An eclectic-pencil blog

Over at my Curlr blog I have added some suggestions for improving the quality of Curl code by enforcing some coding standards.

At the moment in a Curl file containing a class definition, class-side and instance-side expressions can occur willy-nilly.

Some developers have the idea that if a field is only referenced in one method, then declare that field where the method is declared. And declare class-side variables wherever you will.

As Curl still lacks a refactoring browser, this generous approach to coding-as-scripting can and does add to maintenance headaches.

I particularly like my own idea of {i-attributes } and {d-attributes } as those declared once -and-only-once as requiring to be initialized in a constructor or not.

These are not substantive changes such as adding co-expressions or generators as found in ICON and UNICON or becoming more truly multi-paradigm as is distributed OZ. What they address are some of the needs of PITL. In a large codebase, it is important to be able to grasp the intention at a glance so as to move on in confidence (no obvious issue there, so look elsewhere) when refactoring code without the craftsman’s tool.

What I have proposed is not a restriction, but rather only an option to be enforced as a tighter compiler-directive than stringent? is at present.

While never a big fan of UML, I have opted to suggest rational? = true for this meta-level view of Curl class code.

Sometime ago I suggested that a macro for SCURL file headers was overdue. I now have that suggestion:
{curl-class }

What this will mean is that when the header is found, that SCURL file may only contain one Curl class and only a Curl class (no stray globals, proc’s and other flotsam-and jetsam.)

Call it “Curl with class”. || humor

At LogiqueWerks I have replaced the first Curl 6.0 revision of Paul Sheehans ‘Curl Enterprise Framework’ by a slightly re-factored version which uses only the documented sub-applet API’s of Curl 6.0

The original code accompanying the book ‘Enterprise Curl‘ relied on a package ‘IPC’ for inter-process communication which relied on largely undocumented Curl internals. One problem not solved in the port of the framework to Curl 6.0 was that of communication with such transient items as the ’splash screen’ when launched from HTML so as to avoid

Error: InvalidAppletException: Unsupported content-type 'application/octet-stream'

arising from the internal Curl call to Applet.load-next-aux

In the most recent refactoring, launching ‘enterprise-parent.curl‘ results in parent-child calls between that applet and the AppletData of enterprise.curl. However, launching ‘enterprise.curl’ from HTML proved more of a challenge. To allow a simple callback, the package ‘logon’ has now moved within the ‘VMS-MOTORS’ package as has the one ’splashscreen.scurl’ formerly located in a SERVICES package.

In the case of ‘enterprise-parent.curl‘, a parent-callback is used to launch the logon dialog once the splashscreen has run its course. In the case of embedding in ‘enterprise-curl.html‘, the child applet is controlled from a JavaScript function with a call to ‘applet_invoke‘.

The one drawback is that now the Help menu cannot call upon the same SplashScreenManager class to provide the ‘about’ dialog.

The framework is now reduced from 13 to 11 packages. The logon.scurl arguably belonged with the SplashScreenManager in the application code found in VMS-MOTORS where both now reside so as to avoid circular imports from the WINDOWS package.

While the splash screen for ‘VMS MOTORS’ may not be the best illustration of the sub-applet model in Curl, the upside is that there is now a working example which does not use imports from CURL.REMOTE or make explicit reference to ‘remote-connection’ and undocumented Curl such as
{import * from CURL.REMOTE}
{define-remote-class public abstract open ParentInterface
{define-remote-proxy-class ParentProxy}
{define-remote-yxorp-class ParentYxorp}
{remote-method public abstract open {mention-string str:String}:String}
}

But what is available by running ‘enterprise-parent.curl’ with a breakpoint at
set ad.applet-invoke-ready-callback
in the Logon method
{method public {startup-with ad:AppletData}:void
is a view of the underlying sub-applet API features of Curl 6.0 such as
the connection – a TunnelRemoteConnection
the proxy – a TunnelRemoteProxy
none of which are exposed in using only the published API found in the IDE documentation and illustrated in the IDE extended examples.

The LogiqueWerks page also has a link to the example ’shape.curl’ and the various JavaScript tips to be found in ’shape-parent.html’ for the use of AppletData.

The new EnterpriseFramework code now contains an example of a subclass of AppletHistory which might also be of interest.

Several cosmetic features of the example UI for ‘VMS MOTORS’ will see a re-work in the days and weeks to come as my work commitments permit.

Over at blogspot I have a post on inspecting objects ‘live’ in web pages.
What I neglect to mention there is what is really nice if you are used to using Mercury Interactive and that is the fact that inspectors allow you to flip some properties of an object ‘live’ in your web page.

Suppose you have an inspector open on a widget which occurs more than once on your page. With one click on the inspector’s toolbar you get a color-picklist. You can flip the color of the inspected widget’s background or border ‘live’ on the web page in your browser. Ah … that ListBox!

Curl does not use <FORM> elements in web pages; instead, Curl uses the GUI widgets familiar to anyone who has been building Rich applications in Tk, Smalltalk, Swing or most any other GUI framework.

Once you Ctrl-Right-Click on a Curl control on a web page you will see why I now rank the Curl IDE up there with Seaside for Smalltalk or XOTclIDE (the extended OTcl IDE.)

In the Curl web framework a CONTROLLER represents a joystick-type device and a ‘control’ such as a TreeControl is a CONTROL. No “let’s all pretend this is MVC because that is the OOP TLA” mantra. But with anonymous procedures and macros. As I said, COOL. Maybe not REST, but COOL enough with SOAP or XMLHttpRequest. As those of us with teens have had occasion to say, ‘Give it a rest, will ya?…’

PS Comes with built-in profiling, test coverage for QTP and even HTTP monitoring. A web content language from MIT with lambda functions that does not even look like LISP. Not bad. Readable code for literate web programming. Nothing clever embedded in HTML comments.

|| COOL
|| web2.0
|| comments

and

|COOL-tag
|| web2.0
|| documentation
COOL-tag|

Not a kluge and most have not a clue. Oh yes, MIT. CLU. What was her name?