| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| ... | |
| |
|
|
| |
- Added some code to allow variant creation with existential types. (NOTE: Check if it's actually valid)
- Modify var cleanup to leave the var as-is if it has been deleted. (NOTE: Need to find out why a variable is left prior to being deleted)
|
| |
|
|
|
|
|
| |
(slot-names are no longer Text, but Syntax).
- lux/type;bound? no longer does further derefs to check for bound-ness?
- Improved the way 2 type-vars are type-checked.
- When cleaning type-vars, the dereffed typed is no longer cleaned again.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
- Added the ability to export from def.
- Added an optimized lambda macro.
- Finished record analysis & compilation.
- Fixed a bug in tuple, function & record analysis wherein AppT wasn't being performed prior to analysing the expression under analysis.
- Fixed several bugs wherein "fail*" was needed but "fail" was used.
- Added a case for records in base;show-ast.
- Made an improvement for AllT in type;show-type.
- Corrected an error in pattern-matching compilation wherein casts weren't being performed to make sure the source datum was of the necessary type for PM.
- Removed the (now unnecessary) lux/macro.
|
| |
|
|
|
| |
- Renamed a few defs in lux.lux.
- No more type-test shortcut inside lux/analyser/lux.
- Defs are now being classified as either (#ValueD <type>) or (#MacroD <macro>).
|
| | |
|
| | |
|
| | |
|
| |
|
|
| |
of TypeEnv. The performance of the type-checker also improved thanks to that.
|
| | |
|
| |
|
|
|
| |
- Fixed a bug wherein bool pattern-matching was only considered total if there was a wildcard, despite true & false cases where provided.
- Changed the syntax of the jvm-interface special form.
- Refactored the lux;writer, lux;eval-ctor & lux;loader fields of the CompilerState into a single data-structure which is now stored under the lux;host field
|
| |
|
|
|
| |
- Fixed several bugs within lux.lux.
- Renamed "check'" to ":'" and "coerce" to ":!".
|
| |
|
|
|
|
|
| |
- Fixed an error in lux.base/analyse-2
- Modified the analyser so the symbols that identify all of the special forms must mandatorily have "" as their prefix.
- Fixed a bug in the binary operations at lux.analyser.host wherein the types where being omitted.
- Fixed a bug when closing-over variables inside lambda bodies wherein the names of bindings where being stores as (incomparable) arrays, instead of as (comparable) strings.
|
| |
|
|
|
|
|
| |
complexity to explode and cause the compiler to become very slow (solved it by removing type-inference from tuples).
- Also removed type-inference from functions/lambdas.
- Added a small optimization to improve the efficiency of type-checking by not doing a thorough type-check when a global or local binding has a type variant with the same cases as Type, and it's exo-type is also like this (hopefully, it will never happen that someone will exploit this to make the compiler do something weird...)
|
| |
|
|
|
|
|
|
| |
destructuring can be done at do).
- Patterns are no longer put inside a MatchAC structure, but are instead just moved around as lists.
- Code outside of &type can no longer create/delete type-vars directly, but must now rely on with-var & with-vars to manage the life-cycle of type-vars.
- Simplified pattern-matching analysis at lux/analyser/case.
- The LEFT_APP optimization in check* has been replicated on the other side as RIGHT_APP, to attempt to improve performance of pattern-matching.
|
| |
|
|
|
| |
- Fixed a few bugs with types and pattern-matching.
- Fixed a bug wherein primitive-analysis did now unify the primitive type with the exotype.
- Modified lambda-analysis so functions subject to universal quantification can manage better the universally-quantified variables.
|
| |
|
|
| |
- Variants now need types to be specified both when constructing them as when deconstructing them (in pattern-matching).
- Simplified DataT types by no-longer requiring a list of params, so only the tame name is necessary.
|
| |
|
|
|
| |
- Added more cases for totality checking in pattern-matching.
- Fixed a bug when type-checking function application that used the exo-type for both the function being applied as to the application itself.
- Fixed a few minor bugs scattered around that had to do with wrong number of arguments being passed to functions.
|
| |
|
|
|
| |
- :let within |do now uses |let instead of let.
