[XPS-devel] Summary of recent discussions...

[Saem G. <sae...@sh...>]

This is a summary of recent discussions between Reid and Saem.

If any particular topic interests you or you have questions about them, start 
a new thread with that being the focus.

First off, the project isn't dead.

There is also an underlying tone of a little bit of urgency, lest we're left 
behind, jetbrains is already working on MPS 
(http://www.onboard.jetbrains.com/articles/04/10/lop/).

The aim is the to make programming easier for non-programmers, such that they 
can not only be experts in their field but express that expertise to develop 
a computer program.

There are issues on the tool side as well.  Both Reid and Saem agree that the 
format of programs and source code, the way they're stored and the IDEs that 
edit them need work.  Programs and libraries of reusable components should be 
stored in a database of some form.  IDEs should be more powerful such that 
they can be more than a faster means to edit text, rather they should provide 
a lot of helpful suggestions based upon how and what you're doing.  Reid 
agree, but feels the benefits of such aren't as dramatic as Saem.  Instead, 
Reid feels that there will be greater benefit from an IDE working on a more 
macro scale and helping the user combine larger components.  I think in this 
matter, there is room for both, creation versus derivation.

It was agreed that programs are created best with an iterative or evolutionary 
process.  However as the things went further down discussing software 
engineering it was decided that it was too tangential a discussion to persue.  
This would be revisited, however.

The two issues slowing down XPS right now are LLVM going to 1.5 for 
portability features and linking syntax with semantics and generating 
efficient code.

Both agreed that having languages work with each other is going to be 
key.  The issue is how much cooperation is going to be allowed.  Saem 
proposed a C interface which most language can work with.  Reid, basically 
agreed and thought that there should be a few common things (strings and data 
structures) and the rest should be met with you can't do that in XPS from 
XYZ.

Saem put forth the use templates and macros for translation of higher level 
source to core XPL.  This allows one to fairly easily define syntax and 
semantics.  Reid countered with having far too many layers for transformation 
and that code transformation isn't particularly good in terms of optimization 
unless the user does a lot of heavy lifting.  Reid wanted a way to easily 
extend the compiler to understand new languages and language features 
natively by the compiler.  Saem agreed and templates and macros fell from 
being the main thrust of extension.

Namespaces and versioning came up.  Fortunately, namespaces were already 
addressed even at the workspace level.  This should hopefully avoid domain 
specific terminology collisions.  Versioning is still unresolved.  Reid and 
Saem agreed that versioning must be solved by the programming system; 
however, Reid wasn't sure he wanted he wanted it in the language or visible 
by the user, while Saem wanted it as a first class citizen in the form of 
changesets and incorporated into referencing components and code.  Reid 
countered that there are companies which have versioned on per character and 
that mess should be avoided and a DSL might be up to solving the problem.  
Reid felt that it might be best if interfaces and implementation were 
versioned, that and they should play nicely from development, testing, 
integration, staging and production scenarios.

Inheritance and it's pros and cons in terms of describing semantics was 
considered.  Saem initially thought that this would be enough.  Reid 
countered with an example, showing that fuzzy logic is more than a double and 
specialization/categorization wouldn't really cut it, in fact composition 
would be superior in most cases.  Reid added that categorization would be 
completely ineffectual in terms of dealing with completely new types of 
semantics and that is likely the larger issue, creating semantics.