ChangeLog

2010-03-27 04:58 rushing
[r496] tests/parse_0.py more complicated expression
2010-03-27 04:57 rushing
[r495] tests/t_parse.scm rewrote several vcase exprs using pattern matching (with lists)
2010-03-27 04:57 rushing
[r494] tests/t_match7.scm T/insert(): rewrite some vcase exprs with pattern matching
2010-03-27 04:56 rushing
[r493] lib/core.scm impossible(): new error function
2010-03-27 04:56 rushing
[r492] transform.py expand_nvcase() => exp_nvcase()expand_pvcase() => exp_pvcase() this avoids calling the transform more than once
2010-03-27 04:55 rushing
[r491] nodes.py apply_subst(): don't bother putting entries for '_' in the lenv
2010-03-27 04:54 rushing
[r490] match.py kind(): support list patterns "()", "(x . y)", "(a b c d ...)"subst(): don't emit subst for wildcard varsconstructor_rule(): when possible use '_' for wildcards
2010-03-26 19:46 rushing
[r489] lib/string.scm lame string->int()
2010-03-26 19:46 rushing
[r488] Makefile make sure we're clean
2010-03-26 19:45 rushing
[r487] tests/parse_0.py test toy predicate language
2010-03-26 19:45 rushing
[r486] tests/t_string2.scm test string->int
2010-03-26 19:43 rushing
[r485] tests/t_parse.scm added a parse-tree => AST phase using pattern matching.
2010-03-26 03:51 rushing
[r484] tests/t_match9.scm test <match> special form
2010-03-26 03:51 rushing
[r483] tests/t_match8.scm test matching on symbols
2010-03-26 03:50 rushing
[r482] tests/t_match7.scm Use pattern matching to print the tree.
2010-03-26 03:49 rushing
[r481] match.py compile(): we don't need <name>. return vars, codekind(): handle symbols in patterns
2010-03-26 03:47 rushing
[r480] transform.py exp_define(): take some specific work from exp_match()exp_match(): handle rules from either 'define' or 'match'expand_match(): new match special form
2010-03-25 22:27 rushing
[r479] parse/extract.py renamed to gen_irken.py
2010-03-25 22:26 rushing
[r478] tests/f_arity.scm, tests/f_match5.scm, tests/parse_0.py, tests/t_match0.scm, tests/t_match2.scm, tests/t_match4.scm, tests/t_match5.scm, tests/t_match6.scm, tests/t_match7.scm, tests/t_match8.scm, tests/t_nvcase0.scm, tests/t_parse.scm new pattern matching tests
2010-03-25 22:24 rushing
[r477] backend.py insn_vcase(): massively simplified again, using get_safe_typecode(). necessary because nvcase is no longer required to be complete, thus making it possible to leave out a unit constructor and miss an immediate alternative.
2010-03-25 22:22 rushing
[r476] cps.py disentangle pvcase and nvcase, especially since they now have such different shapescompile_primap(): support %%match-error
2010-03-25 22:21 rushing
[r475] parse/README updated
2010-03-25 22:21 rushing
[r474] parse/gen_python_parser.py try to do all the necessary steps in this script, and take a grammar argument
2010-03-25 22:20 rushing
[r473] parse/meta.py alts: build an nary VBAR
2010-03-25 22:19 rushing
[r472] parse/gen_parser.py reworked to (hopefully) generate more predictable names for sub-rules so that we can build practical parsers based on the output.
2010-03-25 22:18 rushing
[r471] header.c comment
2010-03-25 22:17 rushing
[r470] lib/pair.scm map()
2010-03-25 22:17 rushing
[r469] lib/string.scm string-compare(): rewritten a little more clearly now, let_reg makes it ok.
2010-03-25 22:16 rushing
[r468] lib/symbol.scm symbol->stringsymbol-<?
2010-03-25 22:16 rushing
[r467] itypes.py datatype.arity(): return the arity of a constructor
2010-03-25 22:15 rushing
[r466] tests/t_dump_image.scm 'thunk' now takes an argument
2010-03-25 22:13 rushing
[r465] lib/core.scm dump(),load(): can't use a thunk easily since call/cc wants an arg.
2010-03-25 22:12 rushing
[r464] transform.py __init__(): initialize a pattern match compilerfind_datatypes(): scan for datatypes at top level, rather than wherever we might find them.process_formals(): remove variable-arity codeexp_define(): catch pattern-matching definitionsexp_match(): call the match compilerexpand_nvcase(): removed the assumption that all alts are present, which forces us to keep track of which ones are. add support for an 'else' clause.expand_datatype() => parse_datatype()
2010-03-25 22:08 rushing
[r463] solver.py gen_nvcase(): handle the new 'ealt' sub-expression.decompose(): Wow, nasty bug I stuck in there in February that caused it to automatically give a pass whenever the number of args to a predicate didn't match. Yeah, I know!lookup_special_names(): add '%%match-error'
2010-03-25 22:06 rushing
[r462] nodes.py new node type: let_subst used to express variable substitutions so that they can be resolved in a single pass, during the translation into nodes.
2010-03-25 22:05 rushing
[r461] match.py pattern matching!
2010-03-16 03:17 rushing
[r460] parse/gen_irken.py renamed
2010-03-13 04:31 rushing
[r459] tests/t_case0.scm debugging a problem with symbols in code generated for case
2010-03-13 04:31 rushing
[r458] tests/t_string1.scm test mutability of string literals
2010-03-13 04:30 rushing
[r457] tests/t_string0.scm new tests
2010-03-13 04:29 rushing
[r456] tests/t28.exp, tests/t28.scm, tests/t_literal.scm, tests/t_literal0.scm, tests/t_literal1.scm, tests/t_literal2.scm, tests/t_literal3.scm, tests/t_literal4.scm, tests/t_literal5.scm, tests/t_literal6.scm, tests/t_literal7.scm, tests/t_literal8.scm, tests/t_symbols.exp new tests
2010-03-13 04:29 rushing
[r455] notes/literals.txt straight talk about literals
2010-03-13 04:28 rushing
[r454] backend.py emit_constructed_literals(): new (biggish) functionverify():insn_verify(): removed, along with some commented-out callsinsn_constructed(): emit a reference to a constructed literalinsn_make_string(): removed
2010-03-13 04:26 rushing
[r453] cps.py scan_constructed(): scan and collect all constructed literalsgen_constructed(): new insn
2010-03-13 04:25 rushing
[r452] parse/lexer.py reworked to support new literal syntax
2010-03-13 04:25 rushing
[r451] parse/extract.py reworked to support new literal syntax
2010-03-13 04:24 rushing
[r450] header.c placeholder for constructed literalschanged the magic string to a //comment
2010-03-13 04:24 rushing
[r449] pxll.h STRING_TUPLE_LENGTH STRING_HEADER UOTAG UITAG UCON UCON0 UOHEAD: new macros make it easier to construct literalspxll_string: declare the data array to have no length rather than zero length
2010-03-13 04:23 rushing
[r448] itypes.py comments
2010-03-13 04:22 rushing
[r447] transform.py RuntimeLiteral: exception for literals with illegal bits that can't be figured at compile-timeget_constant_binding():add_constants(): removed!expand_exp(): leave strings and symbols aloneexpand_literal(): support the (literal ...) syntaxbuild_literal(): rewritten to support lisp/list literals and 'constructed' literals verifies that the construct can be built at compile-time
2010-03-13 04:17 rushing
[r446] solver.py type a constructed literal
2010-03-13 04:17 rushing
[r445] nodes.py constructed: new node type for constructed literals
2010-03-13 04:17 rushing
[r444] compile.py context: list of constructed literals symbol table
2010-03-07 09:22 rushing
[r443] backend.py emit_gc_copy_regs(): couple of one-off bugs there
2010-03-07 09:15 rushing
[r442] tests/t20.scm renamed to t_parse, still in progress
2010-03-07 09:14 rushing
[r441] backend.py emit_gc_copy_regs(): emit functions to copy registers in and out of the heap as gc roots
2010-03-07 09:13 rushing
[r440] backend.py track the size of allocations inside a function. if the value runs over head_room, then emit a call to check_heap() giving the number of free regsclass function: holds the accumulated allocation sumalloc(): sums up allocations in self.current_funinsn_close(): manage self.current_fun inside each function
2010-03-07 09:11 rushing
[r439] gc.c gc_flip(): now takes <nregs> arg call the generated gc_regs_in() & gc_regs_out()gc_relocate(): returns void
2010-03-07 09:09 rushing
[r438] header.c check_heap():gc_flip(): now takes <nregs> argconst head_room
2010-03-07 09:08 rushing
[r437] pxll.h put head room into a global const variable
2010-03-05 07:35 rushing
[r436] run_tests.py sometimes the lexer gets swapped out
2010-03-05 07:10 rushing
[r435] tests/parse_0.py simple parse test for t_parse.scm
2010-03-05 07:10 rushing
[r434] tests/t_alias.scm tests aliases
2010-03-05 07:10 rushing
[r433] tests/t_letreg0.scm *** empty log message ***
2010-03-05 07:09 rushing
[r432] tests/t_datatype4.scm tests aliases, and chunks of initialized data
2010-03-05 07:08 rushing
[r431] tests/t_datatype3.scm *** empty log message ***
2010-03-05 07:08 rushing
[r430] backend.py insn_vcase(): deref a pointer in GET_TYPECODE() when it's a tuple.
