mikeash.com: just this guy, you know?

Posted at 2012-01-20 16:14 | RSS feed (Full text feed) | Blog Index
Next article: Friday Q&A 2012-02-03: Ring Buffers and Mirrored Memory: Part I
Previous article: Friday Q&A 2012-01-20: Fork Safety
Tags: classic followup guest nostalgia oldschool toolbox
The Mac Toolbox: Followup
by Gwynne Raskind  

Welcome back to Friday, NSBlog readers. Here's the promised followup to my article last week about the Mac Toolbox, where I do a line-by-line analysis of the code from the original article, alongside Mike's Friday Q&A for this week.


Includes and constants and globals, oh my
In the Toolbox days, there were no all-encompassing headers where you do one #import and you have everything you need. Right up until Carbon, this simple truth remained as a holdover from the times when compilers took a major speed penalty for every single header they had to read. Nor did #import exist. Just #include, with all the slings and arrows that flesh was heir to. At least the names were usually straightforward.

    #include <Dialogs.h>        // Dialog Manager (for InitDialogs())
    #include <Fonts.h>          // Font Manager (for InitFonts())
    #include <MacWindows.h>     // Window Manager
    #include <Menus.h>          // Menu Manager
    #include <QuickDraw.h>      // QuickDraw
    #include <TextEdit.h>       // TextEdit (for TEInit())
    #include <Controls.h>       // Control Manager

There also weren't nice string names and outlet connections from nibs to give you access to your resources, so you were best off defining a constant for every resource in your application to make your code even remotely readable. Resource IDs for applications started from 128, and you had better believe following that rule was important. Because resource files were opened as overlays atop each other, the system's resources sat right underneath yours, and if you overrode a system resource by defining another with the same type and ID, chaos could and generally did ensue.

    enum
    {
        kMenuBarID = 128,

        kAppleMenu = 128,
        kFileMenu = 129,
        kEditMenu = 130,

        kWindowID = 128
    };

Then you had global variables. Generally frowned upon in any modern program of today, they were all but a requirement for Toolbox development when not using C++ or some other object-oriented language. There was just no other efficient way to get data where it was going. Certainly the global variable for exiting the event loop was a lot better than returning Boolean from every single event handling function in your code on the off chance an action might lead to a Quit. Not to mention your aevt/quit Apple Event handler (more on that below).

    static AEEventHandlerUPP        oappUPP, odocUPP, pdocUPP, quitUPP;
    static Boolean                  gDone = false;

Alerts and StandardFile
Check out the fancy function for handling an about box! Later versions of OS 8 and 9 brought a better API for presenting simple alerts, but before that you just defined an ALRT (ALeRT) and DITL (Dialog ITem List) pair which described your about box and displayed them. I already violated my own rule about defining a constant for the resource ID, something that would tend to happen a lot when you needed just one more resource for this one little bit of code. Alert() and its cousins such as StopAlert() were used quite a bit for such things as error messages as well.

    static void DoAboutCommand(void)
    {
        Alert(128, NULL);
    }

The API for asking the user to select a file from the hard drive went through quite a few evolutions, along with the data type for storing the location of a file. My knowledge only goes back so far as the System 6 APIs, of which SFGetFile() was the relevant member. This returned an SFReply, containing a volume reference number, directory ID, and a 63-character MacRoman-encoded file name string. This was superseded in System 7 with StandardGetFile(), whose StandardFileReply instead contained an FSSpec, which was really just the same three values packaged up as a fancy structure for easier passing around.

In OS 8.1 (or 8.5, I've forgotten which), Navigation Services and the modern File Manager were introduced, which provided a considerably more featured API, an interface whose merits versus the old StandardFile dialogs were debated, and the FSRef datatype which survived right into some Cocoa code until URLs became the norm for referring to files on disk. A particular quirk of FSRef was that it could not refer to files which did not exist, prompting many coders to come up with their own solutions for holding either a parent directory FSRef and file name (an HFSUniStr255, a brutally contrived data type if ever there was one) or falling back on FSSpec. The latter was common for its simplicity, making FSRef a bit of a mess.

