rfc:ast_based_parsing_compilation_process

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision
Previous revision
rfc:ast_based_parsing_compilation_process [2012/09/07 13:15]
nikic
rfc:ast_based_parsing_compilation_process [2017/09/22 13:28] (current)
Line 1: Line 1:
-====== Request for Comments: Moving to an AST-based parsing/compilation process ======+====== Request for Comments: Moving to an AST-based parsing/compilation process (obsolete) ======
   * Date: 2012-09-04   * Date: 2012-09-04
   * Author: Nikita Popov <nikic@php.net>   * Author: Nikita Popov <nikic@php.net>
-  * Status: Under Discussion+  * Status: Obsolete 
 +  * [[http://markmail.org/message/trt5oz5uioxe3fdv|Mailing list discussion]] 
 +  * Superseded by: [[rfc:abstract_syntax_tree|Abstract Syntax Tree RFC]]
  
 ===== Introduction ===== ===== Introduction =====
 +
 +**Note: This RFC has been superseded by another [[rfc:abstract_syntax_tree|Abstract Syntax Tree RFC]].**
  
 Currently PHP uses a single-pass compilation process, i.e. the parser directly invokes opcode compilation routines. Most other languages on the other hand use an intermediary structure to separate those two phases: The parser only emits an abstract syntax tree (AST), which is then used by a separate compiler to emit instructions. The use of an AST decouples the two phases and as such allows for greater flexibility and deeper analysis. Currently PHP uses a single-pass compilation process, i.e. the parser directly invokes opcode compilation routines. Most other languages on the other hand use an intermediary structure to separate those two phases: The parser only emits an abstract syntax tree (AST), which is then used by a separate compiler to emit instructions. The use of an AST decouples the two phases and as such allows for greater flexibility and deeper analysis.
Line 24: Line 28:
 ==== Elimination of various quirks ==== ==== Elimination of various quirks ====
  
-Currently there is various quirks in the emitted opcodes which can be attributed to the use of a single-pass compiler. Some examples:+Currently there are various quirks in the emitted opcodes which can be attributed to the use of a single-pass compiler. Some examples:
  
   * The NOP opcodes that are inserted in several places. (Yes, this point isn't particularly important)   * The NOP opcodes that are inserted in several places. (Yes, this point isn't particularly important)
Line 42: Line 46:
 With the current single-pass compiler some things are very hard / near impossible to implement. This actively influences syntax decisions. With the current single-pass compiler some things are very hard / near impossible to implement. This actively influences syntax decisions.
  
-Two examples of syntax that is currently not possible, but would be possible with a syntax tree:+A few examples of syntax that is currently not possible, but would be possible with a syntax tree:
  
   * Array destructuring using something like ''[$a, $b, $c] = $array'' instead of a dedicated ''list()'' syntax. This is common in other languages, but not possible in PHP.   * Array destructuring using something like ''[$a, $b, $c] = $array'' instead of a dedicated ''list()'' syntax. This is common in other languages, but not possible in PHP.
   * List comprehensions / generator expressions where the result expression comes first, e.g. ''[x * x for x in list]'' in Python. In PHP only the reverse syntax is possible: ''[foreach ($list as $x) yield $x * $x]''   * List comprehensions / generator expressions where the result expression comes first, e.g. ''[x * x for x in list]'' in Python. In PHP only the reverse syntax is possible: ''[foreach ($list as $x) yield $x * $x]''
 +  * C#-style expression trees (which form the basis for LINQ)
 +
 +Apart from larger syntax limitations the current system commonly also affects smaller syntax decisions. One example here are the strange parentheses requirements for the ''yield'' expression. Those requirements exist solely for technical reasons and would not be required with an AST-generating parser.
 +
 +==== Better error messages ====
  
-Apart from larger syntax limitations the current system commonly also affects smaller syntax decisions.+Currently many things are directly enforced in the grammar which should really be checked during compilation (or a completely separate pass)E.g. if you try to initialize a class property with a non-static value, you'll get a rather unintelligible parse error message, instead of something like ''Cannot initialize property with non-static value''. (And then you obviously go to StackOverflow, ask the question for the five hundredth time and annoy the heck out of me!)
  
 ===== Disadvantages ===== ===== Disadvantages =====
rfc/ast_based_parsing_compilation_process.1347023715.txt.gz · Last modified: 2017/09/22 13:28 (external edit)