2010-03-05 07:07 rushing
[r429] cps.py gen_simple_test(): renamed an arg
2010-03-05 07:07 rushing
[r428] parse/README the forgotten extract.py step
2010-03-05 07:07 rushing
[r427] parse/gen_python_parser.py make the lexer when you make the parser, getting them out of sync is confusing.
2010-03-05 07:06 rushing
[r426] parse/extract.py reworked for normal-variant datatypes.
2010-03-05 07:06 rushing
[r425] tests/t_parse.scm previously known as 't20'.reworked for normal-variant datatypes.
2010-03-05 07:05 rushing
[r424] nodes.py node.alias: slot to hold value for aliases (see transform.py)parse_type.pfun(): handle zero-argument type predicates
2010-03-05 07:05 rushing
[r423] analyze.py find_aliases(): detect 'aliases' like "(define plus +)" and inline them. a bit of a hack, but useful for the parser tables.transform_0_varref(): return the value of the alias
2010-03-03 06:34 rushing
[r422] cps.py safe_for_let_reg(): don't use let_reg if any of the variables will escape
2010-03-03 06:33 rushing
[r421] header.c dump_object(): is declared in lib/core.scm to return undefined, and in fact needs to return undefined (i.e., 'dead') rather than #t
2010-03-03 06:32 rushing
[r420] lib/pair.scm datatype list: rearrange the order so that nil is u0 [to facilitate testing]
2010-03-03 06:32 rushing
[r419] lib/io.scm file:read-buffer(): a bug! found by a test!
2010-03-03 06:31 rushing
[r418] tests/t0.exp, tests/t1.exp, tests/t10.exp, tests/t11.exp, tests/t12.exp, tests/t13.exp, tests/t14.exp, tests/t15.exp, tests/t15.scm, tests/t16.exp, tests/t16.scm, tests/t18.exp, tests/t19.exp, tests/t22.scm, tests/t23.exp, tests/t23.scm, tests/t24.exp, tests/t24.scm, tests/t25.exp, tests/t26.exp, tests/t27.exp, tests/t27.scm, tests/t28.exp, tests/t29.scm, tests/t30.exp, tests/t30.scm, tests/t35.exp, tests/t36.scm, tests/t37.exp, tests/t38.exp, tests/t39.exp, tests/t39.scm, tests/t42.exp, tests/t42.scm, tests/t_bad_inline.exp, tests/t_cexp_wrap.exp, tests/t_datatype.exp, tests/t_datatype.scm, tests/t_datatype0.exp, tests/t_datatype1.scm, tests/t_dump_image.exp, tests/t_empty_vector.exp, tests/t_empty_vector.scm, tests/t_frb0.scm, tests/t_letreg.exp, tests/t_letreg.scm, tests/t_small.exp, tests/t_sorted_list.exp, tests/t_sorted_list.scm, tests/t_stack.exp, tests/t_symbols.exp, tests/t_vector.exp, tests/tak20.exp, tests/tak20.scm expected values for tests, and lots of tweaks to tests in order to make their output verifiable when possible.
2010-03-03 06:30 rushing
[r417] run_tests.py verify the actual output of testsseveral functions for tests that require special attention
2010-03-02 07:50 rushing
[r416] header.c dump_image(): return an int
2010-03-02 07:49 rushing
[r415] tests/t_rsum_0.scm unwritten test
2010-03-02 07:48 rushing
[r414] tests/f_datatype.scm, tests/f_datatype0.scm, tests/f_value_rest.scm, tests/t_cexp_wrap.scm, tests/t_datatype.scm, tests/t_datatype0.scm, tests/t_datatype1.scm, tests/t_datatype2.scm, tests/t_dump_image.scm, tests/t_frb0.scm, tests/t_letreg.scm, tests/t_lex.scm, tests/t_small.scm, tests/t_sorted_list.scm, tests/t_stack.scm, tests/t_symbols.scm, tests/t_vector.scm new tests
2010-03-02 07:46 rushing
[r413] backend.py wrap_in():wrap_out():insn_primop(): handle cexp itypesinsn_test(): changed cexp insn paramsinsn_vcase(): special case when only two branches (use "if")insn_move(): implement the move (fetch/assign) insn
2010-03-02 07:43 rushing
[r412] run_tests.py run tests in smallest-to-biggest order
2010-03-02 07:43 rushing
[r411] cps.py register_rib(): specialized rib type for environment lookup, to support the let_reg() constructcompiler.lexical_address(): handle register ribscompile_exp(): send 'fix' to 'let_splat', and 'let_splat' to 'let_reg' if it's a small leaf expressioncompile_varref():compile_varset(): dispatch to register when neededcollect_primargs(): add optional arg 'reorder' to stop it when used by let_reg (which is actually unused now, urgh)compile_let_reg(): feed init values into registers, extend environment ribcps.gen_move(): new old insn, fetch/assign to registerremove_moves(): tries to cut down on pointless register traffic, doesn't work right yet.walk(): add clause for pvcase/nvcasefind_allocation(): updated to (hopefully) catch all the allocating insns
2010-03-02 07:38 rushing
[r410] lib/pair.scm list: new colon syntax for tags
2010-03-02 07:38 rushing
[r409] lib/frb.scm tree: new colon syntax for tags
2010-03-02 07:37 rushing
[r408] lib/core.scm maybe: new colon syntax for tags
2010-03-02 07:37 rushing
[r407] lib/string.scm new 'raw' predicate cexp syntaxlist->string(): use correct vcase syntax
2010-03-02 07:37 rushing
[r406] itypes.py datatype.__init__(): pass in context simplified by new type parser
2010-03-02 07:36 rushing
[r405] tests/t29.scm new 'raw' predicate cexp syntax
2010-03-02 07:35 rushing
[r404] transform.py expand_datatype(): use the colon :syntax for tags, looks better, easier to readbuild_literal(): removed 'comma' hack, was broken anyway
2010-03-02 07:34 rushing
[r403] tests/t20.scm parser test temporarily disabled until I rewrite it
2010-03-02 07:32 rushing
[r402] tests/t16.scm cosmetic
2010-03-02 07:32 rushing
[r401] solver.py parse_cexp_type(): removedcheck_constraint(): make sure tvar is bound only onceconstraint_generator.go(): call check_constraint()gen_cexp(): rewritten to use itypes, instantiating each cexp
2010-03-02 07:31 rushing
[r400] nodes.py node.pprint(): print the 'L' leaf annotationparse_type(): rewritten to use itypeswalker.walk_exp(): use itypes for cexp
2010-03-02 07:29 rushing
[r399] analyze.py analyzer.__init__(): new attr, inline_multiplier removed vestigial 'pending_inlines'analyze(): find leaf expressionsget_fun_calls(): scale by the multiplier value for this function [adjusting for inlines]inline_threshold: lowered to 13find_inlines.replacer(): set the multiplier when calls > 1 commented out debug printset_multiplier(): adjust the call count for funs called by an inlined funfind_leaves(): find the leaves of the node tree (defined as everything below a funcall), the idea is to identify sub-trees where let_reg is safe to use.