I did not use Navigation Services here, as I was trying to stick to System 7 APIs. Instead, I call into StandardFile, telling it to let me pick only files with the type TEXT, and checking for whether or not the user clicked OK before opening the file (or more exactly, calling the stub function which should have opened it). Note: This is different from the version in the original article, which didn't limit the file type.

Even many Cocoa programmers are still familiar with the OSType concept. In OS 9 and earlier, files typically did not have extensions, and even where they did, only the most modern versions of OS 9 made significant decisions based on them. UTIs did not exist at all. Instead, files had (and indeed, still have, though they're much rarer now) "creator code"s and "type"s. These were ostensibly four-letter codes uniquely identifying the owner and type of a file, of which all-lowercase codes were reserved by Apple. In truth, they were just 32-bit integers (not strings) whose uniqueness was guaranteed by convention at best, and by nothing most of the time. Collisions in creator codes happened all the time, and there was no standardization of file types either. It was little different than problems with three-letter file extensions, just with a little extra space to work in. UTIs were created specifically to address these problems. I leave it to the reader to decide whether that's met with any success.

Note: The practiced C programmer will wonder how you can have a four-character character literal, as these are not supported by the language specification. The spec technically allows for these literals, but to my knowledge they were only ever widely used in Toolbox and Carbon programming, and have no place in Cocoa. Modern compilers will warn when they're used unless explicitly told not to.

    static void DoOpenCommand(void)
    {
        StandardFileReply stdReply;
        SFTypeList theTypeList = { 'TEXT', 0, 0, 0 };

        StandardGetFile(NULL, 1, theTypeList, &stdReply);
        if (stdReply.sfGood)
            DoOpenDocument(stdReply.sfFile);
    }

Apple Events
Apple Events were, and still remain to this day, the mechanism by which AppleScript functions. They were also the first real form of direct interprocess communication, an area severely lacking in the Toolbox before their introduction for the primary reason that before MultiFinder and System 7, only one application ever ran at a time. Ever! Desk accessories notwithstanding. Many applications were written with this in mind, and the cooperative multitasking environment provided by System 7 necessitated a lot of fundamental re-engineering for programs which had gotten too aggressive about owning the whole machine.

Apple Events had a "class" and "event code", which were both another use of those 32-bit character literals. They evolved much later into Carbon Events, becoming the primary mechanism by which the OS communicated with applications, but in the early days existed as the only "asynchronous" way for the OS to tell a program much of anything. Though, as they were dispatched via the event loop like everything else, this was little more than a convenient metaphor.

Four "core" events were "required" for every application. I put "required" in quotes because, in practice, applications could get away without even knowing they existed. A more accurate description would be that they were required for any application which claimed to be "high-level event aware" in its 'SIZE' resource (more on that later).

The first required Apple Event was the "open application" event, corresponding more or less to Cocoa's -applicationDidFinishLaunching: application delegate method. Since this event was not sent if the user had opened the application by double-clicking one of its documents (or later, dragging and dropping a document icon onto the application icon), it was a good place for doing your "default" thing. In this case, I have it do a "new document".

What's this pascal keyword, you might wonder? This was a requirement for just about any callback the Toolbox might call, and told the compiler, "emit code to make this function comply with Pascal's calling conventions instead of C's." Because the Toolbox itself was written in Pascal (or in conforming assembly language), anything it called had to accept parameters and return values as specified by Pascal, or havoc would ensue. Every Toolbox routine was declared pascal in the C headers, when it wasn't just a bit of inline 68K machine code.

    static pascal OSErr     AEOpenApplication(const AppleEvent *theAE, AppleEvent *reply, UInt32 refCon)
    {
        DoNewCommand();
        return noErr;
    }

For the "open documents" Apple Event, corresponding to Cocoa's -application:openFiles: method, I get the "direct object" parameter as a list from the event, count the items in that list, then loop over the list getting each item as a FSSpec pointer and opening the resulting spec. Then I dispose of the list.

In all of this, I ignore quite a few error checks which would make the code all but unreadable. This is almost entirely pure boilerplate code.

