Spring 91 - MACINTOSH Q & A
MACINTOSH Q & A
MACINTOSH DEVELOPER TECHNICAL SUPPORT
Q The TrapAvailable function listed in the "Compatibility Guidelines" chapter of Inside Macintosh Volume
VI contains code that checks the size of the trap table. Do I have to do this if my application doesn't
support the old 64K ROMs?
A Yes, you do. With the introduction of the Macintosh Plus and 128K ROMs, the trap table was
divided into the OS trap table and Toolbox trap table. The number of trap table entries was
increased and the format of those entries was expanded. For additional information on this
expansion see the "Using Assembly Language" chapter of Inside Macintosh Volume IV.
However, the 64K to 128K ROM trap table expansion is not what the code in the
TrapAvailable function is checking for. The Toolbox trap table was expanded yet again when
the Macintosh II was introduced. The Macintosh Plus and SE use the smaller Toolbox trap
table on pre-7.0 systems. To be sure your application is compatible with these machines on pre-
7.0 systems, check the size of the Toolbox trap table.
Q What are the guidelines for determining how much of an image CopyBits can copy to a Macintosh pixel
map at one time, given a particular set of characteristics for the source map and the destination map and
given how much stack space is available? For example, say that we have an 8-bit-deep pixMap to be
copied to a 32-bit-deep pixMap using the ditherCopy mode and expanded by a factor of 4, and we have
45K of stack space.
A CopyBits' stack requirement depends on the width of each scan line (rowBytes). The rule of
thumb is that you need at least as much stack as the rowBytes value in your image (which can
be huge with 32-Bit QuickDraw), with the following additional modifiers: add an additional
rowBytes for dithering; add an additional rowBytes for any stretching (source rect != dest rect);
add an additional rowBytes for any color map changing; add an additional rowBytes for any
color aliasing. The stack space you need is roughly five times the rowBytes of your image. In
general, you're better off processing narrower scan lines. Reducing the vertical size will not
affect stack requirements. Narrow, tall bands (if you can use them) will reduce the stack
requirements.
Q Where can I find documentation on how to write a Macintosh printer driver equivalent to the
ImageWriter® or LaserWriter driver? In particular, how are Printing Manager and QuickDraw
commands translated into calls to the printer driver?
A DTS's "Learning to Drive" document and "SampleWriter" source code, available in
AppleLink's Developer Support folder and on the latest Developer Essentials CD, are helpful
references.
Q How can I determine what hardware device is driven by a particular Macintosh gDevice? I can call
GetDeviceList and GetNextDevice to get the driver reference number of each gDevice but not the
hardware ID. The system 'scrn' resource for the hardware ID of each device doesn't give me the driver
reference number. The device list produced by calling GetDeviceList and GetNextDevice isn't always in
the same order as the 'scrn' resource slot information.
A The following code shows how to obtain the slot number:
DCEHand = (AuxDCEHandle) GetDCtlEntry(DevInfoList[index].gdRefNum);
DevInfoList[devCount].gdSlot = (*DCEHand) -> dCtlSlot;
/* Get slot number. */
Once you have the slot number, you can call the Slot Manager to get the board name and other information you may need to identify the device, as in the following code:
/* Get the board name from the Slot Manager's board sResource. */
spB.spSlot = DevInfoList[devCount].gdSlot;
spB.spID = 0;
spB.spExtDev = 0;
spB.spCategory = 1; /* board sResource sRsrcType value */
spB.spCType = 0;
spB.spDrvrSW = 0;
spB.spDrvrHW = 0;
if (! SNextTypesRsrc(&spB) ) {
spB.spID = 2; /* Found board sRsrc now, get the sRsrcName. */
if ( ! SGetCString(&spB) ) {
/* Let C unravel its own strings. */
for (count = 0;
DevInfoList[devCount].bdName[count+1]
= *((char *)(spB.spResult)+count);
count++);
DevInfoList[devCount].bdName[0] = count;
}
}
Q When I draw a 32-bit Macintosh 'PICT' image from a file to an 8-bit port via an off-screen
GWorld, I use dithered mode in the CopyBits call and the results are quite impressive. If there's
not enough memory to allocate the GWorld, I draw the image directly to the port. But since
there doesn't seem to be any way to tell QuickDraw to use dithered drawing mode, the image
looks horrible. Should I use dithered mode instead of source mode? I don't want to try to parse
the 'PICT' myself, but I thought that maybe a QuickDraw global could be modified in my
StdBits proc to force dithered drawing for that operation only.