2010-03-02 07:25 rushing
[r398] compile.py cps.irken_compiler => cps.cps
2010-02-24 18:13 rushing
[r397] tests/t17.scm free(): cast arg to void*
2010-02-24 18:09 rushing
[r396] tests/t16.scm test datatype version of frb
2010-02-24 18:07 rushing
[r395] tests/t14.scm tests lib/pair.scm
2010-02-24 18:07 rushing
[r394] backend.py insn_nvcase(): based on the code from 2008
2010-02-24 18:06 rushing
[r393] cps.py support vcase/pvcase/nvcase insns
2010-02-24 18:05 rushing
[r392] parse/gen_parser.py record an idea in the comments
2010-02-24 18:05 rushing
[r391] parse/lexer.py gen lexstep.scm
2010-02-24 18:04 rushing
[r390] header.c read_header(): make clang happier
2010-02-24 18:04 rushing
[r389] lib/pair.scm redone with a datatype declaration
2010-02-24 18:03 rushing
[r388] lib/frb.scm redone with a datatype declaration
2010-02-24 18:03 rushing
[r387] lib/core.scm maybe: normal variant/datatype version
2010-02-24 18:03 rushing
[r386] lib/string.scm string->list(): use normal variant version of lists
2010-02-24 18:02 rushing
[r385] lib/symbol.scm use normal variant type rather than poly variants
2010-02-24 18:01 rushing
[r384] pxll.h ctype.h
2010-02-24 18:01 rushing
[r383] itypes.py t_predicate.__repr__(): cleanersum(): not a row type, take *args insteaddatatype: holds info about a datatype declaration
2010-02-24 18:00 rushing
[r382] transform.py add_constructors(): add datatype constructors to the top-level fixexp_function(): remove support for ':type' in the function name [to allow datatype:constructor syntax]expand_vcase(): catch and rename vcase depending on which syntax it usesexpand_pvcase():expand_nvcase(): 'normal' variant caseexpand_datatype(): record the datatype along with info needed to build constructors
2010-02-24 17:56 rushing
[r381] solver.py check_constraint(): another sanity check: ensure that all tvars are properly boundconstraint_generator.__init__(): pass in context rather than the scc graphconstraint_generator.go(): call check_constraint()gen_vcase(): distinguish pvcase / nvcasegen_nvcase(): handle 'normal' vcaseunifier.simplify(): renamed also -> types don't unname tvars that are in types!instantiate_scheme(): made a global funsolver.instantiate_constraint(): option 'generalize' to avoid universal quantification, for the value restriction.lookup_special_names(): implement %dtcon and %nvget
2010-02-24 17:42 rushing
[r380] nodes.py scheme_string(), to_scheme(): removed. we're not much like scheme any more. maybe write a general 'unparse' routine later...def parse_type(): pass in a dict of known tvarswalker.__init__(): pass in context so that so walk_exp() can record datatype declshandle pvcase/nvcase
2010-02-24 17:39 rushing
[r379] analyze.py analyze(): removed 'as scheme' printtransform_0_primapp(): vcase => pvcasetransform_vcase,transform_pvcase,transform_nvcase(): handle both polymorphic and normal variantsis_varargs(): deprecatedinline_application(): disable arg-used-only-once inlining, this needs more analysis to know when it's safe to use. special-case nvget when inliningsafe_nvget_inline(): is this an nvget that's safe to inline?
2010-02-24 17:34 rushing
[r378] compile.py context.datatypes: global set of datatype decls
2010-02-04 00:09 rushing
[r377] parse/README initial readme for the convoluted process of generating a parser
2010-02-04 00:03 rushing
[r376] tests/t_bad_inline.scm demonstrates the problem with inlining varref args that are assigned to [fixed now]
2010-02-04 00:02 rushing
[r375] tests/t_rsum_0.scm exploring the overly powerful nature of row sums.thinking about ditching them, too much pain for not enough gain.
2010-02-03 23:59 rushing
[r374] parse/extract.py output vectors rather than lists, trying to make the typing phase go faster.
2010-02-03 23:59 rushing
[r373] tests/t20.scm added the LR engine - seems to work.
2010-02-03 23:58 rushing
[r372] nodes.py node.__init__(): automatically call 'fix_attribute_names()' 'inline' attr - for the waddell/dybvig inlinerrename(): update the <refs> and <assigns> slots
2010-02-03 23:56 rushing
[r371] analyze.py inline_application(): several problems fixed. 1) if a formal or a 'varref' arg is assigned to, force it to be a 'complex' inline rather than a 'simple' one. 2) commented out the 'complex arg used only once' hack. this was breaking tests/t_stack.scm, obvious in retrospect.substitute(): this was completely failing with varset nodes, fixed.
2010-02-03 23:52 rushing
[r370] compile.py context.nvariant_offset: addition of EMPTY_VECTOR type bumped the position of TC_NIL
2010-01-22 23:11 rushing
[r369] inliner.py waddell & dybvig's inliner, not finished.
2009-12-13 21:30 rushing
[r368] tests/t_empty_vector.scm make sure empty vectors (i.e., TC_EMPTY_VECTOR) work
2009-12-13 21:29 rushing
[r367] compile.py urgh
2009-12-13 21:29 rushing
[r366] compile.py raise the recursion limit whenever the module is imported
2009-12-13 21:28 rushing
[r365] backend.py insn_primop(): support empty vectors by emitting TC_EMPTY_VECTOR when appropriate.
2009-12-13 21:27 rushing
[r364] tests/t20.scm don't crash if no arg is given on the command line (i.e., run_tests.py)
2009-12-13 21:25 rushing
[r363] header.c dump_object(): print TC_EMPTY_VECTOR as #()
2009-12-13 21:25 rushing
[r362] pxll.h TC_EMPTY_VECTOR: new immediate type to represent the empty vector
2009-12-13 21:24 rushing
[r361] compile.py raise the system recursion limit, cps.py is triggering overflow of the default limit of 1000.
2009-12-13 01:08 rushing
[r360] parse/grammar full python grammar, hope to use it soon.
2009-12-13 01:05 rushing
[r359] tests/t28.scm vector and list literals
2009-12-13 01:02 rushing
[r358] tests/t42.scm removed prototype code
2009-12-13 01:02 rushing
[r357] tests/t42.scm testing new generator code
2009-12-13 01:01 rushing
[r356] tests/t21.scm moved stuff from the test into lib/io.scm.
2009-12-13 01:01 rushing
[r355] tests/t20.scm got token filtering/synthesis working, still not running through the parser though. had to take a break to make the solver faster, it absolutely fell down when hit with parser tables.
2009-12-13 00:59 rushing
[r354] backend.py guess_record_type(): cast tuple to list so 'index' will work on it. [tuple.index() appears in python 2.6]insn_primop(): %make-tuple: handle empty vectors (I hope).c_string(): work harder to emit safe strings for C. Unfortunately, we can't use hex encoding since it generalizes to >2 chars (i.e., >1 byte), and thus unpredictably following one with legal digit chars blows up. use octal instead.
2009-12-13 00:56 rushing
[r353] cps.py remove debug print
2009-12-13 00:56 rushing
[r352] parse/meta.py removed unneeded right precedencecorrectly emit 'optional'
2009-12-13 00:55 rushing
[r351] parse/gen_python_parser.py build/emit the irken lexer and parser all from this one file.
2009-12-13 00:54 rushing
[r350] parse/extract.py given the parser data from Parsing.py LR(1), emit some irken code to implement it.
2009-12-13 00:54 rushing
[r349] parse/gen_parser.py handle OPTIONAL in addition to STAR and PLUS.name_literal(): give a default name to literals that aren't passed in.emit_python(): default to right precedence on everything. seems like a bad idea.
2009-12-13 00:52 rushing
[r348] parse/regular.py CONCAT: inadvertently given the same string rep as PLUS. no biggie.
2009-12-13 00:51 rushing
[r347] parse/lexer.py read(): a simple engine to test a lexer before compiling it.gen_scheme_code(): removedfind_sink(): sometimes the sink state is not the second, use this to find it.gen_irken(): send output to a file rather than stdout
2009-12-13 00:49 rushing
[r346] lib/pair.scm use a <cons> function to lower the workload on the type solver
2009-12-13 00:49 rushing
[r345] lib/frb.scm #parens rearranged
2009-12-13 00:48 rushing
[r344] lib/core.scm eq?(): type is ('a 'a -> bool), not, ('a 'b -> bool)putcc(): type is (cont 'a -> 'b) (confirmed from "Compiling with Continuations").make-generator(): rewritten to use getcc/putcc directly. still considered an experiment.
2009-12-13 00:46 rushing
[r343] lib/io.scm still 'modernizing' this file using the new type system
2009-12-13 00:45 rushing
[r342] lib/string.scm string-ref: cast the string to 'unsigned char' to avoid negative char values.string-join: removed. nary funs unlikely any time soon.
2009-12-13 00:44 rushing
[r341] itypes.py t_var: default <in_u> to Falset_predicate.__repr__(): wrap parens around arrow types to make it easier to understand complex function types
2009-12-13 00:42 rushing
[r340] transform.py expand_case(): %eq? -> eq?build_literal(): call a function <cons> rather than inlining :cons, which leads to 'insanely detailed types'. [this probably won't help because the inliner will just undo it, sigh]
2009-12-13 00:40 rushing
[r339] solver.py debug print removed
2009-12-13 00:39 rushing
[r338] solver.py Lots more work to speed up the solver when it's fed large initialized data structures.s_let.counter: count the number of variables that are moved into this letmulti.__init__():multi.free(): compute free variables only when needed.get_compress_key(): used to compress the set of equations in the unifier. experimentally this actually slows things down, so not used.unifier.split(): don't cast sz.vars to a set, it already is one. (avoids a copy) call eq.free() to compute it on the flyunifier.simplify(): avoid another cast to set()solver.__init__(): try_unname - a flag to tell us to try to shrink the current s_let()solver.solve(): if try_unname is set, call do_extra_unname(). use difference_update() to update s_let.vars directly rather than copyingsolver.move_exists(): keep a running counter of the number of variables attached to this s_let(), every 1000 set the try_unname() flag.solver.do_extra_unname(): When solving for a large data structure, the normal split/unname isn't called until it's all over. Even relatively small data structures run into the tens of thousands of tvars. ftv(): no longer needed
2009-12-13 00:26 rushing
[r337] nodes.py tyop
2009-12-13 00:26 rushing
[r336] compile.py context.variant_labels: bit of a hack - pre-define :cons and :nil so they'll match the definitions in pxll.h. This way 'lists' will print out via header.c:dump_object() correctly.
