version 0.9.0
=============

- ``=`` should be overloadable; requires specialization for ``=``
- fix remaining generics bugs
- fix remaining closure bugs:
  - fix evals.nim with closures
  - deactivate lambda lifting for JS backend
  - Test capture of for loop vars; test generics;
  - test constant closures

- GC: marker procs for native Nimrod GC and Boehm GC; precise stack marking;
  escape analysis for string/seq seems to be easy to do too;
  even further write barrier specialization
- dead code elim for JS backend; 'of' operator for JS backend
- const ptr/ref
- unsigned ints and bignums; requires abstract integer literal type: 
  use tyInt+node for that
- implement the high level optimizer
- change overloading resolution
- implement proper coroutines
- implement ``partial`` pragma for partial evaluation; ``hoist`` pragma for
  loop hoisting
- we need to support iteration of 2 different data structures in parallel
- make exceptions compatible with C++ exceptions
- change how comments are part of the AST
- optional indentation for 'case' statement; hm, keep in mind other syntax
  changes that people want; may turn out to be a bad idea
- activate more thread tests
- optimize method dispatchers
- rethink the syntax: distinction between expr and stmt is unfortunate; 
  indentation handling is quite complex too; problem with exception handling
  is that often the scope of ``try`` is wrong and apart from that ``try`` is
  a full blown statement; a ``try`` expression might be a good idea to make
  error handling more light-weight
  people also want ``inc a; inc b``

Bugs
----
- bug: generic assign still buggy
  - special case the generic assign that needs to care about case objects

- bug: returning a tyVar does not mean it is save to return it:
  proc forward[T](x: var T): var T = result = x
  proc p(): var int = 
    var x: int
    # reject this call via alias analysis:
    result = forward(x)

- bug: stress testing basic method example (eval example) 
  without ``-d:release`` leaks memory?
- bug: object {.pure, final.} does not work again!
- bug: tsortdev does not run with native GC?


version 0.9.XX
==============

- implicit ref/ptr->var conversion; the compiler may store an object
  implicitly on the heap for write barrier efficiency; better: 
  proc specialization in the code gen
- shared memory heap: ``shared ref`` etc. The only hard part in the GC is to
  "stop the world". However, it may be worthwhile to generate explicit 
  (or implicit) syncGC() calls in loops. Automatic loop injection seems
  troublesome, but maybe we can come up with a simple heuristic. (All procs
  that `new` shared memory are syncGC() candidates...)
- EcmaScript needs a new and better code gen: simply adapt the C code gen to it
- tlastmod returns wrong results on BSD (Linux, MacOS X: works)
- nested tuple unpacking; tuple unpacking in non-var-context
- 'nimrod def': does not always work?
- implement/generalize the effect system
- test branch coverage
- checked exceptions
- make pegs support a compile-time option and make c2nim use regexes instead
  per default?
- fix implicit generic routines
- 'const' objects including case objects
- improve docgen to use the semantic pass
- 'export' feature (requires improved docgen)
- think about ``{:}.toTable[int, string]()``
- mocking support with ``tyProxy`` that does:
  o.p(x) --> p(o, x) -->  myMacro(p, o, x)
  
  This is really the opposite of ``tyExpr``:
  * For parameter ``tyExpr`` any argument matches.
  * Argument ``tyProxy`` matches any parameter.
  
- nice idea:

  p(a, b): 
    echo a
    echo b
  
  is the same as:
  
  p(a, b, proc() =
    echo a
    echo b)

Library
-------

- wrappers for mongodb; poppler; libharu
- suffix trees
- locale support; i18n module
- bignums

- pdcurses bindings

- for system:
  proc `@` [T](a: openArray[T]): seq[T] = 
    newSeq(result, a.len)
    for i in 0..a.len-1: result[i] = a[i]
    
  --> ensure @[] calls the array version!


Low priority
------------

- ``with proc `+`(x, y: T): T`` for generic code
- new feature: ``distinct T with operations``
- find a way for easy constructors and destructors; (destructors are much more
  important than constructors)
- code generated for type information is wasteful
- resizing of strings/sequences could take into account the memory that
  is allocated
- timeout for locks
- compilation cache:
  - adapt thread var emulation to care about the new merge operation
  - check for interface changes; if only the implemenation changes, no
    need to recompile clients; er ... what about templates, macros or anything
    that has inlining semantics?
- codegen should use "NIM_CAST" macro and respect aliasing rules for GCC
- warning for implicit openArray -> varargs conversion
- implement explicit varargs; **but** ``len(varargs)`` problem remains! 
  --> solve by implicit conversion from varargs to openarray
- implement closures that support nesting > 1


Further optimization ideas
==========================

- To optimize further copies away, you want to gather the additional
  information inlining would provide, but don't inline for code size reasons.


Version 2
=========

- language change: inheritance should only work with reference types, so that
  the ``type`` field is not needed for objects! --> zero overhead aggregation
  BETTER: ``of`` and safe object conversions only work with ref objects. Same
  for multi methods.
  
- explicit nil types?
  * nil seq[int]
  * nil string
  * nil ref int
  * nil ptr THallo
  * nil proc 

- better for backwards compatibility: default nilable, but ``not nil``
  notation:
  
  type
    PWindow = ref TWindow not nil
    
  The problem with ``nil`` is that the language currently relies on it for
  implicit initialization. Initialization is different from assignment. The
  issues can "easily" dealt with by ensuring:
  
    var x = myProc() # checks myProc() initializes every pointer explicitely

- guards for the 'case' statement; generalized case statement;
  a guard looks like: 
  
    case x
    of nkStmtList if x.value == 0: 
  
  a generalized case statement looks like:
    
    case x with `=~`