Friedger's Blog

RE: How many Gears does Curl have?

RobertS — Tue, 30 Sep 2008 16:16:54 GMT

Thanks for keeping us up on Gears as seen from Curl...
I wonder if Richard M-H has a Curl developer contact in Los Angeles, CA?
The LAJUG (Java User Grp) that meets monthly at SUN in El Segundo often has knowledgeable Google folks from Santa Monica ( and an interesting bunch overall with wide-ranging interests and backgrounds)
They would make a great audience against which to bounce off ideas about Gears, Curl and related.

RE: Extending Curl

dmccrae — Fri, 30 May 2008 12:20:18 GMT

I like the idea of using dot notation.

However, I do not have a good answer to the question.

I think 'define-syntax' environment gives sufficient access to compile time types of source fragments that you could detect if a specific fragment has a type amenable to transformation from dot notation to get notation. But doing that for an arbitrary block of code would involve recognizing much of the curl expression grammar, so would not be trivial.

What you described is essentially what Curl does with ActiveXObject.

But I may be misconstruing how you intend the syntax to be used. Maybe a narrower scope would be easier. For example, some kind of destructuring-bind for record iteration (similar concept to 'for-pixel')

Another possibility is just to create a class definition for the record, which would then be strongly typed, and already recognizable by the compiler as accepting dot notation. A 'define-record' syntax could be specialized for that purpose, as alternate to the more verbose 'define-class' equivalent.

RE: Extending Curl

friedger — Sat, 24 May 2008 12:14:27 GMT

Do you have some ideas how to unifiying the get method of RecordData (rs) and the dot notation of class member access, so I would like to translate rs.col1 into rs"col1".
The intention is to use the same where expression for RecordData and for Objects.
Currently I use regular expressions to replace '.col1' by '"col1"' and create a CurlSource from the new string. Is this the best way or should I analyse the CurlSource and work directly on the CurlSource object?

RE: Extending Curl

dmccrae — Thu, 15 May 2008 20:54:40 GMT

I think the idea of a "select" like query syntax on collections is quite interesting. Its often more concise to use a "domain specific language" for specialized operatiI think the idea of a "select" like query syntax on collections is quite interesting. Its often more concise to use a "domain specific language" for specialized operations.

Curl, with the language extensibility offered by macros, affords a lot of possibilities for concise sublangages. For example, the new 'fn' macro is a great example of a concise syntax for functions.

The availability of client side SQL through the CDK also opens the door to new ways of working with objects and records. Perhaps something like a Curl "Active Record" could emerge.

Erik Meijer's LINQ is an ambitious and interesting attempt to address the challenge of the "impedence mismatch" between the relational, document and object paradigms. It is based on expressing "design intent" declaratively, using generic functional operators.

I don't think we've heard the last word on bridging the paradigms, however.

In any case, I encourage you to pursue this idea.

RE: Extending Curl

friedger — Wed, 14 May 2008 07:30:45 GMT

I was thinking of a set of orders or business partners that exist in a kind of business logic layer. But it is probably true that this layer is usually not very thick in a curl applet, is it? And therefore, one works directly on RecordSets and not sets of object models.

RE: Extending Curl

cbarber — Tue, 13 May 2008 21:30:02 GMT

Well, presumably you could always just pick default field names like 'value' and 'key' for use with generic Aggregate-of and Association-of classes, which would allow someone to write something like:

{select * from my-array where value < 42}

The ability of syntaxes to compile types using {syntax-env.type-of x} would allow you to write a syntax that examines the type of the source container and act accordingly.

RE: Extending Curl

rhh — Tue, 13 May 2008 18:51:51 GMT

Well maybe you have a vision that I don't have yet. I agree with your comment about the benefits of APIs that are generic, but I don't really understand how you could run a SQL select statement over an entity such as an {Array-of int}. Don't the individual elements of the collection need to have named fields and such? And if they do, I think that pretty much makes the collection be like a RecordSet. Anyway, that's what I was thinking.

-Bert

RE: Extending Curl

friedger — Tue, 13 May 2008 17:34:41 GMT

The nice thing of having a select statement for more than record sets is that you can express the same filters on any type of collections. It is alright if the select is expanded to a Aggregate-of.filter expression but the developer doesn't have to care.

RE: Extending Curl

rhh — Tue, 13 May 2008 14:26:34 GMT

I think your "select" statement would be a welcome feature and I don't know of anybody who has already implemented something like this. I think it would be very nice if it worked on RecordSets. RecordView is a subclass of RecordSet so presumably that means that your select statement would work on RecordViews as well. I don't think it is worth trying to make it work on Collections, which don't really have the relational structure of RecordSets so it might be difficult to really map SQL semantics onto them. Plus, Collections already have Aggregate-of.filter which does provide some filtering capability.

Perhaps Doug McCrae would like to comment on the relationship between this and the SQLite library that is integrated as part of the Curl v7.0 ("Nitro") beta.

Regarding integration with Office applications, I know we have some ability to communicate with Office applications on Windows via COM, but I don't know what integration hooks or APIs are provided by Google Docs or OpenOffice.

-Bert

RE: Introduction

cbarber — Fri, 02 May 2008 16:18:42 GMT

It seems like this is some form of delegation where x,y,z are delegated to another class instances, but doesn't that instance need to be specified somewhere? That is, x, y and z need to operate on an instance of type OtherClassA and have access to its state and other methods, so presumably you would have to stash a reference to an OtherClassA instance in B or otherwise provide a way to get such an instance.

I have sometimes wanted to write a delegating syntax that would let you implement a given interface by delegating methods to the same methods on a field of the class. This pattern is common when implementing proxy classes but has no direct support in the language currently.