2009-12-11 02:48 rushing
[r335] solver.py unifier.split(): be careful to add eqs to u2 only *after* they've been removed from u, since the act of removing them will blow away their 'in_u' attr.
2009-12-11 01:53 rushing
[r334] solver.py unifier: removed 'self.vars' map, went back to using a tvar attribute, 'in_u'. this has sped up type solving a *lot*, but still not enough to handle the new torture case (predefined variant data structures for the parser). merge(): removed u3: all references removedremember_variant_label(): adjust for the predefined 'negative' entries
2009-12-03 02:35 rushing
[r333] backend.py %vget primop
2009-12-03 02:35 rushing
[r332] run_tests.py don't optimize, we want the tests to run fast
2009-12-03 02:35 rushing
[r331] cps.py compile_primapp(): handle %vget prim
2009-12-03 02:34 rushing
[r330] tests/t11.scm test unary variants
2009-12-03 02:34 rushing
[r329] solver.py type the new %vget primapp
2009-12-03 02:34 rushing
[r328] analyze.py transform_0_let_splat():find_vcase_label_type(): removedtransform_vcase(): use a new primop '%vget' rather than %%cexp (which is harder to type)
2009-12-03 02:32 rushing
[r327] compile.py added the '-tt' (type twice) command line switch
2009-12-02 21:50 rushing
[r326] solver.py cleanups.decode():renumber(): removed
2009-12-02 21:45 rushing
[r325] solver.py find_records(): removed
2009-12-02 21:38 rushing
[r324] tests/t36.scm, tests/t37.scm, tests/t38.scm, tests/t39.scm new tests
2009-12-02 21:37 rushing
[r323] tests/f1.scm, tests/f2.scm, tests/f3.scm, tests/f4.scm, tests/t21.scm, tests/t28.scm, tests/t31.scm, tests/t32.scm, tests/t33.scm, tests/t34.scm, tests/t8.scm new tests, renamed tests
2009-12-02 21:37 rushing
[r322] backend.py insn_vcase(): simplified using get_safe_typecode - necessary because vcase/else may unpredictably refer to nullary constructors ("unit types"), calling GET_TYPECODE(*p) on an immediate data type.
2009-12-02 21:35 rushing
[r321] run_tests.py support tests starting with the letter 'f' that expect to fail
2009-12-02 21:34 rushing
[r320] cps.py dead code
2009-12-02 21:34 rushing
[r319] pxll.h get_safe_typecode(): inline function - used by vcase
2009-12-02 21:32 rushing
[r318] transform.py use constant bindings for symbolsbuild_vector(): call expand_exp() on the whole result rather than piecemealbuild_literal(): fix the list construction stuff to use :cons/:nil
2009-12-02 21:29 rushing
[r317] solver.py some attempts to speed things upclass s_let: keep <vars> in a setgen_vcase(): support else clausesunifier.add(): speed up testssolve(): u3 stuff isn't neededinstantiate_constraint(): correctly handle eqs with >1 variablelookup(): speed up logictyper.go(): find_records(): to be removed
2009-12-02 21:22 rushing
[r316] analyze.py transform_vcase(): implement 'else' clauses
2009-11-30 04:52 rushing
[r315] tests/t22.scm, tests/t23.scm, tests/t24.scm, tests/t25.scm, tests/t26.scm, tests/t27.scm, tests/t28.scm, tests/t29.scm, tests/t30.scm, tests/t31.scm, tests/t32.scm, tests/t33.scm, tests/t34.scm, tests/t35.scm new tests
2009-11-30 04:51 rushing
[r314] run_tests.py in progress - run everything in tests/*.scm [still need to be able to check/compare output]
2009-11-30 04:50 rushing
[r313] tests/t9.scm *** empty log message ***
2009-11-30 04:50 rushing
[r312] tests/t8.scm note that this test currently fails
2009-11-30 04:50 rushing
[r311] tests/t7.scm note that this test currently fails
2009-11-30 04:49 rushing
[r310] tests/t4.scm *** empty log message ***
2009-11-30 04:49 rushing
[r309] tests/t1.scm renamed a variable
2009-11-30 04:49 rushing
[r308] tests/t19.scm uncommented lots of the test
2009-11-30 04:48 rushing
[r307] tests/t16.scm introduced a dependency on lib/pair.scm somehow
2009-11-30 04:45 rushing
[r306] tests/t21.scm moving much further along to having a working io library
2009-11-30 04:45 rushing
[r305] tests/t11.scm give a good workout to variant case, including don't-care binding, inlining, etc..
2009-11-30 04:44 rushing
[r304] backend.py write_header(): support cincludedone(): records2, labels2 (these need to be cleaned up)guess_record_type(): try to disambiguate a row typeinsn_primop(): support %record-setinsn_vcase(): use original arity to compute tag valuesinsn_fetch_const(): vestigial, removed
2009-11-30 04:41 rushing
[r303] cps.py compile_primapp(): support %rsetget_record_tag(): moved from analysis.py to herecompile_record_literal(): nasty bug - forgot to shift the record tagcompile_record_extension(): distinguish between record-set! and record extensiongen_lit(): .type->.ltype
2009-11-30 04:37 rushing
[r302] header.c dump_object(): catch #\eof and print it out neatly
2009-11-30 04:36 rushing
[r301] lib/core.scm error(): just jump to Lreturn rather than calling abort()
2009-11-30 04:36 rushing
[r300] lib/io.scm open(): call zero-terminate on path arg
2009-11-30 04:35 rushing
[r299] lib/string.scm zero-terminate()
2009-11-30 04:35 rushing
[r298] itypes.py get_record_sig(): move to solver.py
2009-11-30 04:35 rushing
[r297] transform.py expand__percentvcase(): enable the use of low-level %vcase prim [used in at least one of the tests]expand_cinclude(): new special formexpand_set_bang(): new %rset translationexpand_begin(): merge sequences when possible
2009-11-30 04:32 rushing
[r296] solver.py major rewrite: finally understand S-LETALL, implemented. stop 'rendering' type schemes into human form just so they'll get chopped back up into tiny pieces again when instantiated.gen_cexp(): handle non-arrow cexp types (e.g., for C constants)gen_vcase(): have to handle this node type in order to re-type after analysis/inliningunifier(): a dict to hold memoized decode results reworked the way we store equations and variables: using mutation on the '.eq' slot was causing a *lot* of confusion, so now we have a map from var->eq.is_free(): no longer neededsplit(): reworkeddecode(): switch from '.eq' slot to var->eq mapsanity(): check invariants in new data structuresimplify(): done much more carefully!prune(): new function to slim a constraint by removing unnecessary equations [why this should be necessary is still a mystery to me]find_free(): find the free variables of a unifier/substreverse_graph(): to help traverse the equation graph bottom up, used by do_letall()do_letall(): implement the S-LETALL rulesolver(): separate the notion of 'verbose' from 'single-stepping'solve(): temporarily disabled the catching of type errors s-solve-id: let self.lookup() instantiate the constraint s-letall: finally implemented this rule. call move_exists() to move any 'older' vars up and out s-pop-let: actually lower the rank (ugh, how embarrassing)build_type_schemes(): removedinstantiate_scheme(): instantiate the schemes returned by lookup_special_names()instantiate_constraint(): instantiate a constraint-based schemelookup_special_names(): added %rset sum->rsum, product->rproductprint_solution(): removedget_record_sig(): redone, probably be able to remove or simplify thistyper.decode(): much simpler decoding using the node.tv variable.
2009-11-30 04:17 rushing
[r295] nodes.py node.pprint(): printing the type out again (to help me debug solver rewrite)literal: use 'ltype' for literal type rather than 'type', which breaks things later onmake_tuple: use 'ttype' for tuple type, ...
2009-11-30 04:15 rushing
[r294] analyze.py transform_0_let_splat():transform_1_application(): disabled, to be removedtransform_vcase(): delay variant type lookup emit fresh tvars for cexp resultsfind_vcase_label_type(): to be removed
2009-11-30 04:11 rushing
[r293] lisp_reader.py read(): support #\eof
2009-11-30 04:11 rushing
[r292] compile.py compile_file(): step_solver arg [need to get compiler options into their own class] added a second type solver pass after inlining
2009-11-17 06:03 rushing
[r291] tests/t21.scm moved code into lib/string.scm, this test is... simpler.