- The analyser now does totality analysis and structures the pattern matching, with the compiler only compiling the generated structures.
- Local bindings with case' can now be prefixed arbitrarily. (Note: must do the same with functions).
|
| |
|
|
| |
compiler code).
|
| |
|
|
|
|
|
|
|
| |
module, instead of the local module. (e.g. ;map, #;Cons, ;All)
- The type-checker now takes into account 2 types: an exo-type (imposed by outside forces) and an endo-type (generated through inference)
- Fixed a few bugs in the analyser and the type-system.
- &type/solve* is now the actual type-checker and &type/solve invokes solve* with an empty fixpoint environment.
- The exo-type of Statements is Nothing.
- variants, tuples and def' are being analyzed properly now.
|
| |
|
| |
- Removed exec, get@' & set@' special forms, as they are not primitive enough as to be part of the language.
|
| | |
|
| |
|
|
| |
lexer nodes & parser nodes.
|
| |
|
|
| |
between raw source-code and the lexing-phase.
|
| | |
|
| |
|
| |
- System works correctly once more.
|
| |
|
| |
- Corrected more bugs in the system.
|
| |
|
|
| |
- Changed indents to symbols.
- Corrected some of the bugs in the system. Many more still awaiting fixes.
|
| |
|
|
|
| |
- Compiler-state is now using Lux record-format.
- Went from Error+Ok to Left+Right.
|
| |
|
|
| |
- Finishing implementing the type-system.
- Migrating more of the data-structures used in the compiler to the ones used by Lux itself.
|
| | |
|
| |
|
|
| |
format as lux data-structures, to ease communication between the compiler & macros.
|
| |
|
| |
- Made a small change in float & double comparisons to make sure NaN < n.
|
| |
|
| |
- The lexer now allows "-" in front of numbers to get negative numbers.
|
| | |
|
| |
|
|
| |
- Fixed a bug when compiling lambda instantiation (the bug had to do with compiling the env variables passed to the closure).
- Fixed a bug in the lexer wherein the backquote (`) wasn't being recognized.
|
| |
|
|
|
|
|
|
|
|
| |
now "jvm-" instead of "jvm;"
- Fixed the bug where the same local vars/registers were getting registered more than once and the class-verifier complained.
- Fixed a bug where the "end label" for pattern-matching bodies was never inserted.
- Simplified the analyser by removing "self" calls and having self be just a local for the "this" object (register 0).
- Removed the lambda-folding optimization.
- The compiler state now holds and environment for naming globally-scoped lambdas.
|
| |
|
|
| |
of using custom classes for that.
|
| |
|
|
|
|
| |
function self-calls are broken).
Also fixed several small bugs, including the one that caused ".apply" to never make progress (always returned a function with count 0).
Also fixed an issue with scopes that caused class-names to be generated improperly.
|
| |
|
|
|
| |
- For some reason, when I fail at compilation due to mistaken use of tags/variants, I don't get the right error reported.
[Fixes]
- Solved a problem where stack values weren't being properly cleaned-up when pattern-matching on variants and branch bodies.
|
| |
|
|
|
| |
- Simplified the code for handling definitions and macros.
- Slightly changed the names of the fields of tuple classes.
- Added a (currently empty) optimization phase.
|
| |
|
|
| |
Codebase still needs to be simplified further, though.
Also, an explicit optimization phase, between analysis and compilation, must be established.
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
| |
- The monadic implementation of macros is almost done.
- Missing error-handling.
[Fixes]
- The output folder is now generated on each compiler run to avoid exceptions thrown by the class-loader.
|
| |
|
|
|
|
| |
- Fixed the errors in the compiler due to its restructuring as monadic code.
- Fixed a bug in the analyser where 'case' forms has "Nothing" as their type.
[Refactor]
- Now there only one way to raise/fold lambda-bodies.
|
| |
|
|
|
| |
- The analyser trying to match function calls first was messing around with variant construction.
[Working on]
- Monadic macros.
|
| |
|
|
| |
- Huge refactoring of the lux/compiler ns.
- No more difference between compiling global lambda definitions vs regular lambdas.
|
| |
|
| |
- Almost finished rebuilding lambda syntax so it always has local names and only takes 1 argument.
|
