Re: Transient environment with standard functions

[Matthew Keeter]

Yes, the graph is equivalent to an s-expr (with subgraphs being nested scopes).

The reason that I’m not just representing the graph as a single expression is 
for
efficiency – I’m doing incremental updates and change tracking, so you can 
update
one of the node's expressions and have the changes propagate through the graph
with a minimum number of re-evaluations.

A similar Python engine is documented here:
http://www.mattkeeter.com/projects/graph/

Regards,
Matt

On Jun 11, 2016, at 11:15 AM, Basa Centro <address@hidden> wrote:

> It seems in essence you are building up a Scheme expression using the
> graph and the code snippets and evaluating that expression.  If this
> is the case, do you even need to use a module system and/or to
> explicitly create environments?  Prima facie, it looks like you are
> trying to reinvent the wheel.  Is your graph really just equivalent to
> an S-exp?
> 
> Scheme is lexically scoped, and creates a "transient environment" for
> you through the use of lambdas.
> 
> Your use of thunks suggests that you are doing caching or lazy-eval.
> 
> I'm not sure this helps, but I hope so.  When thinking about this
> topic, I had this mental image of a GUI for building Scheme
> expressions--like "Visual Scheme". :)
> 
> (Basa)
> 
> On 6/10/16, Matthew Keeter <address@hidden> wrote:
>> The specific use case is for dataflow graphs, where you’re evaluating a
>> bunch of small
>> snippets of code that can refer to each other by name.
>> 
>> I’d like to make an environment in which the variables in the same subgraph
>> are
>> exposed as no-argument thunks.
>> 
>> For example, let’s say I have one subgraph that contains
>>    a = “12” => evaluates to 12
>>    b = “(+ 1 (a))” => evaluates to 13, since the (a) thunk evaluates to 12
>> 
>> and another subgraph which contains
>>    x = “1” => evaluates to 1
>>    y = “(* (x) 2)” => evaluates to 2
>> 
>> If I insert the thunks into (scheme-report-environment 5), they leak from
>> one graph
>> to another – and to be fair, the docs to say that assigning into this
>> environment is
>> undefined behavior.
>> 
>> However, if I make an empty environment with (null-environment), it doesn’t
>> have
>> useful functions like + and *; looks like (make-module) has the same issue.
>> 
>> I'm sure that this is possible in Guile, but I got tired of reading through
>> the source
>> files to hunt down undocumented function that do what I need [1].
>> 
>> -Matt
>> 
>> [1] Another recent incident: How do you programmatically list all of the
>> variables in a module?
>> You search the web, find http://www.draketo.de/proj/guile-basics/#sec-3-2,
>> see a a reference to
>> module-map, which doesn’t exist in the documentation, dig it up in the
>> source to see its
>> arguments, etc…
>> 
>> On Jun 10, 2016, at 6:11 PM, Basa Centro <address@hidden> wrote:
>> 
>>>> First of all: is the "sandboxing" aspect of these environment important?
>>> 
>>> Taylan,
>>> 
>>> Thanks, that's exactly what I meant by "motivation" in my first reply.
>>> (There was a recent, 6 months ago or so, thread on sandboxing in
>>> guile-user by the way.)
>>> 
>>> Was Matt trying to prevent _access_ (inaccessible vs. read-only vs.
>>> read/write) to data?  Or was he trying to prevent only _visibility_
>>> (e.g. for hygeine)?
>>> 
>>> I think we need to know more about "what" and "why" to answer his
>>> question, rather than just "how"--if that makes any sense.  Does
>>> Racket [1] even _really_ achieve what he needs to do (_prevent_ access
>>> for example)?  We don't even know if we don't understand the
>>> higher-level purpose.
>>> 
>>> (Basa)
>>> 
>>> [1] Disclaimer: I like Racket too--it is a great project.  I just hate
>>> to see someone leave Guile because of a minor technicality.
>> 
>>