2009-11-17 05:59 rushing
[r290] lib/string.scm sys:argv: collect into a vector instead
2009-11-17 05:43 rushing
[r289] lib/pair.scm updated for the new type system
2009-11-17 05:42 rushing
[r288] tests/t21.scm test vectors and argc/argv
2009-11-17 05:40 rushing
[r287] lib/frb2.scm, lib/frb3.scm, lib/list.scm, lib/primops.scm, lib/uobj.scm, lib/vector.scm, parse/lex4 lose some old files
2009-11-17 05:39 rushing
[r286] backend.py insn_primop(): implement %make-vector
2009-11-17 05:39 rushing
[r285] cps.py compile_primapp(): handle %make-vector
2009-11-17 05:39 rushing
[r284] header.c vm(): include 'int i' for a loop counter
2009-11-17 05:38 rushing
[r283] solver.py lookup_special_names(): handle %make-vector
2009-11-17 03:44 rushing
[r282] lib/pair.scm redone for new type system
2009-11-17 01:23 rushing
[r281] solver.py solver::solve(): oops, really new syntax in the error handler
2009-11-17 01:14 rushing
[r280] tests/t19.scm test the symbol table
2009-11-17 01:14 rushing
[r279] tests/t20.scm bringing the lexer back to life
2009-11-17 01:13 rushing
[r278] parse/gen_parser.py convert a high-level grammar into a Parsing.py-style grammar
2009-11-17 01:13 rushing
[r277] parse/meta.py parser for python-style meta-grammar
2009-11-17 01:12 rushing
[r276] lib/core.scm updated for the new type system.printn():print(): use proper type variableszero?(): new funerror(): return any type
2009-11-17 01:11 rushing
[r275] lib/io.scm updated for the new type system
2009-11-17 01:10 rushing
[r274] lib/string.scm string-compare(): trying a little to avoid consing, a little more work yet.list->string():string->list(): updated for the new type system
2009-11-17 01:09 rushing
[r273] lib/symbol.scm updated for the new type system
2009-11-17 01:09 rushing
[r272] itypes.py t_symbol:t_continuation: new base typesrproduct():rsum(): special product and sum types just for use with rows
2009-11-17 01:08 rushing
[r271] tests/t18.scm test out array-ref[syntax]
2009-11-17 01:07 rushing
[r270] tests/t16.scm recast the test to use the code in lib/frb.scm, add a few more tests
2009-11-17 01:06 rushing
[r269] transform.py expand__percentvcon(): only tack on the arity onceexpand__percent_percentmake_tuple(): looks like this avoidance is no longer needed
2009-11-17 01:05 rushing
[r268] solver.py parse_cexp_type(): p(): handle '/raw' typesgen_primapp(): catch single-value %vcon constructors, don't wrap the single type in product().gen_make_tuple(): handle this node typesolve(): catch TypeErrors and try to do something reasonable with themprint_type_error(): dump the stack and decode the two types for the user. still sucks but it's better than pdb.lookup_special_names(): use 'rproduct' instead of 'product' for row types... no point in confusing the two rather different types. handle %%array-ref
2009-11-17 01:00 rushing
[r267] analyze.py transform_vcase(): bugs in don't-care vars: bug caused by editing the 'success' node in place bug - don't set alt_formals until success.formals has been editedfind_vcase_label_type(): try harder to find the type of a vcase value variable, for some reason it alternates between being in the varref or the vardef node (in which case we have to pull it out of the c_forall)
2009-11-16 02:53 rushing
[r266] lib/frb.scm rewritten for the new type system (based on tests/t16.scm)
2009-11-14 22:32 rushing
[r265] parse/lexer.py updated for new type system
2009-11-14 00:48 rushing
[r264] parse/Parsing.py Jason Evan's LR(1) parser generator
2009-11-14 00:39 rushing
[r263] tests/t18.scm testing vectors
2009-11-14 00:39 rushing
[r262] tests/t16.scm removed some comments
2009-11-14 00:39 rushing
[r261] cps.py compile_primapp(): %%vector-literal => %vector-literal
2009-11-14 00:37 rushing
[r260] itypes.py vector() type constructor
2009-11-14 00:37 rushing
[r259] transform.py build_vector(): put %vector-literal arity in its name
2009-11-14 00:37 rushing
[r258] solver.py type vector literals
2009-11-10 20:05 rushing
[r257] tests/t17.scm random c interface testing
2009-11-10 20:05 rushing
[r256] tests/t16.scm continuing to port the test case. pretty much done, just need the generator.
2009-11-10 20:04 rushing
[r255] header.c comment changes
2009-11-10 20:03 rushing
[r254] transform.py expand_vcase(): correctly handle sequence bodies
2009-11-10 20:02 rushing
[r253] solver.py constraint_generator: gen_varset()attempts to speed things up: commented out all the self.dprint() calls unifier.decode(): watch for cycles involving tvars only solver.solve(): c=true;s=s_let:compare ranks rather than calling u.is_free()
2009-11-10 20:00 rushing
[r252] compile.py add code to profile the solverremoved read_string()removed inline testsdefault to llvm-gcc only on Darwin
2009-11-09 20:46 rushing
[r251] tests/t4.scm *** empty log message ***
2009-11-09 02:30 rushing
[r250] tests/t16.scm okasaki's red/black trees, first attempt using new variants..
2009-11-09 02:30 rushing
[r249] backend.py insn_vcase(): catch a vcase for a non-existent vcon label, report it with a nicer error
2009-11-09 02:28 rushing
[r248] solver.py unifier.unify_rows(): oops, send rdefault==rdefault through the normal decompose process
2009-11-09 02:27 rushing
[r247] nodes.py node.__repr__(): commented out the code that prints the type of every node [too much noise in the debug dump for complex types]rename_variables.rename(): removed special case code for names starting with '&' and '_'
2009-11-09 02:26 rushing
[r246] analyze.py transform_vcase(): filter out don't-care bindings correctly pull out the type of a variant labelfind_vcase_label_type(): helper for that
2009-11-07 22:02 rushing
[r245] tests/t15.scm test cexp type decls
2009-11-07 22:02 rushing
[r244] backend.py wrap_in(): handle/ignore type vars in cexp types
2009-11-07 22:01 rushing
[r243] itypes.py moved parse_cexp_type to solver.py
2009-11-07 22:01 rushing
[r242] solver.py parse_cexp_type(): moved from itypes.py to here fixed to handle type variableslookup_special_names(): updated some of the greek types in the comments to be more accurate.
2009-11-07 21:59 rushing
[r241] nodes.py parse_type(): support type variables in cexp types complain about inconsistent arrow types
2009-11-07 01:11 rushing
[r240] LICENSE.txt forgot to include this. 8^)
2009-11-07 00:33 rushing
[r239] solver.py debug prints only if self.verbose
2009-11-07 00:33 rushing
[r238] graph.py commented out debug print
2009-11-07 00:04 rushing
[r237] typing.py Yeah! Removed typing.py!
2009-11-07 00:03 rushing
[r236] transform.py removed references to old typing module
2009-11-07 00:02 rushing
[r235] compile.py removed references to old typing module
2009-11-07 00:02 rushing
[r234] nodes.py removed references to old typing module
2009-11-06 23:57 rushing
[r233] tests/t14.scm test/demonstration of lisp lists and associated functions, built with polymorphic variants
2009-11-06 23:56 rushing
[r232] tests/t13.scm demonstrate re-using a variant label for completely different purposes
2009-11-06 23:56 rushing
[r231] backend.py insn_vcase(): insns.params now holds alt_formals, not just types collect (index, tag) pairs use <tag> in each case statement (rather than index)insn_new_vector() => insn_new_tuple()insn_store_vec() => insn_store_tuple()insn_store_tuple(): insns.params now contains an offset into the tupleinsn_store_env(): removedinsn_move(): vestigial, removed
2009-11-06 23:53 rushing
[r230] cps.py gen_vcase(): take the alt_formals list rather than just typescompile_tuple_rands() => compile_store_rands(): coalesce the jobs of these two functions, pass in an offset parameter to handle the difference.compile_vector_rands(): use compile_store_rands()compile_record_literal(): use compile_store_rands() and cut down on register abusegen_store_env(): removed (now done with store_tuple)gen_store_vec() => gen_store_tuple() renamed, generalizedgen_move():gen_save(): vestigial, removedfind_allocation(): catch new_tuple insns
2009-11-06 23:50 rushing
[r229] itypes.py t_base: store the name of the type in an attribute
2009-11-06 23:49 rushing
[r228] transform.py expand_exp(): oops, put that call to build_record() back in there
2009-11-06 23:48 rushing
[r227] solver.py unifier.decode(): watch out for cyclical typesfind_records(): navigate the new universe of other product and row types don't forget to examine top_tv, in case the whole program returns a record. [this function needs to be removed and another way found to collect unique record signatures]
2009-11-06 05:45 rushing
[r226] tests/t12.scm test of apply-lambda inlining
2009-11-06 05:45 rushing
[r225] tests/t11.scm variable-arity variant constructors
2009-11-06 05:45 rushing
[r224] tests/t10.scm variable-arity variant constructors
2009-11-06 05:44 rushing
[r223] tests/t9.scm test the speed of record and variant creation
2009-11-06 05:44 rushing
[r222] backend.py insn_vcase(): changed 'unit' to product()
2009-11-06 05:43 rushing
[r221] cps.py %vcon/label/arity has arity embedded in the name
2009-11-06 05:43 rushing
[r220] itypes.py product(): allow args other than rows
2009-11-06 05:42 rushing
[r219] transform.py fixed the auto-lookup of expansion methods for names with encoded metadata, like %vcon/label/arity.expand__percentvcon(): encode arity into name [feeling prologish]expand_quote(): removed old list/nil code
2009-11-06 05:41 rushing
[r218] solver.py variable-arity variant constructors and variant filters
2009-11-06 05:40 rushing
[r217] analyze.py variable-arity variant constructors
2009-11-06 02:17 rushing
[r216] notes/log.txt *** empty log message ***
2009-11-06 00:20 rushing
[r215] tests/t6.scm use zero? test (makes no speed difference)
2009-11-05 23:47 rushing
[r214] backend.py insn_test(): I got a little too aggressive inlining cexp tests, had to put back the PXLL_IS_TRUE/wrap_out() stuff. at least with llvm-gcc it generates the same code.
