[Erlang 0113] Elixir 编译流程梳理

注意:目前Elixir版本还不稳定,代码调整较大,本文随时失效

     之前简单演示过如何从elixir ex代码生成并运行Erlang代码,下面仔细梳理一遍elixir文件的编译过程,书接上文,从elixir的代码切入,这一次我们主要关注编译流程,一些细节暂时不展开.

 

 

 

-module(elixir).
......
start_cli() ->
  application:start(?MODULE),
  %% start_cli() --> ["+compile","m.ex"]
  'Elixir.Kernel.CLI':main(init:get_plain_arguments()).

　　

   'Elixir.Kernel.CLI'的代码没有找到?它不是erlang代码不在/lib/elixir/src目录下,它是mx代码,位置在:https://github.com/elixir-lang/elixir/blob/master/lib/elixir/lib/kernel/cli.ex 经过一系列参数读取解析之后,最终我们可以定位到执行compile的代码位置,如下:

 

 

defmodule Kernel.CLI do
......
  if files != [] do
      wrapper fn ->
        Code.compiler_options(config.compiler_options)
        Kernel.ParallelCompiler.files_to_path(files, config.output,
          each_file: fn file -> if config.verbose_compile do IO.puts "Compiled #{file}" end end)
      end
    else
      { :error, "--compile : No files matched patterns #{Enum.join(patterns, ",")}" }
    end

　

    这里调用了同一目录下面的Kernel.ParallelCompiler的方法,注意由于Elixir对于代码模块名和文件名没有一致性的要求,找代码的时候要注意一点;对应的ex代码模块是parallel_compile,代码路径在: https://github.com/elixir-lang/elixir/blob/master/lib/elixir/lib/kernel/parallel_compiler.ex 从这里开始,执行逻辑就回到了Erlang代码,:elixir_compiler.file(h)等价的Erlang调用代码是elixir_compile:file(h).

 

defmodule Kernel.ParallelCompiler do
.....
  try do
        if output do
          :elixir_compiler.file_to_path(h, output)
        else
          :elixir_compiler.file(h)
        end
        parent <- { :compiled, self(), h }
      catch
        kind, reason ->
          parent <- { :failure, self(), kind, reason, System.stacktrace }
      end

  elixir_compiler 完成路径解析之后,最终调用了string(Contents, File)方法.我们将string(Contents, File)方法逐步拆解开,elixir_translator:'forms!'(Contents, 1, File, [])首先通过elixir_tokenizer:tokenize 将代码文本解析成为Tokens,然后通过 Forms = elixir_parser:parse(Tokens) 解析成为Elixir AST,注意这里还不是Erlang Abstract Format,这里的Forms我们输出一下看看:

 

 

-module(elixir_compiler).

file(Relative) when is_binary(Relative) ->
  File = filename:absname(Relative),
  { ok, Bin } = file:read_file(File),
  string(elixir_utils:characters_to_list(Bin), File).

string(Contents, File) when is_list(Contents), is_binary(File) ->
  Forms = elixir_translator:'forms!'(Contents, 1, File, []),
  quoted(Forms, File).

　　

 

[{defmodule,
   [{line,1}],
   [{'__aliases__',[{line,1}],['Math']},
    [{do,
      {def,[{line,2}],[{sum,[{line,2}],[{a,[{line,2}],nil},{b,[{line,2}],nil}]},
        [{do,{'__block__',[],[
         {'=',[{line,3}],[{a,[{line,3}],nil},123]},
         {'=',[{line,4}],[{a,[{line,4}],nil},2365]},
         {'+',[{line,5}],[
          {a,[{line,5}],nil},{b,[{line,5}],nil}]}
          ]}}]]}}]]}]

　

      Elixir AST在quoted(Forms, File)函数完成到Erlang Abstract Format Forms的转换,下面我们跟进quoted(Forms,File)方法,在最近的版本中lexical_tracker加入了Scope中;Scope维护了代码的上下文信息,我们暂时有这样一个印象即可,我们先把流程走完,下面就要深入eval_forms(Forms, Line, Vars, S)方法了.

