Nim/doc/intern.txt

=========================================
    Internals of the Nimrod Compiler
=========================================


:Author: Andreas Rumpf
:Version: |nimrodversion|

.. contents::


Directory structure
===================

The Nimrod project's directory structure is:

============   ==============================================
Path           Purpose
============   ==============================================
``bin``        binary files go into here
``nim``        Pascal sources of the Nimrod compiler; this
               should be modified, not the Nimrod version in
               ``rod``!
``rod``        Nimrod sources of the Nimrod compiler;
               automatically generated from the Pascal
               version
``data``       data files that are used for generating source
               code go into here
``doc``        the documentation lives here; it is a bunch of
               reStructuredText files
``dist``       download packages as zip archives go into here
``config``     configuration files for Nimrod go into here
``lib``        the Nimrod library lives here; ``rod`` depends
               on it!
``web``        website of Nimrod; generated by ``koch.py``
               from the ``*.txt`` and ``*.tmpl`` files
``koch``       the Koch Build System (written for Nimrod)
``obj``        generated ``*.obj`` files go into here
============   ==============================================


Bootstrapping the compiler
==========================

The compiler is written in a subset of Pascal with special annotations so
that it can be translated to Nimrod code automatically. This conversion is
done by Nimrod itself via the undocumented ``boot`` command. Thus both Nimrod
and Free Pascal can compile the Nimrod compiler.

Requirements for bootstrapping:

- Free Pascal (I used version 2.2) [optional]
- Python (should work with version 1.5 or higher)

- C compiler -- one of:

  * win32-lcc (currently broken)
  * Borland C++ (tested with 5.5; currently broken)
  * Microsoft C++
  * Digital Mars C++
  * Watcom C++ (currently broken)
  * GCC
  * Intel C++
  * Pelles C (currently broken)
  * llvm-gcc

| Compiling the compiler is a simple matter of running:
| ``koch.py boot``
| Or you can compile by hand, this is not difficult.

If you want to debug the compiler, use the command::

  koch.py boot --debugger:on

The ``koch.py`` script is Nimrod's maintainance script: Everything that has 
been automated is accessible with it. It is a replacement for make and shell
scripting with the advantage that it is more portable.


Coding standards
================

The compiler is written in a subset of Pascal with special annotations so
that it can be translated to Nimrod code automatically. As a general rule,
Pascal code that does not translate to Nimrod automatically is forbidden.


Porting to new platforms
========================

Porting Nimrod to a new architecture is pretty easy, since C is the most
portable programming language (within certain limits) and Nimrod generates
C code, porting the code generator is not necessary.

POSIX-compliant systems on conventional hardware are usually pretty easy to
port: Add the platform to ``platform`` (if it is not already listed there),
check that the OS, System modules work and recompile Nimrod.

The only case where things aren't as easy is when the garbage
collector needs some assembler tweaking to work. The standard
version of the GC uses C's ``setjmp`` function to store all registers
on the hardware stack. It may be that the new platform needs to
replace this generic code by some assembler code.


Runtime type information
========================

*Runtime type information* (RTTI) is needed for several aspects of the Nimrod
programming language:

Garbage collection
  The most important reason for RTTI. Generating
  traversal procedures produces bigger code and is likely to be slower on
  modern hardware as dynamic procedure binding is hard to predict.

Complex assignments
  Sequences and strings are implemented as
  pointers to resizeable buffers, but Nimrod requires copying for
  assignments. Apart from RTTI the compiler could generate copy procedures
  for any type that needs one. However, this would make the code bigger and
  the RTTI is likely already there for the GC.

We already knew the type information as a graph in the compiler.
Thus we need to serialize this graph as RTTI for C code generation.
Look at the file ``lib/hti.nim`` for more information.


The Garbage Collector
=====================

Introduction
------------

We use the term *cell* here to refer to everything that is traced
(sequences, refs, strings).
This section describes how the new GC works.

The basic algorithm is *Deferrent reference counting* with cycle detection.
References in the stack are not counted for better performance and easier C
code generation.

Each cell has a header consisting of a RC and a pointer to its type
descriptor. However the program does not know about these, so they are placed at
negative offsets. In the GC code the type ``PCell`` denotes a pointer
decremented by the right offset, so that the header can be accessed easily. It
is extremely important that ``pointer`` is not confused with a ``PCell``
as this would lead to a memory corruption.


The CellSet data structure
--------------------------

The GC depends on an extremely efficient datastructure for storing a
set of pointers - this is called a ``TCellSet`` in the source code.
Inserting, deleting and searching are done in constant time. However,
modifying a ``TCellSet`` during traversation leads to undefined behaviour.

.. code-block:: Nimrod
  type
    TCellSet # hidden

  proc CellSetInit(s: var TCellSet) # initialize a new set
  proc CellSetDeinit(s: var TCellSet) # empty the set and free its memory
  proc incl(s: var TCellSet, elem: PCell) # include an element
  proc excl(s: var TCellSet, elem: PCell) # exclude an element

  proc `in`(elem: PCell, s: TCellSet): bool # tests membership

  iterator elements(s: TCellSet): (elem: PCell)


All the operations have to be perform efficiently. Because a Cellset can
become huge a hash table alone is not suitable for this.

