Add lexer.rbs #33
Add lexer.rbs #33
Conversation
| @@ -18,8 +20,8 @@ def to_s | |||
| "#{super} line: #{line}, column: #{column}" | |||
| end | |||
|
|
|||
| @i = 0 | |||
| @types = [] | |||
| instance_variable_set :@i, 0 | |||
There was a problem hiding this comment.
[note] steep can not detect a type of instance variable for original codes, even so rbs file has declarations for them. It might the limitation for Struct with block (?).
lib/lrama/lexer.rb:23:6: [error] Cannot find the declaration of instance variable: `@i`
│ Diagnostic ID: Ruby::UnknownInstanceVariable
│
└ @i = 0
~~
lib/lrama/lexer.rb:24:6: [error] Cannot find the declaration of instance variable: `@types`
│ Diagnostic ID: Ruby::UnknownInstanceVariable
│
└ @types = []
~~~~~~
Detected 2 problems from 1 file
| @@ -62,8 +64,7 @@ def self.define_type(name) | |||
| GrammarRules = 3 | |||
| Epilogue = 4 | |||
|
|
|||
| # Token types | |||
|
|
|||
| # @dynamic prologue, bison_declarations, grammar_rules, epilogue, bison_declarations_tokens, grammar_rules_tokens | |||
There was a problem hiding this comment.
In my environment bundle exec steep check passes without this annotation. Is there any benefit for these annotation?
| Token = Struct.new(:type, :s_value, keyword_init: true) do | ||
| Type = Struct.new(:id, :name, keyword_init: true) | ||
| Token = _ = Struct.new(:type, :s_value, keyword_init: true) do | ||
| # @implements Token[SValue] |
There was a problem hiding this comment.
[note] This is an annotation defined in steep not rbs.
| @@ -7,8 +7,10 @@ class Lexer | |||
| include Lrama::Report::Duration | |||
|
|
|||
| # s_value is semantic value | |||
| Token = Struct.new(:type, :s_value, keyword_init: true) do | |||
| Type = Struct.new(:id, :name, keyword_init: true) | |||
| Token = _ = Struct.new(:type, :s_value, keyword_init: true) do | |||
There was a problem hiding this comment.
[note] Inserting _ between a constant and Struct.new is a common workaround when we use Struct to define new class.
| type line_data = [String, Integer] | ||
| type reference = [::Symbol, Integer | String, Token[untyped]?, Integer, Integer] | ||
|
|
||
| class Type[SValue] < Struct[untyped] |
There was a problem hiding this comment.
Do we need SValue type variable? Always Type#id is Interger and Type#name is String.
>> Lrama::Lexer::Token::P_expect
=> #<struct Lrama::Lexer::Type id=0, name="P_expect">
>> Lrama::Lexer::Token::Number
=> #<struct Lrama::Lexer::Type id=15, name="Number">
On the other hand. Token needs type variable because type of s_value depends on the value of Type.
|
| def lex_common: (Array[line_data] lines, Array[Token[untyped]] tokens) -> void | ||
| def lex_bison_declarations_tokens: -> void | ||
| def lex_user_code: (StringScanner, Integer line, Integer column, Array[line_data] lines) -> [Token[String], Integer] | ||
| def lex_string: (StringScanner, String terminator, Integer line, Array[line_data] lines) -> line_data |
There was a problem hiding this comment.
Surely, the type of returned value is [String, Integer]. But this string is a part of line so I'm wondering which is better to use line_data type or [String, Integer]. Do you have any opinion?
Added lexer.rbs.
Discussions:
TypeandTokenhave type parameters namedSValue. I expect them to work well in parser.rb.create_token(Token::Number, str, ...)untypable.patch.rbs, which patches the types ofStringScanner#[]andStringScanner#getch.StringScanner#fetchandStringScanner#getch!, as soutaro remarked.ss[n]with(ss[n] || raise)and so on, but that would be hard to read.Array#[]andArray#firstbecauseArrayis used widely, and such patches may affect other files. Instead, I replacedarray[0...l]witharray.take(l)andarray.firstwitharray.fetch(0).