2009-11-05 23:45 rushing
[r213] analyze.py turn off opt_apply_lambda_to_lettransform_vcase(): don't use let - instead apply lambdas, this will allow the inliner to do the right thinginline_application(): inline the application of a lambda
2009-11-05 22:38 rushing
[r212] tests/t8.scm playing with new vcase syntax
2009-11-05 22:37 rushing
[r211] tests/t7.scm *** empty log message ***
2009-11-05 22:37 rushing
[r210] backend.py removed typecase
2009-11-05 22:37 rushing
[r209] cps.py typecase => vcase
2009-11-05 22:36 rushing
[r208] transform.py expand_colon(): turn :label into (%vcon/label <value>)removed typecaseexpand_vcase(): first attempt
2009-11-05 22:35 rushing
[r207] solver.py renamed %vmake to %vfail, fixed its type
2009-11-05 22:34 rushing
[r206] nodes.py removed typecase
2009-11-05 22:34 rushing
[r205] analyze.py removed typecasetransform_vcase(): catch the impossible %vfail, emit no code
2009-11-05 22:16 rushing
[r204] lisp_reader.py colon variant constructor syntax
2009-11-05 05:23 rushing
[r203] tests/t8.scm testing variants - with vcase syntax
2009-11-05 05:23 rushing
[r202] tests/t7.scm testing variants (pre-vcase/case syntax)
2009-11-05 05:22 rushing
[r201] notes/log.txt *** empty log message ***
2009-11-05 05:18 rushing
[r200] backend.py insn_vcase(): slightly modified version of typecase still more work to do (support products for example)
2009-11-05 05:17 rushing
[r199] cps.py handle %vcon primappcompile_vcase(): similar to compile_typecase()gen_vcase(): new insnflatten(): handle vcase
2009-11-05 05:16 rushing
[r198] transform.py expand_vcase(): placeholder, will be rewritten
2009-11-05 05:15 rushing
[r197] nodes.py vcase node type (will obsolete typecase)
2009-11-05 05:14 rushing
[r196] analyze.py transform_0_primapp(): catch %vcase, call transform_vcase()transform_1_set() removed
2009-11-04 19:36 rushing
[r195] solver.py constraint_generator: gen(): oops, forgot to return value
2009-11-04 19:23 rushing
[r194] solver.py constraint_generator: dynamic node type lookup (i.e., gen_varref() rather than 'if t.is_a (...)'solver: pass context rather than context.datatypes renamed %vextend to %vcon remember_variant_label()
2009-11-04 02:18 rushing
[r193] compile.py using llvm-gcc
2009-11-04 02:17 rushing
[r192] tests/t6.scm commentary on the different output given by gcc-llvm
2009-11-04 02:11 rushing
[r191] tests/t6.scm testing integer loop performance (still looking good 8^)
2009-11-04 02:10 rushing
[r190] tests/t5.scm testing simple conditionals
2009-11-04 02:10 rushing
[r189] tests/tak20.scm the tak classic
2009-11-04 02:10 rushing
[r188] backend.py done(): removed pointless verbose commentinsn_test(): support 'simple conditionals' again by inlining a cexp test rather than casting to boolean, moving to a register and testing that register against PXLL_FALSE
2009-11-04 02:08 rushing
[r187] cps.py make 'simple conditionals' work again
2009-11-04 02:05 rushing
[r186] header.c removed verify() function
2009-11-04 01:10 rushing
[r185] backend.py commented out more verify calls
2009-11-04 00:54 rushing
[r184] irken.html updated/modified
2009-11-04 00:33 rushing
[r183] itypes.py unit type
2009-11-04 00:23 rushing
[r182] tests, tests/t0.scm, tests/t1.scm, tests/t2.scm, tests/t3.scm, tests/t4.scm redoing tests
2009-11-04 00:22 rushing
[r181] tests redoing tests
2009-11-04 00:18 rushing
[r180] backend.py pass in compile contextremoved obsolete node.typecheck supportemit_typecheck(): removeddone(): emit the lookup_field() functioninsn_primop(): implement %record-get, %extend-tuplecommented out some self.verify() calls (to be removed)removed support for safety>3 calls to verify
2009-11-04 00:15 rushing
[r179] cps.py pass in compile contextcompile %raccess, %rextendcompile_record_literal():compile_record_extension(): turn these into %make-tuple calls
2009-11-04 00:14 rushing
[r178] header.c lookup_field() prototypevarref_safe(), varset_safe(): removed
2009-11-04 00:14 rushing
[r177] itypes.py t_str => t_stringt_undefinedget_record_sig(): given a product (rlabel (...)), extract the signature of the record (i.e., its sorted list of labels)
2009-11-04 00:12 rushing
[r176] transform.py transformer: pass in compile context removed obsolete klass support support record literals, record access build_record(): transform a record literal into a set of primops expand_class(): removed
2009-11-04 00:10 rushing
[r175] solver.py removed unused fun
2009-11-04 00:10 rushing
[r174] solver.py updating the comments a bitgenerator.gen(): attach tvar at each node treat primapp like app n-ary let_splat handle sequence map literals through base_typesunifier.split(): replace compress local fun with self.decode() when splitting out a var, see if it has a node attached and 'solve' its type using self.decode()solver: pass in context.datatypes instantiate(): treat ints like tvars lookup_special_names(): put this back - record ops are primapps nowtyper: pass in context after running the solver, decode any remaining tvars find_records(): scan all types for complete record definitions, remember them along with the set of all labels
2009-11-04 00:03 rushing
[r173] nodes.py removed obsolete typecheck attrremoved obsolete 'klass' stuffrename_variables() now takes a datatypes arg [to avoid renaming constructors]
2009-11-04 00:02 rushing
[r172] analyze.py undo temporary hacks for row-related prims
2009-11-04 00:02 rushing
[r171] analyze.py pass the compilation context downcomment out the get/set transformshandle row-related primitives [fixme]
2009-11-04 00:00 rushing
[r170] notes/rows.txt *** empty log message ***
2009-11-02 01:43 rushing
[r169] lisp_reader.py support for record literals
2009-11-02 01:43 rushing
[r168] compile.py new 'context' class to hold information across all the various passes
2009-10-28 23:36 rushing
[r167] notes/log.txt sp
2009-10-28 23:32 rushing
[r166] t1.c, test.image, test.img first release in a while!
2009-10-28 23:32 rushing
[r165] notes/log.txt *** empty log message ***
2009-10-28 23:31 rushing
[r164] tests/tc1.scm *** empty log message ***
2009-10-28 23:31 rushing
[r163] tests/tc0.scm *** empty log message ***
2009-10-28 23:29 rushing
[r162] itypes.py types
2009-10-28 23:28 rushing
[r161] notes/classes.txt thoughts on classes and a circular dependency
2009-10-28 23:27 rushing
[r160] tests/t0.scm use %%cexp
2009-10-28 23:26 rushing
[r159] solver.py interactive testing only
2009-10-28 23:26 rushing
[r158] compile.py new constraint-based type inference engine
2009-10-28 23:25 rushing
[r157] gc.c gcc warnings
2009-10-28 23:25 rushing
[r156] header.c gcc warnings
2009-10-28 22:57 rushing
[r155] solver.py removed unneeded test cases and prim_env support
2009-10-28 22:56 rushing
[r154] solver.py first version that limps along and types irken nodes. still no rows.
2009-10-28 22:55 rushing
[r153] graph.py dependency graph and fix-reordering code from old HM typing.py module
2009-10-19 21:15 rushing
[r152] cps.py renamed lambda_tree => nodes
2009-10-19 21:15 rushing
[r151] transform.py renamed lambda_tree => nodes
2009-10-19 21:15 rushing
[r150] typing.py renamed lambda_tree => nodes
2009-10-19 21:15 rushing
[r149] analyze.py renamed lambda_tree => nodes
2009-10-19 21:15 rushing
[r148] compile.py renamed lambda_tree => nodes
2009-10-19 21:14 rushing
[r147] tests/y.scm Y ask Y?