We use a mixture of bitset and hash table for this. The hash table maps *pages*
to a page descriptor. The page descriptor contains a bit for any possible cell
address within this page. So including a cell is done as follows:

- Find the page descriptor for the page the cell belongs to.
- Set the appropriate bit in the page descriptor indicating that the
  cell points to the start of a memory block.

Removing a cell is analogous - the bit has to be set to zero.
Single page descriptors are never deleted from the hash table. This is not
needed as the data structures need to be periodically rebuilt anyway.

Complete traversal is done in this way::

  for each page decriptor d:
    for each bit in d:
      if bit == 1:
        traverse the pointer belonging to this bit


Further complications
---------------------

In Nimrod the compiler cannot always know if a reference
is stored on the stack or not. This is caused by var parameters.
Consider this example:

.. code-block:: Nimrod
  proc setRef(r: var ref TNode) =
    new(r)

  proc usage =
    var
      r: ref TNode
    setRef(r) # here we should not update the reference counts, because
              # r is on the stack
    setRef(r.left) # here we should update the refcounts!

Though it would be possible to produce code updating the refcounts (if
necessary) before and after the call to ``setRef``, it is a complex task to
do so in the code generator. So we don't and instead decide at runtime
whether the reference is on the stack or not. The generated code looks
roughly like this:

.. code-block:: C
  void setref(TNode** ref) {
    unsureAsgnRef(ref, newObj(TNode_TI, sizeof(TNode)))
  }
  void usage(void) {
    setRef(&r)
    setRef(&r->left)
  }

Note that for systems with a continous stack (which most systems have)
the check whether the ref is on the stack is very cheap (only two
comparisons). Another advantage of this scheme is that the code produced is
smaller.


The algorithm in pseudo-code
----------------------------
To be written.


The compiler's architecture
===========================

Nimrod uses the classic compiler architecture: A scanner feds tokens to a
parser. The parser builds a syntax tree that is used by the code generator.
This syntax tree is the interface between the parser and the code generator.
It is essential to understand most of the compiler's code.

In order to compile Nimrod correctly, type-checking has to be seperated from
parsing. Otherwise generics would not work. Code generation is done for a 
whole module only after it has been checked for semantics.

.. include:: filelist.txt

The first command line argument selects the backend. Thus the backend is
responsible for calling the parser and semantic checker. However, when 
compiling ``import`` or ``include`` statements, the semantic checker needs to
call the backend, this is done by embedding a PBackend into a TContext.


The syntax tree
---------------
The synax tree consists of nodes which may have an arbitrary number of
children. Types and symbols are represented by other nodes, because they
may contain cycles. The AST changes its shape after semantic checking. This
is needed to make life easier for the code generators. See the "ast" module
for the type definitions.

We use the notation ``nodeKind(fields, [sons])`` for describing
nodes. ``nodeKind[sons]`` is a short-cut for ``nodeKind([sons])``.
XXX: Description of the language's syntax and the corresponding trees.


How the RTL is compiled
=======================

The system module contains the part of the RTL which needs support by
compiler magic (and the stuff that needs to be in it because the spec
says so). The C code generator generates the C code for it just like any other
module. However, calls to some procedures like ``addInt`` are inserted by
the CCG. Therefore the module ``magicsys`` contains a table
(``compilerprocs``) with all symbols that are marked as ``compilerproc``.



Generation of dynamic link libraries
====================================

Generation of dynamic link libraries or shared libraries is not difficult; the
underlying C compiler already does all the hard work for us. The problem is the
common runtime library, especially the memory manager. Note that Borland's
Delphi had exactly the same problem. The workaround is to not link the GC with
the Dll and provide an extra runtime dll that needs to be initialized.



How to implement closures
=========================

A closure is a record of a proc pointer and a context ref. The context ref
points to a garbage collected record that contains the needed variables.
An example:

.. code-block:: Nimrod

  type
    TListRec = record
      data: string
      next: ref TListRec

  proc forEach(head: ref TListRec, visitor: proc (s: string) {.closure.}) =
    var it = head
    while it != nil:
      visit(it.data)
      it = it.next

  proc sayHello() =
    var L = new List(["hallo", "Andreas"])
    var temp = "jup\xff"
    forEach(L, lambda(s: string) =
                 io.write(temp)
                 io.write(s)
           )


This should become the following in C:

.. code-block:: C
  typedef struct ... /* List type */

  typedef struct closure {
    void (*PrcPart)(string, void*);
    void* ClPart;
  }

  typedef struct Tcl_data {
    string temp; // all accessed variables are put in here!
  }

  void forEach(TListRec* head, const closure visitor) {
    TListRec* it = head;
    while (it != NIM_NULL) {
      visitor.prc(it->data, visitor->cl_data);
      it = it->next;
    }
  }

  void printStr(string s, void* cl_data) {
    Tcl_data* x = (Tcl_data*) cl_data;
    io_write(x->temp);
    io_write(s);
  }

  void sayhello() {
    Tcl_data* data = new(...);
    asgnRef(&data->temp, "jup\xff");
    ...

    closure cl;
    cl.prc = printStr;
    cl.cl_data = data;
    foreach(L, cl);
  }


What about nested closure? - There's not much difference: Just put all used
variables in the data record.