Perhaps you can see how all of this was intended to produce a type-agnostic flexible event system, but in practice all it produced was a complicated batch of APIs which were fiendishly difficult to use correctly in all cases. Maybe the OS sent me an FSRef instead of an FSSpec, and this was an OS version which didn't include a working coercion handler to turn one into the other. Maybe there was no direct object, or the direct object was just a single file. Most of the assumptions this code makes were only safe for System 7.5 and earlier, and check out how many parameters I have to declare junk variables for because NULL was not a safe thing to pass.

    static pascal OSErr     AEOpenDocuments(const AppleEvent *theAE, AppleEvent *reply, UInt32 refCon)
    {
        FSSpec          file;
        AEDescList      docList;
        SInt32          items, i;
        Size            siz;
        AEKeyword       kwd;
        DescType        typ;


        AEGetParamDesc(theAE, keyDirectObject, typeAEList, &docList);
        AECountItems(&docList, &items);
        for (i=1; i<=items; i++)
        {
            AEGetNthPtr(&docList, i, typeFSS, &kwd, &typ, (void *)(&file), sizeof(FSSpec), &siz);
            DoOpenDocument(file);
        }
        (void) AEDisposeDesc(&docList);
        return noErr;
    }

I didn't even think about supporting printing (a truly monstrous task in Toolbox days), so I just return an error from the "print documents" handler. If I'd chosen to allow it, I could have used the same boilerplate from the open documents event to get the list of files. Yes, the exact same code, save that I'd call DoPrintDocument(file) instead.

    static pascal OSErr     AEPrintDocuments(const AppleEvent *theAE, AppleEvent *reply, UInt32 refCon)
    {
        return errAEEventNotHandled;    // Don't support printing
    }

Finally, the "quit application" event. This is not equivalent to -applicationWillTerminate:, because if this event handler doesn't do anything, the application won't quit! I exit the event loop from here, of course.

Mind you, I wouldn't get this event if I chose "Quit" from the "File" menu (not the application menu, there was no such thing), unless I dispatched it to myself manually.

    static pascal OSErr     AEQuitApplication(const AppleEvent *theAE, AppleEvent *reply, UInt32 refCon)
    {
        gDone = true;
        return noErr;
    }

Initialization
Surely, you must be wondering, at least the system initializes itself on startup. Maybe there's just one function that needs calling, as with NSApplicationMain() or RunApplicationEventLoop()!

Sorry, no. In the Toolbox days, you did everything yourself. Let's take a look at the breakdown:

    static void Initialize(void)
    {

First off, maximize my application's zone (this does not mean what you think it does) and ask for more master pointers.

Maximizing my zone means that I'm asking for my heap to be as large as it can be before doing anything else. This saves the Memory Manager from having to grow it later as my allocations increase, a potentially time-consuming task. Why isn't this the default? Old design decisions from the days of tiny memory space.

Master pointers were used to implement handles. A handle is a double-pointer which allows a block of memory to be relocated at need without having to update every reference to that block. This theoretically prevented memory fragmentation, but it could get ugly with the need to lock handles before accessing their contents. There was no reference or depth count on locking, making passing handles around a very fussy business.

        MaxApplZone();
        MoreMasters();
        MoreMasters();

Now, initialize all the basic Toolbox managers. Start with QuickDraw, and give it the address of the GrafPort passed by the OS. That's the very ancient equivalent of an NSGraphicsContext, for you Cocoa devs. You'd think if it were passed by the OS, you wouldn't need to give it back to the OS, but the nature of global variables in the Toolbox world made this the easier way. In theory, you could have passed a different port, but I've never seen nor heard of such a thing, and I can't imagine what the use would have been. By the time Carbon rolled around, this entire initialization had been completely done away with.

        InitGraf(&qd.thePort);

Initialize the Font Manager, the Window Manager, the Menu Manager, and TextEdit. These all do what they say.