2009-10-19 21:01 rushing
[r146] lambda_tree.py, nodes.py renamed lambda_tree.py to nodes.py
2009-10-19 20:57 rushing
[r145] notes/log.txt *** empty log message ***
2009-10-19 20:52 rushing
[r144] README, lib/rbtree.scm, pottier_remy.py readme is html, removing unneeded files, preparing for new solver
2009-10-19 20:45 rushing
[r143] lib/core.scm clearing out mods in prep for the new type solver
2009-10-19 20:44 rushing
[r142] lib/io.scm clearing out mods in prep for the new type solver
2009-10-19 20:44 rushing
[r141] lib/string.scm clearing out mods in prep for the new type solver
2009-10-19 20:44 rushing
[r140] tests/product_0.scm clearing out mods in prep for the new type solver
2009-10-19 20:43 rushing
[r139] tests/io_0.scm clearing out mods in prep for the new type solver
2009-10-19 20:43 rushing
[r138] tests/t_callcc_2.scm clearing out mods in prep for the new type solver
2009-10-19 20:42 rushing
[r137] tests/step5.scm clearing out mods in prep for the new type solver
2009-10-19 20:42 rushing
[r136] typing.py clearing out mods in prep for the new type solver[this file will be going away! 8^)]
2009-10-19 20:23 rushing
[r135] analyze.py clearing out mods in prep for the new type solver
2009-10-19 20:22 rushing
[r134] lambda_tree.py class-related stuff I'm going to have to replace anyway
2009-10-19 20:22 rushing
[r133] lisp_reader.py support postfix array notation
2009-10-19 20:22 rushing
[r132] compile.py clearing out mods in prep for the new type solvertrying out llvm/clang
2009-10-19 20:21 rushing
[r131] cps.py clearing out mods in prep for the new type solver%%product-ref (will probably be replaced)
2009-10-19 20:20 rushing
[r130] parse/lexer.py table-based dfa using strings
2009-10-19 20:19 rushing
[r129] pxll.h clearing out mods in prep for the new type solver
2009-10-19 20:19 rushing
[r128] transform.py clearing out mods in prep for the new type solver
2009-10-19 20:18 rushing
[r127] solver.py checking in the pottier/remy solver
2009-09-30 21:35 rushing
[r126] README *** empty log message ***
2009-09-30 21:34 rushing
[r125] irken.html converted from README, updated
2009-07-19 23:30 rushing
[r124] pottier_remy.py simply typed lambda calculus, constraint generation. still need to write a solver (which here I think just means unify()).
2009-01-05 21:31 sam
[r123] notes/log.txt *** empty log message ***
2009-01-05 21:28 sam
[r122] cps.py pxll => irken
2009-01-05 21:27 sam
[r121] compile.py pxll => irken
2009-01-05 21:25 sam
[r120] a.out, test, test.c, typing.backup.py, typing.backup2.py, typing_attr_type.py removing these old files which don't belong in the repository.
2009-01-05 21:24 sam
[r119] tests/frb.scm got the generator code working, fixed the nth problem with polymorphism.renamed 'tree:' => 'node:'
2009-01-05 21:23 sam
[r118] cps.py pass verbose down into compiler and 'find_allocation()'
2009-01-05 21:22 sam
[r117] lib/core.scm added <pair> datatype
2009-01-05 21:22 sam
[r116] lib/symbol.scm use node: rather than tree: prefix for the frb
2009-01-05 21:21 sam
[r115] lib/frb.scm put identifying 'node:' prefix on many of the top-level funs [may formalize this syntax later with some kind of package/module/structure system]node:make-generator: in order to pass key/val pairs through the generator, we have to package them up using the <pair> constructor.
2009-01-05 21:19 sam
[r114] typing.py removed some outdated commentsproduct.__init__(): handle procedure typesinitial_type_environment(): pass <name> down to build_type_scheme()build_type_scheme(): when printing verbose info, print the name of the function we're schemeing for.
2009-01-05 21:17 sam
[r113] compile.py pass verbose down to cps.py
2008-12-22 21:38 sam
[r112] README *** empty log message ***
2008-12-22 21:29 sam
[r111] . renamed the project
2008-12-22 21:19 sam
[r110] test new user product datatypes
2008-12-22 21:19 sam
[r109] trying to grok why aliases are borked
2008-12-22 21:18 sam
[r108] new string tests
2008-12-22 21:18 sam
[r107] new io tests
2008-12-22 21:18 sam
[r106] updated for new typing
2008-12-22 21:16 sam
[r105] updated for new typing
2008-12-22 21:16 sam
[r104] updated for new typing
2008-12-22 21:16 sam
[r103] updated for new typing
2008-12-22 21:16 sam
[r102] updated for new typing
2008-12-22 21:15 sam
[r101] updated for new typing
2008-12-22 21:15 sam
[r100] hand-translated for the new typing [need to fix the lexer generator]
2008-12-22 21:15 sam
[r99] wrap_in(): support a '/raw' annotation to stop an object from getting auto-wrapped. (needed to access a string *object*)insn_primop(): make-tuple: handle dead constructors (maybe a bad idea?) TC_USEROBJ+tag: multiply tag by 4!insn_new_vector:insn_store_vec: used for large vector literals
2008-12-22 21:12 sam
[r98] compile_primapp(): if a vector literal has more than 5 elements, compile it as a sequence of assignments - otherwise a 100-element vector needs 100 registers to build!compile_vector_literal(): fill in a vector literal like it were a series of function arguments.analyze(): renamed to 'find_allocation()', and updated/fixed.walk_function(): handle the special case of 'typecase' (similar to 'test')
2008-12-22 21:09 sam
[r97] re-arranged to not place a limit on typecodes (i.e., user tags will continue past TC_USEROBJ and TC_USERIMM...)
2008-12-22 21:07 sam
[r96] dump_object(): don't forget to divide user tags by 4allocate(): calling check_heap() here is a Bad Idea, removed even this commented code to stop me from doing it.
2008-12-22 21:06 sam
[r95] updated for new typing
2008-12-22 21:05 sam
[r94] updated for new typing
2008-12-22 21:05 sam
[r93] updated for new typing
2008-12-22 21:04 sam
[r92] updated for new typing
2008-12-22 21:04 sam
[r91] updated for new typing
2008-12-22 21:03 sam
[r90] h-m version of lib/pair.scm ('pair' doesn't correctly describe this any more)
2008-12-22 21:02 sam
[r89] TC_USERIMM comment was wrongTC_LAST removed
2008-12-22 21:02 sam
[r88] datatype: support <product> datatypesnil: emit '() as (list/nil) rather than atom ('nil', 'nil')add_constants_and_classes(): prepending constants... now probably not needed due to typing.reorder_fix().
2008-12-22 21:00 sam
[r87] class product: support user-defined product types... optional name gen_constructor(): get_datatype_constructors(): new methodsget_type_variable(): wrapper for the constructor that makes these act more like 'pure' types (i.e., so they act more like the integers they used to be). I kept running into a problem where two tvars with the same serial were conflicting with each other.the_type_map: pulled out the bottom-most layer of the type environment. To me it seems much less confusing to have a separate data structure recording all the 'named' types.wildcard(): renamed to parse_cexp_type() support function types in %%cexp (needed this for call/cc)check_partition(): needed to suss out bugs in partition_fix()reorder_fix(): use the dependency graph to physically re-order the values in a fix... this was needed in order to correctly handle e.g. a top-level variable that needs to call a function to initialize itself... [in Scheme these are usually set to #f, but with typing we can't be so sloppy]
2008-12-22 20:53 sam
[r86] alpha_convert(): removed [now lives in lambda_tree.py]transform_1_fix(): coalesce cascading fix (just like we do with let_splat)check_args(): removed, no longer needed!various notes/reminders, this file still needs lots of updating work, mostly to get it to use the two graphs created by typing.
2008-12-22 20:51 sam
[r85] add_constructors(): commented out class code added support for product datatypes
2008-12-22 20:49 sam
[r84] pxll=>irkennew notes on The Plan
2008-12-22 20:29 sam
[r83] pxll=>irken
2008-12-22 20:29 sam
[r82] pxll=>irken
2008-12-18 01:57 sam
[r81] tests that should fail to typecheck
2008-12-18 01:55 sam
[r80] contrived example meant to have at least one strongly-connected component
2008-12-18 01:55 sam
[r79] testing the SCC algorithm using the graph from CLR
2008-12-18 01:54 sam
[r78] analyzer.__init__(): vars dict is now passed inanalyzer.add_constructors(): moved to lambda_tree.pytransform_0_typecase(): layout of alt_formals changed
2008-12-18 01:53 sam
[r77] cleaned up some outdated comments
2008-12-18 01:52 sam
[r76] flesh out this test of polymorphism and union types, by instantiating the red-black tree and several functions using a few different types.
2008-12-18 01:51 sam
[r75] expand_typecase(): the 'typecase' node was cleaned up a bit
2008-12-18 01:50 sam
[r74] Turns out that H-M polymorphism works by continually embedding letrecs inside of each other. Rather than force the user to do this, we compute what that embedded set would look like, and then emulate that *only for the purposes of typing*.rearranged the file more logicallytype_of(): fix: 'virtually' partition the fix into topologically-sorted groups of strongly-connected parts.build_dependency_graph(): build_call_graph(), moved from analyze.py to here.transpose(): make the transpose of a adjacency-matrix graphstrongly(): find the strongly-connected components of the graph using Kosaraju's algorithm. returns the components in topologically-sorted order.partition_fix(): using the SCC graph computed by strongly(), sort each set of functions/vars in a fix in dependency order.
