Monthly Archives: February 2023

5 year anniversary! And happy new year 2023

It feels crazy to think that today it’s been a whopping 5 years since the project got rebooted to its current iteration! Sorry again for slow progress between late summer and the Christmas holidays, a lot of things going on so we have had to juggle between different things and try to prioritise them. But past couple months have been again quite productive, and we have some nice updates to share:

File system improvements

One important feature that was added to NativeFS has been ability to create directories using DirCreate, something that not only installer applications, but also some apps generally depend on. It’s still not perfect (for example extracting the THINK Pascal installer archives made with SmallerInstaller causes the directories to get a bit haywire), but basic use cases such as creating Desktop & Trash folders seems to work.

Also, some groundwork has been laid to support deletion of files, as the AVL tree routines for node deletion were added – but actual deletion of file system items is still a work-in-progress task, which needs to be done carefully and is not to be rushed considering errors and bugs in this code might cause a bit headache to the end-users.

And also, when doing test runs on some Installer VISE installers, we found out that the NativeFS DRVR’s Control handler had used incorrect csCodes for the logical/physical drive icons (we were using 0x16 and 0x17, while correct ones were 0x15 and 0x16), this tiny mistake caused the whole installer to crash…oops! Another quite important bug fix was adding missing IODone call to the NativeFS DRVR’s Status handler, as that caused basically the entire Device Manager to hang…

List Manager multi-select

Not really a tough feature to implement, but one that has been a TODO for long time, list manager multi-selection has been finally implemented. A very good test case for this has been ResEdit, as it supports multi-selection both in the 2-dimensional icon grid, and in the 1-dimensional list view

List multi-select in action in ResEdit

SeedCFill & CalcCMask

Also as part of filling some missing traps, SeedCFill and CalcCMask were implemented. Like the aforementioned list multi-selection, ResEdit again was a wonderful test case for this, as the color editor PACK 1 in it uses these for the color flood-fill tool:

SeedCFill & CalcCMask in ResEdit’s PACK 1 color pixmap editor

Technically, the SeedCFill and CalcCMask actually call SeedFill and CalcMask, and to do that, they first map the color pixmap passed to the routine to a 1-bit image by using a custom color search proc to a temporary bitmap, which replaces matching colors with 0 and non-matching with 1 (i.e. doing floodfill on certain color gets flooded only to exactly matching pixels, while being bound by any other colors)

Application zone teardown updates & fixes

A bunch of fixes and missing features were added to the InitApplZone & SetApplBase, which makes doing transfer/shell launch to other apps way more reliable:

  • Font caches are flushed in SetApplBase and InitApplZone. Sometimes recently used font might be left cached in the application zone, and doing Launch/Chain to new app might accidentally use the cached (but not obliterated from memory) font, causing random crashes.
  • All color grafports, palettes, and VBL tasks located in application zone are disposed/closed

These are quite important fixes, as the aim is to provide soon the much-promised “generic” MACE runtime, which would contain an application shell (i.e. Finder replacement) to reliably run applications from

Communications Toolbox foundation

Weirdly, the Comm Toolbox has a bunch of very non-related routines included, which some applications seem to depend on. The first one we encountered was some Installer VISE installers using the CountDITL selector to get number of items in dialog item list, which the designers of the toolbox interestingly put as part of the Communications Toolbox. Also, the dispatcher mechanism they use is a very non-standard one (they basically pass parameter block pointer in A0 register, which contains the actual arguments of the routine – which might vary in size and number depending on the API call – and expecting result in D0 register). Although there’s just this one call implemented, the foundation for Comm Toolbox was done so that other routines can be added as needed later.

SoftFPU support

Even though Pekka is working hard on the 6888x FPU support, we did interesting experiment to see how well the SoftFPU control panel would work on MACE. Basically, that utility hooks as the F-line exception handler, using SANE calls to the actual math, and passing those results as emulated FPU instruction results. So, emulation inside emulation, kind of. Getting this done was actually quite easy, as the SANE emulation already covers so many calls, so we just had to hook up the F-line exceptions, and implement two missing SANE FP68K selectors (FNEXTX and alternate FX2X), and we were able to run a test app written in THINK Pascal with native 68881/68882 code generation with correct results:

SoftFPU running on MACE

In the above test application, the Compile Options dialog shows that 68881/68882 codegen is active, the code contains math for 9 / 2, and the Text window shows the result of 4.5 (in scientific formatting, as the Dec2Str routine still has some unfinished features)