        InitFonts();
        InitWindows();
        InitMenus();
        TEInit();

Initialize the Dialog Manager. Pass NULL so there's no resume procedure. A resume procedure was, pre-System 7, a function that could be called by the OS when the infamous bomb dialog popped up and the user clicked "Resume". Generally, the function would do a jump to main(), restarting the application. This was the only safe thing to do, though in olden times programs would sometimes actually attempt recovery. This is the equivalent of trying to restore normal operations from a SIGSEGV handler in Cocoa code. Good luck with that. From System 7 on, it was unsupported to pass anything other than NULL for this, and I believe that doing otherwise had no effect anyway.

        InitDialogs(NULL);

Finally, set the cursor to the standard arrow and show it.

        InitCursor();

Now make sure there's no leftover cruft in the event queue from anything the user might have done between launch time and initialization. Computers were slow enough back then to make this worth doing.

        FlushEvents(everyEvent, 0);

Install handlers for the four "required" Apple Events. What is a UPP, you might ask? A UPP is a "Universal Procedure Pointer", a convoluted bit of trickery and hacking that the Mixed Mode Manager used to make sure the Toolbox could talk to 68K and PowerPC programs equally regardless of which architecture the program itself or the particular Toolbox routines involved was using. Apple sidestepped the whole problem during the PowerPC to Intel transition by just recompiling everything as a fat binary, a much more viable option in the Panther and Tiger days than in times when hard drive sizes were still measured in the tens of megabytes.

        oappUPP = NewAEEventHandlerUPP(AEOpenApplication);
        AEInstallEventHandler(kCoreEventClass, kAEOpenApplication, oappUPP, 0L, false);
        odocUPP = NewAEEventHandlerUPP(AEOpenDocuments);
        AEInstallEventHandler(kCoreEventClass, kAEOpenDocuments, odocUPP, 0L, false);
        pdocUPP = NewAEEventHandlerUPP(AEPrintDocuments);
        AEInstallEventHandler(kCoreEventClass, kAEPrintDocuments, pdocUPP, 0L, false);
        quitUPP = NewAEEventHandlerUPP(AEQuitApplication);
        AEInstallEventHandler(kCoreEventClass, kAEQuitApplication, quitUPP, 0L, false);

Finally, load up our 'MBAR' resource and set the menu bar hold the menus it lists. Add the installed desk accessories to the Apple menu, and draw the menu bar.

AppendResMenu() is a bit interesting. The original implementation of this Toolbox call did exactly what it said: Appended the names of every resource of the given type to the given menu. With 'DRVR', this meant each installed desk accessory, as those were stored as DRVR resources. However, in System 7, it became possible to put other things in the Apple menu. So, AppendResMenu() had its meaning overloaded so that when drivers were added to an Apple menu, it would instead add all the items that belonged in an Apple menu.

        SetMenuBar(GetNewMBar(kMenuBarID));
        AppendResMenu(GetMenuHandle(kAppleMenu), 'DRVR' );
        DrawMenuBar();
    }

Menu commands
Menu commands are passed around in their raw form as 32-bit integers encoding the menu's resource ID in the high 16 bits and the item number in the low 16 bits. To figure out what to do, one extracts these two words and branches based on them. Yes, this means your code is intimately and directly tied to the exact layout of items in your menu resources. You remembered to make a constant for every menu ID and item number, right? No? Wow, are you in for a fun time.

You may notice that for all items in the Apple menu that I didn't define, I call the OpenDeskAcc() function with the name of the item. This is the second half of the overloaded AppendResMenu() call from earlier; it was overloaded similarly to open the appropriate Apple menu item rather than just desk accessories.

At the end, regardless of what happened, I call HiliteMenu(0). Yes, that's right, the Toolbox doesn't un-hilite the menu title for you. You have to do it yourself.