2008-12-18 01:44 sam
[r73] moved alpha conversion and add_constuctors() into this file
2008-12-18 01:43 sam
[r72] compile_file(): pulled alpha conversion out of analyze.py, now runs *before* typing. pass the dict of all variables into analyze, which still needs it.
2008-12-13 04:05 sam
[r71] *** empty log message ***
2008-12-13 04:01 sam
[r70] *** empty log message ***
2008-12-13 04:00 sam
[r69] yet another version of function red-black trees, this time using algrebraic datatypes.
2008-12-13 03:59 sam
[r68] wrap_in(): automatically convert <pxll_string> to <char *>insn_typecase(): GET_TYPECODE() will *not* return zero for a pointer value, so use 'default' rather than 'case 0'.
2008-12-13 03:58 sam
[r67] compile_typecase(): use exp.vtype, not the (possibly tvar) value typenode.print_info(): typecase: print the assigned vtype
2008-12-13 03:57 sam
[r66] type_program(): don't bother with any of the code that tries to attach types to nodes... the type_of() wrapper seems to do a good enough job now.type_of(): typecase: type=>vtype in typecase node set vtype when sorting the alts, also sort the alt formals! push everything back into the node correctly
2008-12-13 03:53 sam
[r65] transform_0_typecase(): allow 'dontcare' bindings
2008-12-13 03:52 sam
[r64] typecase: changed the 'type' attr to 'vtype', less confusing
2008-12-13 03:52 sam
[r63] read(): allow #u for the 'undefined' type
2008-12-12 01:33 sam
[r62] *** empty log message ***
2008-12-12 01:31 sam
[r61] support unit typesinsn_typecase(): emit 1-2 switch clauses depending on the mix of unit and tuple types.
2008-12-12 01:30 sam
[r60] support unit types<typecase> insn needs to pass the type object down
2008-12-12 01:30 sam
[r59] support unit typesdump_object(): fall-through for TC_USERIMM+n
2008-12-12 01:29 sam
[r58] support unit typesnew #define for TC_USERIMM, user-defined immediate types (i.e., unit types) start there.
2008-12-12 01:29 sam
[r57] support unit types
2008-12-12 01:29 sam
[r56] support unit types
2008-12-12 01:28 sam
[r55] *** empty log message ***
2008-12-12 01:28 sam
[r54] support unit types
2008-12-11 02:57 sam
[r53] *** empty log message ***
2008-12-11 02:56 sam
[r52] updated to new syntax
2008-12-11 02:56 sam
[r51] updated to new syntax
2008-12-11 02:56 sam
[r50] new
2008-12-11 02:56 sam
[r49] new
2008-12-11 02:55 sam
[r48] compile_typecase(): change the name of the attr again!
2008-12-11 02:55 sam
[r47] typecase: new syntax - require name of type
2008-12-11 02:54 sam
[r46] class product: handle polymorphic productsclass union: handle polymorphic unionsinitial_type_environment(): build polymorphic datatypes correctly removed the 'nil' base typeunion.get_datatype_constructors(): just return the namesunify(): fail correctly when two base types differtype_of(): typecase: handle the new explicit-type syntaxinstantiate_type(): handle products & unions
2008-12-11 02:48 sam
[r45] compile_typecase(): typecase node attribute renamed
2008-12-11 02:48 sam
[r44] typecase: new syntax - require name of type
2008-12-10 23:20 sam
[r43] *** empty log message ***
2008-12-10 23:17 sam
[r42] compile_typecase(): typecase node attribute renamed
2008-12-10 23:17 sam
[r41] full algebraic datatypesexpand_datatype(): new syntaxexpand_typecase(): new syntax
2008-12-10 23:16 sam
[r40] full algebraic datatypesclass product: minimal support for a product typelookup_subst(): don't store None as the final value of a tvar. 8^)union.gen_constructor(): handle product sub-typesunify(): remember original values of t1&t2 for debugging watch for type aliases and dereference them [this might be the trick I need to do recursive types correctly]type_of(): union: unify using a product sub-type when appropriate typecase: unify using a product sub-type when appropriate
2008-12-10 23:05 sam
[r39] transform_0_typecase(): new 'variant' of typecase
2008-12-10 23:04 sam
[r38] full algebraic datatypestypecase: changed 'varname' attr to 'variant', though I think I need to undo it... I thought I had a way of avoiding the requirement that it be a varref.
2008-12-10 23:02 sam
[r37] first test of full algebraic datatypes
2008-12-10 01:46 sam
[r36] *** empty log message ***
2008-12-10 01:37 sam
[r35] *** empty log message ***
2008-12-10 01:37 sam
[r34] *** empty log message ***
2008-12-10 01:35 sam
[r33] insn_primop(): new variant of make-tuple supporting variant-type tagsinsn_typecase(): new insn, generates a switch
2008-12-10 01:34 sam
[r32] supporting union/sum datatypescompile_typecase(): new methodgen_typecase(): new method, insn typepretty_print(): special-case <typecase> along with <test> and <close>
2008-12-10 01:33 sam
[r31] dump_object(): handle TC_USEROBJ+ [this needs to be done to gc as well]
2008-12-10 01:32 sam
[r30] comment describing how variant types are tagged
2008-12-10 01:32 sam
[r29] expand_datatype(): new methodexpand_typecase(): new method
2008-12-10 01:30 sam
[r28] supporting union/sum datatypesclass union: union/sum type.type_of(): let_splat: extend the rib one var at a time application: handle rator_type of <union> typecase: unify itapply_tenv(): walk let* ribs backward
2008-12-10 01:25 sam
[r27] supporting union/sum datatypesanalyze(): call add_constructors if there are any datatype defnstransform(): added a 'stage' arg, to allow differents sets of transforms for different stages of optimization.add_constructors(): add union/sum constructorstransform_0_typecase(): hack to wrap typecase with a let that pulls out the (single) variant object. this is likely to go away when I switch to sum-of-products.
2008-12-10 01:22 sam
[r26] supporting union/sum datatypesnode.flag: boolean flag, used in analyze to stop infinite recursive transform. I'll probably end up removing this.constructor, typecase: new node types
2008-12-05 01:35 sam
[r25] type_of(): primapp/%%array-ref: return the array's item type, not the array type itself.
2008-12-05 01:33 sam
[r24] build_vector(): trying this: allow expressions rather than just constants
2008-12-04 11:42 sam
[r23] rewrote all <subst> code using <type_variable> instances rather than an assoc of ints, including the 'path' compression suggested by the secret eoplv1 draft chapter. probably still some more cleaning up to do.
2008-12-04 10:19 sam
[r22] read(): bug when reading the symbol '-'
2008-12-04 06:45 sam
[r21] insn_primop(): %array-set
2008-12-04 06:44 sam
[r20] compile_primapp(): emit %%array-set
2008-12-04 06:44 sam
[r19] expand_set_bang(): emit %%array-ref
2008-12-04 06:43 sam
[r18] test assignment
2008-12-04 06:41 sam
[r17] array assignment
2008-12-04 04:54 sam
[r16] array types
2008-12-04 04:48 sam
[r15] analyze(): point out in a useless comment how this is currently borken.
2008-12-04 04:44 sam
[r14] support %array-ref prim, plus range check
2008-12-04 04:44 sam
[r13] compile_primapp(): spun this into a separate fun added support for %%array-ref. eventually I'll get the 'number of percent signs' problem figured out.8^)
2008-12-04 04:41 sam
[r12] range_check(): new function for checking runtime array index
2008-12-04 04:41 sam
[r11] build_vector(): instead of calling the 'vector' function, use a %%vector-literal primappremoved unused gen_constructor() function
2008-12-04 04:40 sam
[r10] added an array type. still need to support set! on array refs
2008-12-04 04:35 sam
[r9] reader.read(): support postfix array-reference syntax. a little ugly.
2008-12-04 04:35 sam
[r8] use '-m64' instead of '-arch ppc64', this seems to work on intel and ppcdefault to -m64 on os x
2008-12-02 03:55 sam
[r7] don't tar up the svn dir
2008-12-02 03:43 sam
[r6] analyzer.replace(): when recomputing the node size after replacing the sub-nodes, count this node as one, not zero.analyzer.inline_application(): verbose print reformatted so the size is easier to read
2008-12-02 03:41 sam
[r5] node.__init__(): default the size to one, not zero
2008-12-02 03:40 sam
[r4] removed some prints, protected others with 'if verbose:'removed add_typechecks() function
2008-12-02 03:15 sam
[r3] *** empty log message ***
2008-12-02 03:00 sam
[r2]
2008-12-02 02:58 sam
[r1] pxll import

Generated by svn2cl 0.12