rytina / xtext-typesystem Goto Github PK

When the language allows to specify language constructs which can refer each 
other, it is possible that the type-inference algorithm does not terminate.

E.g., when your language allows to specify mutually recursive functions like

function even(X) { return X == 0 ? true : odd(X - 1) }

function odd(X) { return X == 0 ? false : even(X - 1) }

then for inferencing and checking the type of even the type of odd has to be 
known and vice versa.

With the current implementation, this leads to non-termination / stack overflow.

Hi,

Isn't there a bug with the compareTypeDispatcher?

It is declared with 4 arguments but invoked with only 3…

I noticed that because I cannot get such a compatible method to be called in my 
code…

Thank you

It's more a question than an issue. I just couldn't find some forum.

Instead of a type class I have a certain structure as type (a "type 
description" containing more information than the core type) and I want to 
interpret it myself in typeToString(). When I register it with useCloneAsType() 
the original object is destroyed and a new empty one is created. Of course, 
when I don't register it I get the error "no type calculation rules found for 
EClass TypeDescription".

So my question is: is there a way to keep the original objects returned by 
type() methods?

Thanks and also thanks for this great extension of Xtext!

Michael

We downloaded the trunk from the typesystem library and had to do some changes 
to make it compatible with Xtext 2.8 API. Find changes in attached patch..

compareTypeDispatcher defines compareTypes with 4 arguments, that is type1, 
type2, checkKind and trace. DefaultTypesystem (rev. 31) only invokes it with 
only 3 arguments, the last one missing. You maybe remove the trace in 
compareTypes, or add the trace to all invocations. For the last solution, I 
have added a patch.

Cheers,

Jens

Hi Markus,

could you please add the following patch to your type system plug-in?
We need this extension for one of our language constructs during type checking.

The ValidationMessageAcceptor gets per default the index -1 in the check method 
of the CDCTCAdapter class, so there is no possibility to provide an index just 
like for the error or warning methods of the AbstractDeclarativeValidator class 
in Xtext.

I'm trying to build my bundles with tycho (http://www.eclipse.org/tycho/) so 
they can be build with maven. I'm getting the following error because of the 
typesystem bundle dependency which could not be resolved.

[ERROR] Cannot resolve project dependencies:
[ERROR]   Software being installed: my.bundle 1.0.0
[ERROR]   Missing requirement: my.bundle 1.0.0 requires 'bundle 
de.itemis.xtext.typesystem 0.0.0' but it could not be found

I'ts very hard to fix this if no p2 repository exists.

Guide how to create a p2 repo: 
http://maksim.sorokin.dk/it/2010/11/26/creating-a-p2-repository-from-features-an
d-plugins/

BTW: Thanks for providing the bundle.

It will be nice to provide an interface which will allow to customize dynamic 
error messages. Currently all the ensure... methods allow to customize only 
static error messages.
Client may want to provide custom error messages depending on the checkable 
context.

InterpreterGenerator.xpt file has no explicit resource encoding set, so it 
inherits the default encoding (MacRoman for mac). This incorrectly reads the << 
>> characters, which in turn breaks generation.

The fix is simply to explicitly set the Resource Encoding to ISO-8859-1 on the 
InterpreterGenerator.xpt file (right click -> Properties -> Resource). 
Alternatively the resource encoding can be set on the entire project.

Currently the InterpreterGenerator assumes that the metaclass and interpreter 
packages are named according to a fixed pattern relative to the name of the 
grammar. The attached patch allows for the metaclass and interpreter packages 
to be explicitly set in cases (like mine) where the metaclass name does not 
follow the assumed pattern. If these values are not set, then the previous 
default pattern is used.

What steps will reproduce the problem?
1.lessons
2.phone.mobile.net
3.rain

What is the expected output? What do you see instead?


What version of the product are you using? On what operating system?


Please provide any additional information below.

sources were deleted from version control in changeset 
http://code.google.com/a/eclipselabs.org/p/xtext-typesystem/source/detail?r=35

newbie launch

xtext-typesystem seems to not work with Xtext 2.0.

Which versions are supported?

There is a bug in the DefaultTypesystem.java leading to subtypes not being 
properly detected.

In method findSubtypes() a call is made to findSupertypes(). This is a 
copy&paste error and should be a recursive call to findSubtypes().

To reproduce, just define an INTEGER type as a subtype of FLOAT (or similar). 
Any occurrence of an INTEGER variable or constant is rejected in places where 
FLOAT is expected.

This bug is present in 1.3 and all 2.x branches.

There is now a feature for the type system plugins (yay!) - can we have a 
feature for the interpreter & interpreter generator as well so that they can be 
installed via the update site?