 

附Scpoe定义 https://github.com/elixir-lang/elixir/blob/master/lib/elixir/include/elixir.hrl

-record(elixir_scope, {
  context=nil,             %% can be assign, guards or nil
  extra=nil,               %% extra information about the context, like fn_match for fns
  noname=false,            %% when true, don't add new names (used by try)
  super=false,             %% when true, it means super was invoked
  caller=false,            %% when true, it means caller was invoked
  module=nil,              %% the current module
  function=nil,            %% the current function
  vars=[],                 %% a dict of defined variables and their alias
  backup_vars=nil,         %% a copy of vars to be used on ^var
  temp_vars=nil,           %% a set of all variables defined in a particular assign
  clause_vars=nil,         %% a dict of all variables defined in a particular clause
  extra_guards=nil,        %% extra guards from args expansion
  counter=[],              %% a counter for the variables defined
  local=nil,               %% the scope to evaluate local functions against
  context_modules=[],      %% modules defined in the current context
  macro_aliases=[],        %% keep aliases defined inside a macro
  macro_counter=0,         %% macros expansions counter
  lexical_tracker=nil,     %% holds the lexical tracker pid
  aliases,                 %% an orddict with aliases by new -> old names
  file,                    %% the current scope filename
  requires,                %% a set with modules required
  macros,                  %% a list with macros imported from module
  functions                %% a list with functions imported from module
}).

　　

quoted方法最近的变化是使用elixir_lexical:run包装了一下,之前的版本简单直接,可以先看一下:

quoted(Forms, File) when is_binary(File) ->
  Previous = get(elixir_compiled),
% M:elixir_compiler Previous undefined
  try
    put(elixir_compiled, []),
    eval_forms(Forms, 1, [], elixir:scope_for_eval([{file,File}])),
    lists:reverse(get(elixir_compiled))
  after
    put(elixir_compiled, Previous)
  end.

　　现在quoted是这样的:

quoted(Forms, File) when is_binary(File) ->
  Previous = get(elixir_compiled),
  try
    put(elixir_compiled, []),
    elixir_lexical:run(File, fun
      (Pid) ->
        Scope = elixir:scope_for_eval([{file,File}]),
        eval_forms(Forms, 1, [], Scope#elixir_scope{lexical_tracker=Pid})
    end),
    lists:reverse(get(elixir_compiled))
  after
    put(elixir_compiled, Previous)
  end.

quoted方法里面我们需要重点关注的是eval_forms方法,在这个方法里面完成了Elixir AST到Erlang AST转换,Elixir表达式通过 elixir_translator:translate被翻译成对应的Erlang Abstract Format.之后eval_mod(Fun, Exprs, Line, File, Module, Vars)完成对表达式和代码其它部分(比如attribute,等等)进行组合.

eval_forms(Forms, Line, Vars, S) ->
  { Module, I } = retrieve_module_name(),
  { Exprs, FS } = elixir_translator:translate(Forms, S),

  Fun  = eval_fun(S#elixir_scope.module),
  Form = eval_mod(Fun, Exprs, Line, S#elixir_scope.file, Module, Vars),
  Args = list_to_tuple([V || { _, V } <- Vars]),

  %% Pass { native, false } to speed up bootstrap
  %% process when native is set to true
  { module(Form, S#elixir_scope.file, [{native,false}], true,
    fun(_, Binary) ->
    Res = Module:Fun(Args),
    code:delete(Module),
    %% If we have labeled locals, anonymous functions
    %% were created and therefore we cannot ditch the
    %% module
    case beam_lib:chunks(Binary, [labeled_locals]) of
      { ok, { _, [{ labeled_locals, []}] } } ->
        code:purge(Module),
        return_module_name(I);
      _ ->
        ok
    end,

    Res
  end), FS }.

