above mentioned issue).

Thank you for working on this! :) If you need details on how we use it specifically for WoW or have tips for us to change our workflow, I'd be happy to get back to you on that.

[SalavatR]

Thanks for the great server/extension! With the latest version of the VS Code extension (3.0.0) I get a lot of undefined field errors in classes. e.g. I actually use middleclass for classes, but that used to work fine in previous versions. Is this the same unsupported case as "find definition in methods"? As I use classes heavily, this is raising a lot of warnings. Is there a workaround or do you recommend that I revert to a previous 2.X version?

You can choose:

1. Use ---@field field boolean under ---@class x
2. Revert to 2.X
3. Waiting for supporting this case, maybe next week

Hello! is any update for this issue?

---@class my_window

---@param self my_window
local function lcf1(self)
   local a = self.variable ----- Undefined field `variable`.Lua Diagnostics.(undefined-field)
end

---@param self my_window
function init(self)
   self.variable = true
   lcf1(self)
end

in some cases we have more about 100 variables in self, and it is not comfortable to declare all as ---@field

[Arxareon]

@SalavatR I've reverted to using version 2.6.8 until this gets resolved without needing to use excessive documentation for references to be recognized. It works fine for me even though I wish I had some of the new features available as well (like the recognition of aliases - but I can always collapse them manually now) - so unless you need something new, I suggest reverting the extension back to 2.6.8 for the time being (not sure if sumneko is planning to revert any of these changes in one way or another, so it might be indefinitely).

[sumneko]

in some cases we have more about 100 variables in self, and it is not comfortable to declare all as ---@field

You can use

---@param self my_window
function init(self)
    ---@class my_window
    self = self
    self.variable = true
    lcf1(self)
end

This is to prevent fields from being declared unexpectedly.

[SalavatR]

in some cases we have more about 100 variables in self, and it is not comfortable to declare all as ---@field

You can use

---@param self my_window
function init(self)
    ---@class my_window
    self = self
    self.variable = true
    lcf1(self)
end

This is to prevent fields from being declared unexpectedly.

Thanks! What about performance of this? And will LLS analize all of usings of param?

[sumneko]

@SalavatR LLS only analize fields in ---@class, this is both strict and performance oriented.

[Iyadriel]

@SalavatR I've reverted to using version 2.6.8 until this gets resolved without needing to use excessive documentation for references to be recognized. It works fine for me even though I wish I had some of the new features available as well (like the recognition of aliases - but I can always collapse them manually now) - so unless you need something new, I suggest reverting the extension back to 2.6.8 for the time being (not sure if sumneko is planning to revert any of these changes in one way or another, so it might be indefinitely).

I was able to make references across namespaces work in the latest version by simply using the following syntax in every file:

---@class ns
local ns = select(2, ...)

Does this work for you?

Edit -- Never mind, this only seems to work for a depth of one level.

[bandaloo]

@SalavatR I've reverted to using version 2.6.8 until this gets resolved without needing to use excessive documentation for references to be recognized. It works fine for me even though I wish I had some of the new features available as well (like the recognition of aliases - but I can always collapse them manually now) - so unless you need something new, I suggest reverting the extension back to 2.6.8 for the time being (not sure if sumneko is planning to revert any of these changes in one way or another, so it might be indefinitely).

I was able to make references across namespaces work in the latest version by simply using the following syntax in every file:

---@class ns
local ns = select(2, ...)

Does this work for you?

Edit -- Never mind, this only seems to work for a depth of one level.

it would be really nice to have a @namespace instead of having to use @class for simple modules. it is a little funky to mark a table as @class when you're not interested in using it as a type or instantiating it, and it makes no reference to self. if you don't use @class, it's easy to make a typo and access a field that doesn't exist; it will be treated as unknown. @class does the job though! it's just slightly more feature-rich than what i want most of the time.