- missing import should always show up in the summary of errors. eg
ERROR import X ERROR fn X(1) doesn't exist
ERROR can't find type for import X
-- if a macro has a bug, the stacktrace is ugly
100 char or less per feature, points included syntax
- (points) description
points logrithmicy difficulty 0-3
- 0 know what to do, how to do it, and it's trivial < 15 min
- 1 know what to do, how to do it, but not trivial < 1/2 day, > 1 hour
- 2 know what to do, but not how to do it, < 1 day, > 1/2 day
- 3 don't know exactly what to do, or how to do it, est is undefined
- mirah-parser maven artifact mirah/mirah-parser#16
- Java 8 ...
- look at java 9 features we could use
- update mvn compiler plugin
- add regression test for ant task example!
- rm implicit ant dependency... break out ant support as separate maven artifact
- mvn artifacts: -- mirah-core -- mirah-parser -- mirah-ant -- mirah
-
add generic warnings / error message facility -- right now we've got errors and they're ok. but it'd be better if we had something better.
-
make Mirror type systems addDefaultImports more obvious
-
faster mirah compiler build: src => class is tricky, I could make src => jar, jar => jar tho
-
switch to minitest
-
tests for imports at different scopes
-
shade asm lib for dev happiness
-
keep version in fewer places whew. Current is version.rb various pom.xmls some random mirah file in src
-
automated tests for distribution artifacts
- verify bytecode version of compiler should have dist jar check that does this as part of acceptance suite
-
silence logging from JRuby interfaces, unless overridden.
-
get rid of all outputs that escape from tests
-
move build artifacts to javalib for gem so that we don't depend on dist directory
-
clean up javalib dir
-
don't use 1.5 as java compile target
-
make it obvious when ASTs have been dropped during macro expansion
-
add timing to phases
-
break compiler into phases
- sort files by import graph (could cache this)
- load macros
- load type systems
- load types for System, & predefs maybe?
- load scoped macros / extensions
- visit script body / class bodies, skipping method bodies
- compile macros
- visit method bodies
- resolve proxies
- change captured locals into binding fields
- clean ast up
- maybe after parse, check built files to see if they need to be rebuilt, and ignore them otherwise
- investigate using Java 8 Lambda impl hooks for closures.
-
fix import not found so that it says can't find com.blah for import com.blah, or at least both package scoped & absolute packages. mirah#267
-
warn about unused imports mirah#268
-
initial not found should not be error in fine logging mirah#269
-
log bindings as they are created
-
don't warn on self.initialize mirah#270
-
can't find method error could list nearly matching methods mirah#271
-
error on not all methods implemented for interfaces mirah#272
-
improve duplicate name/sig error when multiple method defs w/ same sig mirah#273
-
code sources from macro expansion should know both the macro location and the invoke location and report both on errors so that users know where to look to debug
-
-v should only show the version, it currently says no main method too. :/
-
add -Werror where warnings become errors
-
make block mirror type string a little nicer somehow eg Can't find method java.util.concurrent.locks.ReentrantLock.synchronize(org.mirah.jvm.mirrors.BlockType)
- variables w/ name package are not allowed :/ mirah-parser#18
- case expressions mirah-parser#17
- lambda literals mirah-parser#19
-
each macro for Maps
-
Reflection Macros
-
AST formatter that converts back to something the parser can parse
-
default toString?
-
default hashCode / equals
-
default val/no type that defers to indy eg def foo(x: int, y=1) end should work
-
allow defining equals using def ==(other)
-
attr_reader/et al as varargs instead of just Hash. It would mean that the type info would have to come from somewhere else
-
goto: headius wants it
-
synchronize intrinsic ala java's
-
file scoped macros / extensions
-
extension syntax
-
public / private / protected / package scope helpers
#macro def self.public()
#end
-
subclass access to super protected fields w/ @syntax
-
(0) add a gets function like Ruby
-
(2) a more "batteries included" set of imports
-
(3) generics generics generics
-
(???) ARGV
-
(???) field access from closures, synthesize synthetic accessors
-
(???) self in closures should be lexical self
-
covariant interface impl overrides
-
macro helpers for inserting things into initializers
-
macro helpers for add method if not already there
-
reverse macro on collections
-
inner classes don't get nested names class A class B end end
-
understand Java 8 default methods
-
default args on non-last position params. eg when a method takes a block(read: functional interface), it should be allowed to have default args before the block arg.
-
do jruby style method lookup
-
add env var hash constant ala Ruby's ENV.
