adding the ability to grow the value stack as needed
the net effect is that calls use much less space, and stack frames
can use all available heap space. the only downside is that C builtins
must be aware that the stack can change out from under them if they call
lisp code. currently the only example of this is table.foldl.
also fixing bug where exceptions failed to unwind the gc handle stack.
making branch destinations relative so we can access the code
using *ip instead of code[ip]
doing the same to bp (argument access)
small improvement to map, builtin->instruction
better profiling
using copy-list to implement append2
fixing bug in op_vargc when nrequired==MAX_ARGS
making fetching of multibyte numbers from bytecode portable
some cleanup in compiler
various bug fixes
language changes:
• constant symbols no longer shadow everything
• eval* removed
• vararg lists always allocated on entry, dotted argument
lists not preserved
new applyn() entry point
so recursion depth is limited only by the process stack limit.
reorganizing evaluator so the same code is used for evaluating
and pushing arguments for both builtin functions and lambdas.
for now this is slower, but it was done in preparation for
Things To Come.
adding list-head
implementing the calling convention for long argument lists in
bytecode compiler. arguments are broken down into a nest of
list and nconc calls.
also implementing vararg builtins.
fixing bug in ios, not initializing readonly flag
updating string and sizeof to use new strstream functions
removing some redundant numeric type init functions
enabling type-specific print and relocate behavior
allowing GC to be triggered by large buffer allocations
adding hash table constructor and print function
renamed some functions
explicitly represented in an fltype_t struct, and symbolic
types are hash-consed. a lot of code is smaller and simpler
as a result. this should allow more features in less space
(both code and data) going forward.
changing \DDD and \x escape sequences to read bytes instead of
characters
re-fixing uint64 cast bug
adding Paul Hsieh's hash function, to be evaluated later
some cleanup
moving some library code around for size optimization
now using == instead of flt_equals for float comparison, mostly
for hash compatibility
better text representation of infs and nans
removing construction of low-level numeric types directly from strings
adding hash function corresponding to equal
better way to initialize builtins
moving advanced bitvector ops to separate compilation unit
rearranging hash table code so it can be specialized for different
comparison functions
one good way to bloat a piece of software is to add several
ASCII pictures of the mona lisa
after all that's the whole point of bitvectors
some prettyprinting tweaks
more uniform way to handle forwarding pointers. fix forwarding
of gensyms
:keyword symbols
JeffBezanson