A You can install a bitsProc bottleneck procedure to get all the CopyBits calls
when the picture is being played back. One of the parameters to the
bitsProc call is the mode. You can install a procedure that passes
ditherMode to the original StdBits proc.
Q Why does a call to gestaltNuBusConnectors return zero slots when there actually is
a NuBusTM card or a PDS (Processor Direct Slot) card in the Macintosh IIsi slot?
A A call to gestaltNuBusConnectors returns zero slots because there's no way
to determine if there's a NuBus card or a PDS card in the Macintosh IIsi
slot. Gestalt can't assume that there's always a NuBus or PDS slot, so it just
says there's no slot. However, Apple recommends that you always use the
Slot Manager instead of Gestalt to search for cards, after first checking to
see that the Slot Manager trap is implemented. The Slot Manager will safely
do all the necessary work for you whether a card with a valid declaration
ROM is installed or not, and you can search for the card using a variety of
criteria. This technique will allow you to locate NuBus cards, but has the
added benefit of being able to find PDS cards that contain declaration
ROMs.
The Macintosh SE/30, IIfx, and LC have the Slot Manager and PDS slots.
It's safe to assume the Slot Manager will be resident in all future machines
that have either NuBus or PDS slot capability.
Q Do you have any available tools or test programs for testing how well our Macintosh
application responds to core Apple events? We need a test application that will, for example, send events to open a document and print it within our application.
A One of the best tools to test Apple events like 'odoc' and 'pdoc' is the
Finder®. The System 7.0 beta CD version of the Finder provides excellent
'odoc' and 'pdoc' testing capabilities. That's what most of DTS uses to test
sample applications.
There are no tools designed specifically to send Apple events to an
application as a test. Try using some of the DTS sample applications on the
Developer Essentials disc to test your application. One of the easiest to use
and adapt is the TrafficLight 3.0 sample. It already will send its own
MoveWindow Apple event to any application, allowing you to see how your
application is handling "foreign" events.
Also, you can adapt the CreateAndSendAppleEvent function in
AppleEvents.c to send any type of event you'd like, which allows you to test
events that may be specific to your application.Q Why do I get a bomb when I create a Macintosh filename starting with a period (.)?
A Macintosh filenames are not allowed to begin with a period, to avoid possible confusion with
driver names, which must begin with a period. (This restriction does not apply to folder names.)
Ideally, the Finder should catch this possible error and require the file to be renamed, but it
doesn't. Future versions of the Finder should catch this potential problem, but until then users
must remember not to begin a filename with a period. See Macintosh Technical Note #102,
HFS Elucidations, for details.
Q Can the refNum returned by FSOpen ever be 1? What is the range or format of legal refNums?
A Macintosh file reference numbers (refNums) are currently positive and that means any positive
number, including 1. This doesn't mean that they won't change in the future, however. To
maintain system compatibility, use refNums only as they're intended to be used.
Q Calling Create from an INIT causes the INIT to be called twice if it alphabetically follows the file it
creates. Is there a workaround?
A This particular limitation of the startup process occurs when you create files in the same folder
as the INIT that alphabetically precede your INIT file. HFS orders files alphabetically, and the
startup process does its thing by incrementing the ioFDirIndex field of a PBGetFileInfo call.
Your INIT file (for example, file #5) is getting opened, your INIT is called, and it creates a file.
The file it creates, for example, is file #3--that is, it's alphabetically before your INIT file. Now
your INIT file is file #6, and when you return, it increments the ioFDirIndex value from 5 to--
you guessed it--6, opens that file, and runs your INIT again.
The workaround is to avoid creating files in the same folder as the INIT that alphabetically
precede your INIT file. Also, the file you create should be placed in the Preferences folder within
the System Folder, for both System 6 and System 7. Check for a Preferences folder within the
System Folder using FindFolder, and create one if it doesn't exist.
