New features (albeit very minor ones):

  * Eliminated tail recursion in favor of loops in the generated C code; 
    no dependency on gcc for implementing this feature anymore.

  * Modified the rts queue handling so that messages with identical 
    deadlines/baselines are circulated wherever possible (even though 
    this cannot affect schedulability).

  * Changed program startup code to exclude module initializations 
    from work attributed to the startup message.

  * Made the number of cpu cores a dynamically selected parameter instead 
    of a hard coded value.

  * Changed name of primitive type 'PID' to 'OID' in accordance with 
    documentation.

  * Wiped out the 'while' syntax.

  * Extended the possibilities to use record stuffing patterns in 
    structs that themselves are defined using stuffing syntax.  

  * Allowed internally generated state types to appear abstractly in 
    module interfaces.

  * Implemented the static Delay rule for subtype coercions, which 
    transforms coercion selections from forward references into forward 
    references in their own right.  Not as nice as a dynamic Delay rule, 
    but better than ill-founded recursion generated by the compiler.

  * Changed the script ./timberc to run the latest compiler build 
    relying on a previous install for a valid datadir.

  * Changed the monomorphism restriction so that single variable 
    right-hand sides are allowed to be polymorphic, but non-function 
    recursion is always monomorphic.

  * Changed some defs in Prelude to use struct stuffing in a more direct 
    way.

  * Separated the Monad and Applicative type classes to better align with 
    Haskell wisdom, and updated the parsing combinator files accordingly.

  * Adapted Data.Objects.Dictionary to use monadClass.

  * Added function fromJust to Prelude.

  * Added an implementation of "The Santa Problem" to the examples 
    collection.


Bug fixes (mostly listed in internal form, with little connection to the 
examples that trigger them):

  * Fixed Prelude bugs in 'replicate' and the Applicative Maybe instance.  

  * Moved the start-gc-test to a more proper place *before* a worker might 
    go to sleep.  

  * Fixed bug in the POSIX rts leading to active threads not being 
    properly registered in the runQ.

  * Fixed race condition bug in the code for allocating new heap segments.

  * Fixed bug in the placement of interrupt airholes in scanTimerQ.

  * Switched to using octal escape sequences in the generated C code, as 
    well as some standard escapes, for the purpose of avoiding C/Haskell 
    mismatches (and the ambiguous length of hexadecimal escapes).

  * Fixed bug causing rts function Raise to crash when called from a non-Cmd 
    context.

  * Fixed bug causing incorrect gcinfo being generated for extensions to the 
    Msg type.

  * Fixed a bug in the splitting of switches into nullary/non-nullary cases 
    in Prepare4C.

  * Fixed a bug in Prepare4C causing loss of dependencies between generated 
    bindings.

  * Added missing indirect dependencies on free type variables to the 
    lambda-lifter.

  * Fixed the Kindle translation of Class, Cmd and Request types so that it 
    works even if the saturated type expression is obtained only after a 
    type substitution.

  * Reworked the arity-adapting pieces of Core2Kindle to work correctly even 
    in the higher-order case.

  * Fixed incorrect entering of polymorphic closures in Core2Kindle.

  * Fixed bug in Type2: polymorphic quantifiers not already present in the 
    input to Type2 were not properly encoded in the generated terms.

  * Removed printing of what would seem like scoped type variables in the 
    type inference output, and also removed prefixing of "_" to truly scoped 
    variable occurrences in type declarations.

  * Prevented term reduction from choking on static divide-by-zero terms.

  * Removed erroneous inlining of state variables.

  * Corrected the code for type-expression duplication in the presence of 
    free unification variables. (The parsing combinator tests work again!)

  * Fixed unexplicable slip that avoided using the special code for reducing 
    case alternatives (no eta-reduction must be allowed here).

  * Removed incorrect resetting of the 'ticked' constraint attribute which 
    caused incomplete removal of internally generated coercion constraints.

  * Fixed bug causing incorrect pruning of substitutions in the type-
    checker.

  * Fixed accidental reordering of mutually recursive binidings during 
    dependency analysis.

  * Made type syntax checks generate proper error messages instead of 
    internal errors (and added an explicit error message for illegal use of 
    type wildcards).


Various code cleanups:

  * Removed superfluous file env.h

  * Introduced a custom made Exception type for error handling in the 
    compiler.

  * Added test files related to recent bug fix history.

  * Added "linkLibs" as a separate configuration parameter in each rts.

  * Made the original form of a symbolic name (re)appear in the generated 
    C code as a comment appended to the generated name.

  * Added primitive closure names for arities 1-3.