    static void HandleMenuSelection(long response)
    {
        short       menu = (short)((response & 0xFFFF0000) >> 16),
                    item = (short)(response & 0x0000FFFF);
        Str255      name;

        switch (menu)
        {
            case kAppleMenu:
                if (item == 1)
                    DoAboutCommand();
                else
                {
                    GetMenuItemText(GetMenuHandle(menu), item, name);
                    (void) OpenDeskAcc(name);
                }
                break;
            case kFileMenu:
                if (item == 1)
                    DoNewCommand();
                else if (item == 2)
                    DoOpenCommand();
                else if (item == 4)
                    DoSaveCommand();
                else if (item == 5)
                    DoSaveAsCommand();
                else if (item == 7)
                    DoCloseCommand();
                else if (item == 9)
                    gDone = true;
                break;
            case kEditMenu:
                if (item == 1)
                    ;
                else if (item == 3)
                    DoCut();
                else if (item == 4)
                    DoCopy();
                else if (item == 5)
                    DoPaste();
                else if (item == 6)
                    DoClear();
                else if (item == 8)
                    DoSelectAll();
                break;
        }
        HiliteMenu(0);
    }

Event handling
There are lots of different kinds of events the OS can tell you about! In Carbon, there were literally hundreds of potential Carbon Events you might be sent. But even in the Toolbox, there were plenty of things that might happen...

    static void HandleMouseDown(EventRecord *event)
    {
        WindowPartCode      part;
        WindowPtr           window;

The mouse was clicked? Well, let's figure out which window it was clicked in and what part of that window.

        part = FindWindow(event->where, &window);
        switch (part)
        {

It was in the menu bar! Call a Toolbox function to synchronously track the menu drag and return the 32-bit response, and pass that off to our menu command handler.

            case inMenuBar:
                HandleMenuSelection(MenuSelect(event->where));
                break;

It was in a window the application doesn't own? Well, let the system handle it, then...

            case inSysWindow:
                SystemClick(event, window);
                break;

It was in a window the application does own! Pass it off to our own window click handler.

            case inContent:
                DoWindowClick(window, event);
                break;

In the title bar? Better track that window drag, and make sure to tell the Toolbox what the boundaries of the drag have to be.

            case inDrag:
                DragWindow(window, event->where, &qd.screenBits.bounds);
                break;

Oh, the window's close box? Track that drag, and if the mouse is released therein, call the close handler.

            case inGoAway:
                if (TrackGoAway(window, event->where))
                    DoCloseCommand();

Were there other places it could land? Plenty, but we don't have to worry about them in a really simple program. Handling the zoom box, or the expand/collapse box, or in later versions of the OS, the proxy or toolbar icons, was more trouble than it was worth!

        }
    }

Hey, a key was pressed! If the command key was down, treat it as a menu equivalent and handle that.

    static void HandleKeyPress(EventRecord *event)
    {
        if ((event->modifiers & cmdKey))
            HandleMenuSelection(MenuEvent(event));
    }

Okay, now for the more general cases of events:

    static void HandleEvent(EventRecord *event)
    {
        switch (event->what)
        {
            case mouseDown:
            {
                HandleMouseDown(event);
                break;
            }
            case keyDown:

autoKey? Oh, that's just if a key was held down.

            case autoKey:
            {
                HandleKeyPress(event);
                break;
            }

Update event! That's the result of [view setNeedsDisplay:YES] on an entire-window scale. If you don't call BeginUpdate() and EndUpdate(), even if there's nothing to actually do, you'll just keep getting update events ad infinitum.

            case updateEvt:
                BeginUpdate((WindowPtr)event->message);
                EndUpdate((WindowPtr)event->message);
                break;

Hey, the user put a disk in the floppy drive! If something went wrong, you'd better make sure they get the message, and be sure to pick some arbitrary point on the screen to show the error dialog at. Be sure to load up the Disk Initialization package first, and to unload it when you're done. Can't have it taking up memory you're not using.

The Disk Initialization package could be used for manually initializing floppies, verifying the formatting, and wiping them with zeroes. There were also a couple of extended routines in System 7.5+ for formatting disks with different filesystems.

            case diskEvt:
                if ((event->message & 0xFFFF0000) != noErr)
                {
                    Point       pt = { 100, 100 };

                    DILoad();
                    DIBadMount(pt, event->message);
                    DIUnload();
                }
                break;

Activate events just mean your window was brought forward or moved backward, usually by your own code.

            case activateEvt:
                break;

OS events had two subtypes: suspend (you're no longer the frontmost application) and resume (you're the frontmost again). There was also a "convert clipboard" flag in System 7+ which told the application to make sure the contents of the clipboard made sense, as they might have changed.