Some updates to get Civilization 2 to work…almost

One cool game we also have been wanting to play for a long time is legendary Civilization 2, and we took a small steps in getting closer to being able to play that game someday soon:

  • Time Manager’s “hidden” microseconds trap (A093). This one was a bit hard to find documentation for, but we were able to produce close-enough results to keep Civilization 2 happy, which uses it for some reason. Basically the call returns 64-bit microseconds counter, with low 32 bits in D0, and high 32 bits in A0 register.
  • Added DriverGestalt (43) status call to NativeFS, and dgDeviceType (‘devt’) to it, which Civilization 2 seems to use to detect the media type of volume it was launched from (to see if it was started from the CD-ROM or Hard Disk)
  • CopyPalette trap handler (only partial implementation, but seems to be enough for Civ2 startup)
  • Added some more fake process manager data to GetProcessInformation results, in this case basically faked the processAppSpec by using HGetVol to get active directory, and CurApName from low mem…really dirty, but luckily enough again for Civ2. When process manager is someday added, this hack will be replaced by a proper implementation

With these done, we did get the game to progress quite far already:

There are still some big issues in the game, such as missing for some reason the entire backdrop images from the start-up menus, fonts looking ugly because truetype is not implemented (bitmap fonts have the pixelation on non-exact sizes), and color picture recording has still missing features causing city screens to crash.

Newer Technology GURU: “GUide to Ram Upgrades”

One other nice application that we were also experimenting is NewerTech’s GURU, which not only at the time was nice tool for checking out what RAM/VRAM upgrade options you had for Macs, but also doubles as nice mini-encyclopdia of various Macs, clones – and even printers – with their specs and nice color icons included.

GURU running on MACE

To get GURU running, there were a couple things needed:

  • We had again to temporarily tweak the GetProcessInformation’s processSignature (which sadly also WarCraft 2 uses, so both this app and that game need a proper implementation for this
  • Interestingly, the tiny and very harmless-looking popup menu triangles (seen in the screenshot on right-hand side) were actually 16-bit RLE-compressed pixmap images, so we had to implement both the 16-bit pixel translator, and the RLE decompressor in the Color QD Picture bits opcode handler
  • Also, the About Box in GURU used TESetText for styled textfield, which was previously a TODO, so that had to be implemented (it basically just replaces existing styles with a single style run)

MacWrite (finally) works now!

We also took some time to dig into other long-time known issues, one of which the MacWrite crashing on newline issue. Interestingly, two separate bugs did surface:

  • The biggest issue was that doing post/pre-increment of registers in middle of execution was not handled properly, in MacWrite’s case the offending instruction was ADDA.W (A0)+,A0, where value from memory pointed was read, A0 was supposed to be incremented by the word size (2 bytes), and the result added to A0 register. The bug caused the increment to get missed before adding the value to A0, causing it to be 2 bytes too low. This caused a bunch of havok in MacWrite, as the app used its own data structures to keep style data for rows, and this made – among memory corruption – the font size to go to a whopping 255 pixels when adding a new line on keyboard
  • The “justify” text alignment was one of rare cases where spExtra of GrafPorts was used to do sub-pixel (fixed-point) spacing of text, so we didn’t notice before Font Manager was missing a byteswap in handling of spExtra, which caused the cached spExtra to get totally crazy values
  • A rare case of resource map not being initialized properly surfaced on document saving, which was fixed
  • And also, document saving in MacWrite seems to depend on IUSetIntl trap (to apparently write localization data to the document), that trap had to be implemented

With these fixes, MacWrite finally works:

MacWrite works now!

Strategic Conquest

Although most of the features we had tested in Strategic Conquest were working, the map drawing had glitches: polygon lines were missing. We did some investigation about this, and found out that MapPoly trap only mapped the polygon points, but totally forgot to do the mapping for polyBBox. This in turn caused the polygon drawing to assume points were out of range, and fail to render anything in most of cases. With this fixed, the map seems to work fine now (at least in the B&W version):

Strategic Conquest MapPoly works now!

Arkanoid has sound!!!