-
Thread extensions: Thread.start {}, which does t=Thread.new{};t.start;t
-
macro for typecheck & cast. replace the x.kind_of? Y; Y(Object(x))
-
a[-1] should work
-
array ext size
-
make sure that macro lookup looks at super classes
-
"foo: #{x || y} acts weird", my guess is that interpolation doesn't generate the right code sometimes
-
lambda's can't exist in methods! !! by which I mean that they can't close over things in methods. You can have them there, they just don't close. >:[]
-
parser can't deal w/ unquotes in hash literals when key is unquote quote { puts
key: 1 } -
implicit void methods w/ explicit return generate incorrect bytecode. Let's see if that's true of explicit. nope.
-
import * doesn't always work correctly
-
Inner class issues. Two inner classes w/ the same name will confuse the type system class A class B def b @b = 1 end end end class C class B def b @b = 1 end end end
-
if with no else has type of the bool expression not type of the body. It should return nil w/ the body type, or just be an error This might be just a error reporting issue ERROR: Invalid return type org.mirah.typer.Scope, expected org.mirah.typer.ResolvedType if parent ^^^^^^^^^
-
constants are not referrable outside the defining scope :/ because they are private
-
referring to classes as constants causes crashes java -jar dist/mirahc.jar -cp test/fixtures/ -e 'puts org::foo::A' blows up w/ a asm error java -jar dist/mirahc.jar -cp test/fixtures/ -e 'puts org::foo::A.class' doesn't
- delegation macro library
- (???) testing framework
- (???) fix up shatner
- (???) fix dubious
- name extension / macro classes after method name
- break annotations / macro build path to make javac happier currently macros are compiled to same tree as runtime code. We should make it so you can compile them to a different dir. Additionally, the annotations used for macros are unfriendly to javac, so we should put them in the macro dir
-
overriding interface req type atm :( apparently. -- nope wrong method name interface Bar def bar(i: int):void;end end class Foo; implements Bar def bar(i) puts i; end end
-
ArrayList.new(list) should do generic inference, perhaps, so T of list -> T of the resulting object
-
when dealing in generics, error messages should highlight generic types somehow
blocks don't introduce new scopes when they are part of a macro currently :/
#fuzzy aka cast scoping for rescue clauses eg e = nil begin rescue X => e
rescue Y => e
end puts e # e is LUB(X, Y)
#behavior literate, exec spec list o example code + what the compiler says when you give it to the compiler things like errors, & behavior
fix maven plugins
what I want is to be able to say this might be a field access, if a field w/ that name exists on the class, then use it if it's a constant that's in scope that's not on the current class target, use it
def a CONST end
CONST=1
fixing closures
add block scope block scope "self" is captured self if in closure self may also be closure type, for method lookup. I think this means that the proxy node for calls might need another elem
what's happening is hit 1st block look for parent binding, none create binding for 'm' scope resolve block type, build closure hit 2nd block look for parent binding, walks up to 'm' but in m's scope y is not captured, so not added to binding
def m # b = Binding1.new() r do # Closure1.new b; b2 = Binding2.new y = 1 # b2.y = 1 r do # Closure2.new b2 y += 4 # b2.y += 4 end puts y end end
update binding/closure generation
one top level binding, 2 closures def m # b = Binding1.new() x = 1 # b.x = 1 r do # Closure1.new b x += 1 # b.x += 1 r do # Closure2.new b x += 4 # b.x += 4 end end puts x # puts b.x end
one tl binding, one closure binding def m # b = Binding1.new() x = 1 # b.x = 1 r do # Closure1.new b; b2 = Binding2.new y = 1 # b2.y = 1 r do # Closure2.new b2 y += 4 # b2.y += 4 end x += y # b.x += b2.y end puts x # puts b.x end
binding shared across closures either pull binding construction above 1st ref, or copy binding back after. Not sure if I like that it's probably better to pull binding up, because you don't know when things are run def m x = 1 r { x += 1 } r { x += 2 } puts x end
def m x = 1 l { x += 1 } l { x += 2 } puts x # => 1 run_ls puts x # => 4 end
SimpleNodeVisitor sucks. It doesn't seem to work unless you define all the things to visit :(
sometimes implicit nil in return screws everything up, and you get a stack error from the jvm
Hash (things w/ init_list) needs a position, or they blow up w/ NPE somewhere
can't have block args w/ name interface
wierd error for VV def foo: int if a < b end end
compile rake task doesn't fail, if it doesn't work for some things, it goes on anyway in generics. hmm