            case osEvt:
                break;

kHighLevelEvent is the generic dispatch point for Apple Events. AEProcessAppleEvent() calls through to any registered Apple Event handlers.

            case kHighLevelEvent:
                AEProcessAppleEvent(event);
                break;

A null event just means nothing's happening! Since nothing happens on its own in Toolbox-land, we check whether there's a frontmost window, and if there is, make sure any controls it contains get their share of idle time. This, for example, makes insertion points in text fields blink and indeterminate progress bars do their barbershop animation.

            case nullEvent:
                if (FrontWindow())
                    IdleControls(FrontWindow());
                break;
        }
    }

And finally, the event loop itself. Herein we find the infamous WaitNextEvent() function, System 7's answer to the inadequacies of the older, not-multitasking-aware GetNextEvent(). WNE fetched the next event from the event queue which matched the requested set of events, allowed the specified amount of time to background processes, and returned whether or not there was any event to handle.

If there was no event to handle, we do the same thing as we do for a nullEvent. I don't actually remember what circumstances would distinguish the false return from WaitNextEvent() from receiving a valid nullEvent, I'm afraid. In either case, loop until something says it's time to quit.

    static void RunEventLoop(void)
    {
        EventRecord         event;

        while (!gDone)
        {
            if (WaitNextEvent(everyEvent, &event, 30L, nil))
                HandleEvent(&event);
            else
            {
                if (FrontWindow())
                    IdleControls(FrontWindow());
            }
        }
    }

main
And finally, the ever-important entry point of the application, main() itself!

Oh. All it has to do is call the initialization and the event loop. Even in System 7 days, there wasn't usually much for main() to clean up once the event loop was done.

main() didn't take any parameters in the Toolbox universe. There was no command line and no concept of environment variables, so there was really nothing to pass. What little information the OS passed directly to the program was in that qd global.

    void        main(void)
    {
        Initialize();
        RunEventLoop();
    }

Conclusion
All I can say is, it really took a lot to get simple things done in the old days. But, it also felt a lot closer to the machine itself, with control possible over just about everything if you wanted it. Cocoa's really gotten away from that concept, as has programming in general. I don't miss the extra work, but I sometimes get nostalgic for the feeling of speaking the OS' own language.

That's all from me this week, I'll be back next week with a Friday Q&A. Meanwhile, enjoy Mike's article for this week, and as always, thanks for reading!

Did you enjoy this article? I'm selling whole books full of them! Volumes II and III are now out! They're available as ePub, PDF, print, and on iBooks and Kindle. Click here for more information.

Comments:

I remember having to get into just the tip of this iceberg a few years ago when I started working on Core Audio and Quicktime -- prior to 10.4 you still had to use Handles and FSRefs in order to do simple things like iterate over components.

I think you still have to use FSRefs if you want to write code that does an asynchronous filesystem copy in the background and callsback the client with progress reports (with FSCopyObjectAsync()); also the old Carbon File Manager remains the only way to "set" filesystem comments without ducking into Apple Events or scripting, which work, but have, at best, same-day service.
Then you had global variables. Generally frowned upon in any modern program of today, they were all but a requirement for Toolbox development when not using C++ or some other object-oriented language. There was just no other efficient way to get data where it was going. Certainly the global variable for exiting the event loop was a lot better than returning Boolean from every single event handling function in your code on the off chance an action might lead to a Quit.


This seems to be a criticsm of procedural programming languages, not the Mac Toolbox, and it's also an incorrect generalization; you are taking poor programming practices that many programs used and extrapolating them to all. There's no good reason to keep the standard AppleEvent handler UPPs as global variables, it's pointless. A boolean for exiting from your event loop ("gDone") can be replaced by any number of superior constructs.

Handling the zoom box, or the expand/collapse box, or in later versions of the OS, the proxy or toolbar icons, was more trouble than it was worth!


Care to elaborate? You got collapse/expand for free unless you asked to handle it specially. The proxy icon was something like an additional inProxyIcon: case, where you then passed the event to a Window Manager function named something like HandleProxyIconEvent(). Not worse than diskEvt handling, surely.