One issue that was also mysterious for long time was why Arkanoid had no audio, while (almost) every single other black & white game/application had sound. After some researching, we found out that being super smart (and bizarre), instead of the standard VLB queue item (like 99.9% of games), or the not-so-common Lvl1DTVIACA1 interrupt handler (which Fokker Triplane does), Arkanoid actually hooks directly to the AutoInt1 vector on the 68K exception vector list, which normally would be handled by the Mac OS to invoke the VBL interrupts. After adding a level of indirection to the AutoInt1 vector (and adding possibility to later add other interrupt handlers to this level), Arkanoid sounds works perfectly! Sadly we did not have time to get a nice video capture of this for this update 😦

THINK Pascal zone check fixed

Another long-time issue that had been bothering us was that THINK Pascal kept saying in debug mode all the time that all zones were damaged, so we finally took time to dig into the THINK Pascal code and look for their zone check code and see what it was doing. Luckily, the only part that was failing was the trailer check, which checked that the bkLim block was supposed to be always 12 bytes in 24-bit mode, was getting triggered because when GrowZone was adjusting the zone trailer, it accidentally wrote 0 (zero) as trailer size, instead of the expected 12 (ironically InitZone did set it up properly). Although this bug had no functional effects to anything else but the zone check, the THINK Pascal was way more happier now, and looks like the LightsBug debugger is working perfectly, zone browser now included!

LightsBug working nicely after zone checks pass

Other bug fixes

There were a couple bugs/missing features that were fixed, with varying effects:

  • Fixed mask overflow in 32-bit pixel data stretching blitter, and implemented TODOs for some 32-bit pixel data colorization variables – These caused black buttons in some F/A Hornet 2.0 dialogs
  • ScrollRect passed wrong offsetPt to ShieldCursor – this caused some cursor artifacts (this bug did not manifest itself until after the actual software cursor code was added recently)
  • The WDEF 0 (standard window defproc) did not offset the grow box drawing correctly when a window grafport was offset to a non-zero origin (like in PhotoShop) – this caused the bottom-right growbox to get drawn in wrong position on the desktop)
  • 1-bit cursor data saving was missing from SetCCursor – This caused the cursor update to not get triggered if doing InitCursor or SetCursor after doing SetCCursor
  • Fixed issue in RestoreBits, where the menu bar was invalidated in RestoreBits even though it was restored, which in turn caused recursive call to DrawMenuBar when had a menu highlighted (the kMBDFMessageHilite message in MBDF code called RestoreBits which in turn did InvalMenuBar call, which caused the DrawMenuBar to get triggered from the second HiliteMenu call inside the DrawMenuBar) – Fixes the issue of MenuBarPattern code getting stuck in infinite recursion in Civilization, as it tried to double-patch itself from the recursed DrawMenuBar call
  • Added missing call to SystemEvent in GetNextEvent, which caused some events, such as key press events, to not get passed properly to Desk Accessories – Fixes the issue of Key Caps DA not receiving text input
  • Allow BlockMove to receive negative lengths. After researching how the List Manager in some 68k Mac OS versions handle the case of having zero rows but non-zero columns, it seems that they do pass negative lengths (maybe on purpose) to BlockMove with no side effects, and BlockMove ignores those – Fixes the unncessary negative length assertion in ResEdit with empty resource files
  • vRefNum case was missing for GetWDInfo calls in NativeFS, which caused ResEdit to not be able to open files from root directory of a volume
  • Do proper setup of stack pointer on application launch to BufPtr, because this was missing (and SP was left in boot-time default), it cut amount of available memory to apps down to 75% of total RAM! This also surfaced another bug that the startup-time boot grafport (used to draw any stuff during INIT loading etc) was not closed, so any code assuming that all PortList entries were valid were randomly crashing
  • Added style support for TEDelete, so using backspace key in SimpleText works now

What next?

It’s always hard to estimate what will be done and when, but there are still some key goals we’re aiming in the near future:

  • To get the public binaries finally updated after we have time to figure out not only a nice, cheap and practical CDN or other way update the files
  • When electricity prices go down (which got crazy during winter, so had to turn off most of extra computer equipment), work on automating the builds for different platforms, instead of manually building everything (even though CMake does a lot of the hard lifting)
  • Get Sound Manager 3.x to work, so we can start putting out the color version too
  • Try to make sure ReserveMem & NewHandle combos work as expected, right now this is causing THINK Pascal to run out of memory – this is something we want to make sure works, before we move to:
  • 32-bit memory manager. Something that’s needed not just to get more than 8 megabytes of accessible memory, but also to get some picky games (such as Doom) to run, AND to allow:
  • Direct memory access support, for avoiding the overhead of virtual memory interface, potentially speeding up emulation quite a lot. This requires 32-bit memory model to work efficiently.
  • Update build system to include creation of the System file from resource scripts – it’s still being handled as binary file that’s edited using native resource tools, which is not only cumbersome, but also makes it difficult to have variants of the System file for different system types – when we at some point do add process manager support, the way for example desk accessories and fonts are handled needs adjustment, as in process manager-aware (System 7 and later) those are external files instead of ones in System like in System 6 and earlier
  • Work on the so-many missing features (which do also cause known bad behaviour or crashes)
  • Debug and research the problems and crashes which we know about, but haven’t yet figured out what is causing them (such as F/A Hornet and F117A drawing issues, ball missing bug in Loony Labyrinth, timing issues in Monkey Island and Prince of Persia 2…)
  • Prepare a Discord server for public access
  • Many, many, ….many other things

