This previous weekend, while helping Pukka to fix a mysterious coordinate calculation bug in Fool’s Errand intro, I decided to experiment with one weird idea: How about adding support for System Extensions and Control Panels? Therefore, yesterday I added the After Dark ‘cdev’ control panel to MacPaint app’s build settings to be included in the System Folder, and took some time to look what it would take to make it work.

Basically, INITs and cdevs are loaded during the startup by Mac OS by checking a few bits (such as file type, visibility, etc), opening the resource map of each file, and executing the first ‘INIT’ code resource found. At this experimental stage, we don’t yet have full INIT/cdev scanning, but instead have just a hard-coded filename to load INIT code from which works ok enough for this test case.

Our boot-time heap wasn’t quite properly set up yet, as the boot stack was too high, not leaving enough space between MemTop and BufPtr, which on the first attempt of running the INIT caused After Dark’s InitZone call to actually end up at top of the stack, causing all sorts of havoc. Besides that, there were a ton of bugs/improvements:

• 68k-to-68k trap calls did not work correctly (i.e. calling a trap patched with 68k from 68k code)
• InitZone parameter block handling was completely broken from the last year’s trap refactor, so it was fixed
• Memory Manager’s Handle validation had not taken into account the case where handle’s “relative handle” field would point to invalid memory location in some “fake” handles (as one in After Dark), for which validation was added
• Empty stubs were added for Notification Manager calls to allow After Dark to call them during activation/deactivation of screen saver
• Locking of 68k mutex was not done early enough, so After Dark’s VBL handler was making the initial call Launch trap to have invalid register values
• After Dark uses grafProcs to call QuickDraw, so one missing case used by it was added (ovalProc for ovals)

After that bunch of minor tweaks, we have now at least After Dark working nicely!

Although the INIT code itself works, there is not yet any UI for displaying cdevs (Control Panels), so modules cannot be configured yet, and thus only the default “Starry Night” built-in module is displayed.

There’s also a short video to demonstrate After Dark running:

Although not very usable feature at the moment, this will definitely allow interesting possibilities in the future, as theoretically we should be able to run applications requiring certain application-specific extensions to be present in the System Folder. This experimenting with After Dark has anyway allowed us to find a number of important bugs (as previously mentioned), and also works as a very good “compatibility” test to see how well our Toolbox API and 68k implementation can handle one very quirky control panel, as it patches a ton of toolbox traps and does a lot of low-level stuff which all appear to be compatible with our emulation.

This week we reached a major milestone! From the encouragement on the work on HyperCard on 68KMLA forums, we got inspired to attempt tackling one major obstacle that was preventing it from working – file system write support!

As we mentioned earlier last year, when we last fall added the file system API, we did so using abstraction on top of AppleDouble file format. As Mac file system has the special dual nature of being split to data and resource forks, the AppleDouble format helps managing this in the emulated file system. Data fork, as mentioned earlier, is saved directly in its own file, while resource fork is merged with other file metadata in the AppleDouble header file. With this distinction, adding support for writing to data forks was really just trivial mapping of file manager write calls to the data file. Resource fork writing is a bit tricky, but we currently force the AppleDouble headers to keep resource fork as last element in the file, which allows us to offset also write operations from the beginning of resource data, and allows us to expand/shrink file easily, as end of resource fork is the actual end of header file. (One side note, setting EOF of files doesn’t have a standard C file operation, so we’re for now using the POSIX extension “ftruncate” which works at least on Mac OS X, but may lead to portability issues.)

Currently there’s very limited set of features for writing to file system implemented, basically just “Create” and “Write” file manager calls. For example, resource manipulation does not yet work as it will require a number of routines added to the Resource Manager to support it. However, one test application that directly benefited from this new feature was our beloved MacPaint, as it requires two scratch files to be present on startup volume/folder to function. This was nice, as it allowed us to add a few other missing features, such as PackBits (which also benefitted other apps and picture recording).

Although all tools don’t yet work perfectly, most of basic drawing functions are solid enough to not only allow drawing, but the experimental file writing also allows saving of MacPaint files, which appear to be readable by the Preview application in development host system Mac OS X 10.12!

Additionally, many other applications have seen improvements from this, as now we can save Civilization games and continue playing them! (Although Standard File package is still using hard-coded SFReply records).

The past week we’ve been busy working on optimizations on the 68k CPU and various improvements to the toolbox API. One of the improvements has been adding support for Apple’s MacApp MethodDispatch routine, which is used by Object Pascal and C+- (yes, that’s “C plus minus”, Apple’s proprietary C++ version which was briefly out before official C++ came to Mac). Basically, MethodDispatch is the virtual function table dispatcher commonly seen in C++ programs, but instead of being bundled in the application, it is actually implemented by the operating system as a toolbox trap – although with special calling convention, which cannot be called directly through A-line traps, but needs to be invoked through jump from 68k code. It handles selecting the correct implementation for class functions which can be inherited and overridden.

With the new MacApp method dispatching support, we should now be able to try running any MacApp-based applications – one cool such application, which is written in Objective-Pascal, is the first version of Adobe Photoshop:

Most of the dialog boxes appear to display (almost) correctly, although color picker, text input and popup menu support are missing so there’s little that can be done here:

And just to test out how it works, we did a quick “Hello world” drawing using Photoshop drawing tools inside MACE:

Although implementing the method dispatching routine was fairly trivial, some time was taken by other features used by Photoshop, such as crude International Utilities/Script Manager support, PrGlue dispatcher for printing (or as in current case, telling application that printers are not available), proper Gestalt support, and various bug fixes even to some fairly low-level things as handling of empty regions in UnionRgn calls, rewriting the 80-bit “Extended” floating-point format conversion in SANE, etc… A few other test applications did benefit of those additions, but we’ll post more about them later!

Ps. In case you are interested how MacApp-based application code looks like, the original Object Pascal source code for the first Mac Photoshop was at some time ago published as open-source and is available on github: https://github.com/amix/photoshop

After some effort, we now have basic shape drawing working in MacDraw. One fun part of this was implementing FixDiv, which uses the binary long division algorithm like the original one on Mac. MacDraw uses this routine excessively to handle document grid sizing and coordination of all shapes drawn on the canvas.

There was also a minor bug in MenuSelect causing zero item selections in MacDraw’s custom menus, but after a quick fix we can now also use them.

Contrary to many other applications, MacDraw uses a different approach to implementing custom menus – instead of providing a MDEF and assigning it to the menu using the menu definition proc ID in resource, it has its MDEF code as part of the program, and creates a fake handle which it assigns to the menu manually. Because of this, we had to tweak the memory manager handle operations to silently fail on such handles, and instead return the proper memAZErr (-113) in D0/MemErr.

Still no text drawing though, and there’s minor tweaking required for line drawing, but it’s looking promising at the moment.