Of course, the Carbon Event Manager eliminated most of this tedium, while still allowing you to hook in at a low level if you needed it.
Minor quibble: the "open application" event corresponds more closely to the -applicationOpenUntitledFile: application delegate method. In fact, Apple pretty clearly documents how the standard Apple Events are translated into Cocoa: https://developer.apple.com/library/mac/#documentation/Cocoa/Conceptual/ScriptableCocoaApplications/SApps_handle_AEs/SAppsHandleAEs.html
Chris: I can only speak for my own experience, and my Toolbox skills were never what I'd call top-rate. I never could have said for certain whether UPPs needed to be saved over time :). Much of the code I used here was partially copy-pasted from multiple ancient projects. As for handling the various other window parts, I can honestly say I simply don't remember how the Appearance-aware Window Manager did things, and how much of the work it took over.

Ken: I didn't know about that document; thanks for the link!
Hi,

 another nice installment in the series, thank you! I have a few more corrections:

1) Stop bashing globals. They're the right tool for the right job. [NSApplication sharedApplication] is effectively a struct of globals. It also contains its own gDone-equivalent.

2) One great advantage of UPPs in fat binaries was that they let you mix-and-match emulated 680x0 code and PowerPC code in a single application (something the "Universal Binaries" of the PowerPC/Intel switch didn't support). So you could load a plug-in that is written in 680x0 into a PowerPC app or vice versa.

This is actually what Apple did: They only ported the most important system routines to PowerPC, and everything else still ran in emulation. Then with every system release, more was ported.

3) There is a usability reason that HiliteMenu(0) needed to be called: Menus were supposed to remain highlighted while a menu command was busy doing its work. E.g. if you selected "About" the Apple menu would not un-highlight until you closed the about panel. This was a nice bit of breadcrumbs-style feedback, reminding you where you got a window from.

However, it became problematic when alerts and dialogs started handling the menu bar (originally, you couldn't use menus while a modal dialog was up, not even copy and paste).

4) You should probably add a HiliteWindow() call to the activate event, otherwise your app is going to look really weird. Well, it would, if you had any SelectWindow() calls anywhere that re-order windows on clicks (which also needed to be done manually on content- and title bar clicks).
Uli: Thanks for all your corrections!

1) I admit, I've been a little cruel to globals, mostly because I've been bitten by their misuse in the past.

2) The Mixed Mode Manager was extremely clever in making things transparent that way, and I duly salute its ingenuity :). But in my opinion, the extra complexity it added to any code that used callbacks (and there was a lot of such code in the Toolbox) was only a reasonable tradeoff versus recompilation (such as in PPC/Intel) when space limits were considered.

3,4) I'd forgotten about both those bits :). As I mentioned, it's been a long time since I used this stuff, and I only wish I could justify the time it'd take to become proficient again.
There were, of course, several main concerns in Mac OS Classic which prevented the plain recompliation of all of the code for PowerPC

1) Much of the toolbox was written in 68k assembly
2) Much of the toolbox was written in such a way that it had to be shared between all running programs
Great article. I have spent very little time with old-style Mac programming, so this was a great little taste.

Re: four-character character literals:
I think you're a little off-base here; you can definitely find these in Cocoa, and I've never had a compiler complain about them (though admittedly I probably have only used them in Apple gcc/clang).

For example, you can specify OSTypes with four-character literals. Also they're used extensively in Cocoa's support for Apple Events; types like AEEventClass and DescType are found in NSAppleEventDescriptor, NSAppleEventManager, SBObject, etc.

Comments RSS feed for this page

Add your thoughts, post a comment:

Spam and off-topic posts will be deleted without notice. Culprits may be publicly humiliated at my sole discretion.

Name:
The Answer to the Ultimate Question of Life, the Universe, and Everything?
Comment:
Formatting: <i> <b> <blockquote> <code>.
NOTE: Due to an increase in spam, URLs are forbidden! Please provide search terms or fragment your URLs so they don't look like URLs.
Code syntax highlighting thanks to Pygments.
Hosted at DigitalOcean.