　　

最后完成编译和加载的重头戏就在module(Forms, File, Opts, Callback)方法了:

 

%% Compile the module by forms based on the scope information
%% executes the callback in case of success. This automatically
%% handles errors and warnings. Used by this module and elixir_module.
module(Forms, File, Opts, Callback) ->
  DebugInfo = (get_opt(debug_info) == true) orelse lists:member(debug_info, Opts),
  Final =
    if DebugInfo -> [debug_info];
       true -> []
    end,
  module(Forms, File, Final, false, Callback).

module(Forms, File, RawOptions, Bootstrap, Callback) when
    is_binary(File), is_list(Forms), is_list(RawOptions), is_boolean(Bootstrap), is_function(Callback) ->
  { Options, SkipNative } = compile_opts(Forms, RawOptions),
  Listname = elixir_utils:characters_to_list(File),

  case compile:noenv_forms([no_auto_import()|Forms], [return,{source,Listname}|Options]) of
    {ok, ModuleName, Binary, RawWarnings} ->
      Warnings = case SkipNative of
        true  -> [{?MODULE,[{0,?MODULE,{skip_native,ModuleName}}]}|RawWarnings];
        false -> RawWarnings
      end,
      format_warnings(Bootstrap, Warnings),
      %%%% ModuleName :'Elixir.Math' ListName: "/data2/elixir/m.ex"
      code:load_binary(ModuleName, Listname, Binary),
      Callback(ModuleName, Binary);
    {error, Errors, Warnings} ->
      format_warnings(Bootstrap, Warnings),
      format_errors(Errors)
  end.

到这里编译的流程已经走完,附上几个可能会用到的资料,首先是elixir_parser:parse(Tokens)的代码,你可能会奇怪这个模块的代码在哪里?这个是通过elixir_parser.yrl编译自动生成的模块,你可以使用下面的方法拿到它的代码:

Eshell V5.10.2 (abort with ^G)
1> {ok,{_,[{abstract_code,{_,AC}}]}} = beam_lib:chunks("elixir_parser",[abstract_code]).
{ok,{elixir_parser,
[{abstract_code,
........
{...}|...]}}]}}
2> Dump= fun(Content)-> file:write_file("/data/dump.data", io_lib:fwrite("~ts.\n", [Content])) end.
#Fun<erl_eval.6.80484245>
3> Dump(erl_prettypr:format(erl_syntax:form_list(AC))).
ok
4>

　　　

 下一步,就要深入几个非常有意思的细节了

 

%% TODO

1. elixir_aliases的设计
2. elixir_scope 的设计
3. elixir macro 相关的几个话题:hygiene unquote_splicing

 今天先到这里.

 

 

 

 

 

最后,小图一张:

 

马背上的Godiva夫人

主人公:戈黛娃夫人Lady Godiva，或称Godgifu，约990年—1067年9月10日
作者：约翰·柯里尔（John Collier）所绘，约1898年

据说大约在1040年，统治考文垂（Coventry）城市的Leofric the Dane伯爵决定向人民征收重税，支持军队出战，令人民的生活苦不堪言。伯爵善良美丽的妻子Godiva夫人眼见民生疾苦，决定恳求伯爵减收徵税，减轻人民的负担。Leofric伯爵勃然大怒，认为Godiva夫人为了这班爱哭哭啼啼的贱民苦苦衷求，实在丢脸。Godiva夫人却回答说伯爵定会发现这些人民是多么可敬。他们决定打赌——Godiva夫人要赤裸身躯骑马走过城中大街，仅以长发遮掩身体，假如人民全部留在屋内，不偷望Godiva夫人的话，伯爵便会宣布减税。翌日早上，Godiva夫人骑上马走向城中，Coventry市所有百姓都诚实地躲避在屋内，令大恩人不至蒙羞。事后，Leofric伯爵信守诺言，宣布全城减税。这就是著名的Godiva夫人传说。