We will as always update our progress on this blog, hopefully a bit more frequently. Your support and encouragement has been very inspiring, and we’re trying as soon as possible to share more and more of the stuff we’ve been making with you.

Full list of changes since last post

2023-02-15 14:56:38 +0200 • Fix 24-bit zone trailer & add bkLim validation
2023-02-15 03:45:35 +0200 • Add dummy ADBOp trap (for SoftWindows)
2023-02-14 21:47:21 +0200 • Expose & use low-level AutoInt1 vector for VBLs
2023-02-13 23:34:21 +0200 • Add LMKbdType and use it for Gestalt & SysEnvirons
2023-02-13 00:27:16 +0200 • Also adjust polyBBox in MapPoly
2023-02-12 17:40:45 +0200 • Style support for TEDelete
2023-02-12 17:39:11 +0200 • Disable font width table logging in debug mode
2023-02-08 23:48:12 +0200 • Merge branch 'master' of
2023-02-08 23:47:43 +0200 • Remember to close the startup-time GrafPort
2023-02-08 06:12:54 +0200 • Merge branch 'master' of
2023-02-08 06:12:50 +0200 • Handle TEMeasure start in middle of style run
2023-02-08 06:12:02 +0200 • Don't return random value if FindFolder fails
2023-02-08 06:11:09 +0200 • Some cleanup in HighLevelFSDispatch
2023-02-08 06:10:30 +0200 • Don't ignore opIFore and opIBack
2023-02-06 02:47:11 +0200 • Reset stack pointer A7 in non-multitasking Launch
2023-02-06 00:48:02 +0200 • Handle vRefNum case in GetWDInfo
2023-02-05 23:15:40 +0200 • Implement IUSetIntl selector for IntlUtils Pack6
2023-02-05 22:29:59 +0200 • Fix rare case of res map handle not initialized
2023-02-05 22:26:45 +0200 • Fix ADDA.W/L postincrement same as destination bug
2023-01-30 12:28:55 +0200 • Fix byteswap bug on spExtra handling in FMSwapFont
2023-01-29 01:26:40 +0200 • Add dummy stub for HMRemoveBalloon (for Excel 3.0)
2023-01-28 08:35:17 +0200 • Implement TESetText styled textfield case
2023-01-28 00:35:35 +0200 • Some process manager placeholder hacks for GURU2.6
2023-01-28 00:34:03 +0200 • Use 16-bit RLE decoder for PICT packtype 3
2023-01-28 00:33:37 +0200 • Add 16-bit RLE decompressor for QD Pictures
2023-01-28 00:32:33 +0200 • Add 16bit direct to indexed color pixel translator
2023-01-26 21:06:19 +0200 • Merge branch 'master' of
2023-01-26 21:06:17 +0200 • Fix 68k Bxx opcode with 32-bit extension
2023-01-26 01:07:16 +0200 • Fix return value of Microseconds trap
2023-01-25 14:45:34 +0200 • Hack face processAppSpec in GetProcessInformation
2023-01-25 13:52:55 +0200 • Partial CopyPalette (0xAAA1) trap implementation
2023-01-25 03:32:02 +0200 • Add dummy Lower/Strip/Upper/StripUpperText handler
2023-01-25 03:01:56 +0200 • Handle 'devt' driver gestalt in NativeFS
2023-01-25 03:00:19 +0200 • Return controlErr on non-support NFS Control codes
2023-01-25 02:41:18 +0200 • Add DriverGestaltParam & dummy handler in NativeFS
2023-01-25 02:39:21 +0200 • Fix broken return from NativeFS Status handler
2023-01-25 02:38:02 +0200 • Implement Microseconds trap (A093) for Time Mgr
2023-01-23 03:10:20 +0200 • Expose some rsc manager stuff for SetResourceSize
2023-01-23 03:09:21 +0200 • Merge branch 'master' of
2023-01-23 03:08:20 +0200 • Log also 68k program counter in trap debug msgs
