May 93 - WAMADA Notes
The February WAMADA meeting was held a week late to avoid conflicting with MADACON. As a result, members who traveled to San Diego had a chance to report on the latest news from the conference. It will suffice to say that no one leapt for the phone to change their current development plans. The expected release of another MacApp update was the piece of news most immediately useful to our members. Component Software seems to be off to a good start with their plan to involve third parties in the enhancement of Component Workshop™. This could become a key weapon in their upcoming battle for market share with Symantec's Think C++. Bedrock is still on the horizon, and the viability of a single framework which provides a "native" look and feel on more than one platform remains barely tested. At the 50,000 foot level (a common Apple metaphor for "the big picture") the concept of writing frameworks to encapsulate your application's knowledge is the latest programming paradigm. Dave Wilson was the most public advocate of this idea, Taligent the most subtle.
Threads and Asynchronicity
The second part of the February meeting focused on ways of providing multiple execution paths in MacApp applications. Yvon Perreault of Cactus Inc. (a large-format color printing company) presented some sample code for the use of Apple's Threads package to provide asynchronous polling of PPC I/O completion calls. One "gotcha" with using threads is that segments can not be unloaded while a thread is executing, and since the main program is also a thread, MacApp's auto-unload feature had to be disabled.
Phil Flack of McDonnell Douglas Aerospace showed his code which provides event handling while the mouse is held down in controls. This was used to allow the monitoring of real-time devices which can transmit data to the Mac at any time. Phil overrode Button, CheckBox and Radio controls to set the control's action proc to a function which calls PollEvent during tracking. The key "trick" to making this work is to call Focus after PollEvent has returned. If the routine which is calling PollEvent is performing drawing, you need to also save off and restore the visRgn, clipRgn, and state of the fUpdating flag during the call.
Phil and coworker Mark Gerl also showed some code for a TEventHandler class called TTimedHandler which serves as an intermediary for TBehaviors on the application's idle chain. Since TBehaviors are not descended from TEventHandler, they can not directly be part of the idle chain. By using a TTimedHandler, a TBehavior can indirectly add itself to the idle chain. The TTimedHandler just passes the DoIdle call through to the TBehavior. Lastly, they described how the use of a state machine within idle methods can be used to break up processing of lengthy tasks without destroying the responsiveness of the user interface.
The March WAMADA meeting fell four days after the infamous blizzard of '93, and squarely on Saint Patrick's Day. As a result, attendance also fell. Still, almost a dozen attendees gathered and, to start the meeting, tackled the subject of becoming an official MADA affiliate. The chief concern was that an admittance charge for non-members would drive people away, rather than encourage them to pay for membership. This lead to a discussion on how to structure MADA's fees and benefits. One proposed benefit of becoming an official affiliate is that we would receive a small share of the profits from MADA products sold at meetings. Since this is an inconvenient way to purchase software, it was suggested that a member's affiliation be recorded when they sign up, and that the affiliate get a cut from any MADA purchase which they made. Electronic access to bulletin boards is a common and often essential part of our jobs, so it was suggested that MADA establish a board on one or more of the commercial systems and limit access to members only. (Using existing systems removes the costs of billing and some administration from MADA, and provides the most widespread service). Also suggested was a more staggered fee structure, where a higher fee gains you increasing access to MADA's benefits, namely: printed information, books and software, on-line services, meetings and conferences. The question of where affiliate meetings fell in this list of services was left unanswered. Since the board member writing the affiliates rules (Leslie Jeffries) is also in charge of WAMADA, we agreed that becoming an affiliate was a good thing to do, but we wanted to see a more complete design of the implementation before making a formal declaration (typical OO programmers, huh?).
Our March speaker was Dan Corenzwit of Mitem Corporation. Mitem makes MitemView™, a communications product which people often associate exclusively with HyperCard. While HyperCard is the original interface for MitemView, it is no longer the only one. Now in its fourth generation, MitemView can make use of HyperCard, 4D, and C++. Note that I said "make use of" rather than "be used by". This is because Mitem considers MitemView (MV) to be a framework rather than a library. Dan credits Dave Wilson's session at MADACON for putting the difference in perspective. A library provides services which your application must determine how and when to call. A framework provides functionality that calls your application's additions at the appropriate time.
In the case of MV the functionality provided is the ability to recognize the arrival of messages and associated data from remote machines (usually mainframes) and then call your code to handle the message and its data. Its primary market is people who need to use "legacy" applications (all those mainframe COBOL applications which communicate with hardwired terminals) on their micros. Dan explained that MV views the remote system as a non-sequential, non-deterministic state machine and attempts to linearize the complexity of moving between states. MV works asynchronously (hmmm, I see a trend here!). It accepts input from a variety of communication services (SNA, TCP/IP, etc.) examines the "presentation" of the data (sort of a generalized terminal emulation) and matches incoming patterns with pre-built messages which it then sends to you. MV is able to update itself on the fly, from information provided by the sender, if a new configuration is needed to handle changes in the presentation.
To support MacApp, Mitem provides several C++ classes, along with code which is installed in your application as resources. The MV communication objects are not part of the target chain. This allows interactions with the remote machine to work in parallel with user interface activity. When MV recognizes a pattern it sends your code a message in the form of a message name. This name is turned into a method call by way of a registration table which your application builds. The result is that your application becomes a sort of way station between two users: the person running it, and the remote machine sending in data.
In combination with the previous month's discussion of handling asynchronous I/O, I found the asynchronous nature of MitemView its most interesting feature. Dan showed that responding to messages from MV in the proper order can improve response time, by taking the remote machine's processing time into account. As operating systems (including the Mac's) move towards providing multi-threading and multi-processing services, those of us who long ago became comfortable with event-based programming will have to take the next step to working with these new services. Thinking back to MADACON, I don't recall many vendors touting multi-threading as a factor in the design of their frameworks (SmalltalkAgents is the only exception that comes to mind). Let's hope this changes as fast as the new OSs change.
If you're in the D.C. area, and interested in object oriented programming, give WAMADA a visit. We meet every third Wednesday at McDonnell Douglas in Tyson's Corner, Virginia, beginning around 7:15 p.m. For a map, send a message to JEFFRIES.L on AppleLink, or call Leslie at (301) 340-5126 during business hours (EST).