Re: Anniversary hack: Zile editor with Guile support

["Gary V. Vaughan" <gary@gnu.org>]

Hi Ian,

On 25 Mar 2013, at 13:24, Ian Price <ianprice90@googlemail.com> wrote:
> "Gary V. Vaughan" <gary@gnu.org> writes:
>> I think we are misunderstanding one another somewhere :(
>> 
>> I'm talking about this:
>> 
>>  1. user writes some Guile Scheme, and feeds it to Lua Zile
>>  2. Lua Zile runs zlc to translate Guile Scheme to Lua
>>  3. Lua Zile runs luac to compile output of (2) to Lua bytecodes, and
>>     saves the result to disk
>>  4. Lua runtime executes file of Lua bytecodes.
>> 
>> I really don't see how Tree-IL fits in here, or makes loading and executing Guile
>> Scheme sources from Lua Zile any easier.
>> 
>> What am I missing?
> 
> Two observations:
> 
> 1. The frequent references to Guile Scheme (as opposed to just Scheme)
> imply that Guile compatibility is desirable.
> 
> 2. The most compatible Scheme implementation with Guile is, quite
> obviously, Guile.
> 
> So, then, my advice is quite simply to maximise the use of Guile, and
> minimise the use of new code when writing this Guile-Scheme->Lua
> compiler.
> 
> Since Guile is written as a hierarchy of translators, you don't need to
> implement Scheme per se, but can instead implement one of the languages
> to which we can compile Scheme already, and ask Guile to compile Scheme
> to that for you.
> 
> e.g. guild compile foo.scm --to=tree-il --output=foo.tree-il
> 
> Tree-IL is the "default" option, as it is relatively close to Scheme,
> but you won't have to deal with things like hygienic macro
> expansion. However, you can pick whichever one you find easiest to
> translate to Lua or Lua Bytecode.
> 
> I actually intend to follow a similar approach with a
> guile-in-the-browser project I would like to write, however, I am
> probably going to implement a vm, and have it interpret Guile's assembly
> language directly.
> 
> I'm not sure I can be clearer than that. If it wasn't enough, well, just
> pretend I didn't say anything :)

Okay, the penny has finally dropped.  And that does look like a more
compatible approach, so I will certainly experiment with it, when I reach
this item on my TODO list :)

Thank's for your patience!

Cheers,
-- 
Gary V. Vaughan (gary AT gnu DOT org)