2023-01-23 03:07:21 +0200 • Update TimeManager.c
2023-01-23 03:06:42 +0200 • Work on SetResourceSize
2023-01-17 22:25:49 +0200 • F-line handler for M68kHelper glue module
2023-01-17 22:04:05 +0200 • Implement FNEXTX and alternative FX2X selector $10
2023-01-17 22:01:38 +0200 • Fix f-line dispatcher installation in 68k cpu code
2023-01-17 00:27:06 +0200 • Add CountDITL, and a VERY bizarre A08B dispatcher
2023-01-16 14:06:31 +0200 • Fix invalid icon csCode in NFS + add assert checks
2023-01-16 13:54:23 +0200 • Fix byteswap bug
2023-01-16 13:41:18 +0200 • Remove VBL tasks in app zone in InitApplZone
2023-01-16 13:34:27 +0200 • Dispose palettes in app zone in InitApplZone
2023-01-16 13:22:09 +0200 • Close all open grafports in InitApplZone
2023-01-16 13:21:38 +0200 • Flush font caches in InitApplZone
2023-01-16 13:02:15 +0200 • Flush font caches in SetApplBase
2023-01-16 13:01:45 +0200 • Add SLCheckApplZoneAddress, for app zone cleanup
2023-01-16 13:01:04 +0200 • Detect use of ResErrProc in resource manager
2023-01-16 13:00:29 +0200 • Add (partial) font cache flushing in Font Dispatch
2023-01-16 12:59:34 +0200 • Implement FCMPL selector in SANE FP68K trap
2023-01-09 23:32:54 +0200 • Implement QDExtensions's GDeviceChanged selector
2023-01-09 23:22:39 +0200 • Implement QDExtensions's OffscreenVersion selector
2023-01-08 23:17:28 +0200 • Added AVL Tree node deletion routines
2023-01-08 13:49:24 +0200 • Implement SeedCFill + CalcCMask QD traps
2023-01-06 16:57:46 +0200 • Fix handling of zero & negative BlockMove lengths
2023-01-06 11:24:27 +0200 • Check VCB flags in SetFileInfo/SetCatInfo
2023-01-01 17:46:21 +0200 • List Mgr's LClick multiselect (shift+cmd) support
2022-12-29 11:06:57 +0200 • Implement RsrcMapEntry trap handler
2022-12-20 00:02:35 +0200 • Don't return mouseMovedMessage unless app4 bit set
2022-12-18 14:05:14 +0200 • Implement SysEdit trap handler
2022-12-18 14:04:49 +0200 • Fix issue of GetNextEvent not calling SystemEvent
2022-12-17 00:25:55 +0200 • Fix unneeded invalidate of menu bar in restorebits
2022-12-14 21:29:16 +0200 • Save 1-bit cursor data in SetCCursor
2022-12-14 16:21:12 +0200 • Fix wDrawGIcon incorrect offset in nonzero origin
2022-12-13 21:49:42 +0200 • Proper fix for horizontal stretch pixel masking
2022-08-24 02:02:08 +0300 • Fix ShieldCursor offset in ScrollRect
2022-08-22 23:21:44 +0300 • Improve relative mouse mode host cursor behaviour
2022-08-22 22:11:12 +0300 • Merge branch 'master' of
2022-08-22 22:11:02 +0300 • Add dummy MIDI driver module (for MIDI Manager)
2022-08-22 22:09:16 +0300 • Fixes to 32-bit direct-to-8bit color QD blitter
2022-08-22 22:08:03 +0300 • Fix mask overflow in 32-bit horizontal CQD stretch
2022-08-21 13:30:13 +0300 • Fix sound manager header
2022-08-05 03:58:52 +0300 • Add empty dummy MIDIGetClients selector
2022-08-05 02:58:50 +0300 • Refactor filesystem AVL tree code & add validation
2022-08-05 02:57:45 +0300 • Cleanup SoundDispatch & add SndDispVersion handler
2022-08-01 17:28:51 +0300 • Merge branch 'master' of
2022-08-01 17:28:46 +0300 • Implement DirCreate file manager trap
2022-07-31 23:09:20 +0300 • Re-export traps & selectors for status update