Q During a Macintosh application's life, does the value of A5 change? Why does SetCurrentA5 have to set
A5 to CurrentA5? Aren't A5 and CurrentA5 the same while an application is executing non-interrupt-
time code?
A A5 is not necessarily always equal to CurrentA5. Because the Macintosh operating system and
Toolbox don't need to access your application's jump table or global variables (which A5 points
to), they often use A5 for other purposes, except for the parts of the Toolbox that need to
perform graphics operations and use the QuickDraw globals defined by your application.
Because the operating system or Toolbox can change A5, you must make sure it's set correctly
if the operating system or Toolbox ever calls your code in the form of a callback. For example,
if you make an asynchronous File Manager call, your I/O completion routine must call
SetCurrentA5 for your I/O completion routine. Other places where you may want to call
SetCurrentA5 are in trap patches, your GrowZoneProc, custom MDEFs, WDEFs, or CDEFs,
or control action procedures.
Please note that even SetCurrentA5 is not sufficient when writing code that's executed at
interrupt time, such as VBL and Time Manager tasks. When an interrupt occurs that your
application wants to handle, there is no guarantee that the low-memory global CurrentA5
belongs to your application. The interrupt could occur while some other application is running
under MultiFinder®. In this case, you should use other approaches to setting A5. Please see
Technical Note #180, MultiFinder Miscellanea, for further information.
Q My Macintosh application receives information from the serial port, keeping as much information in
memory as possible. However, if I do run out of memory, my GrowZone routine is allowed to delete older
information. Does the operating system allow the GrowZone routine to be called recursively? That is, if
in handling a GrowZone call, my code does something that requires memory, such as bringing in a font
that isn't there, can I call the GrowZone routine again or would doing so cause an error? Am I allowed
to update things on the screen, such as windows and text in windows, while in my GrowZone routine?
A Neither the GrowZone routine nor the Macintosh system is reentrant. There is very little that
you can safely do in a GrowZone routine. The system expects that the typical extent of a
GrowZone routine is to check a table or some other predesignated criteria and then to release
or not to release any memory blocks it can. Any sort of user interface is well beyond the
reasonable scope of this level of routine.
Keep a list of handles to blocks of text that you're willing to purge, so that your GrowZone
routine can look through them when called and then release one or more blocks. Any more
than that is really stretching GrowZone too far. By the way, SetGrowZone should not be assumed
to return a result. Inside Macintosh is incorrect
in saying that D0 contains a result code on exit.
Q Do I need to call StripAddress when using SwapMMUMode to switch to 32-bit mode? I would like to
assume that all my pointers are valid 32-bit quantities when in 32-bit mode.
A Yes, though it would be nice to be able to assume that all your master pointers are valid 32-bit
quantities when in 32-bit mode, the assumption would only be true of master pointers that were
created in a zone created while you were in 32-bit mode, and even this would only be true on
systems that have a 32-bit Memory Manager (the Macintosh IIci, IIfx, LC, and IIsi).
You need to call StripAddress on any master pointers that were created in a
24-bit world that you expect to use while in 32-bit mode. You also need to call StripAddress on
distinct master pointers that were created in 24-bit mode if you want to compare them,
regardless of whether you're currently in 32-bit mode.
Q When building a standalone code resource (such as an 'XCMD', 'INIT', 'CDEV', or 'CDEF') with the
MPW Linker, the main entry point is specified with the "-m" option. So why does one also need to make
sure that the main routine is the first one in its object file, and that that object file is the first one linked?
Why isn't "-m" enough for the Linker to figure this out?
A The MPW® Linker generally places modules in the output file in the order that they appear in
the sequence of files to be linked. When building an application, the Linker builds the jump
table after building all the code resources; thus, it doesn't need to have "seen" the main
procedure first in order to place the main procedure's entry first in the jump table. For
standalone code resources (which have no jump table), the main procedure must be first in the
code resource. The simplest way to make sure this happens is, as you say, to make sure that the
main procedure is the first one the Linker sees.
Q Why does MPW Pascal convert SINGLE numbers to extended, create a temporary area for them, and
then pass a pointer to those extended numbers on the stack?
A The conversion of single-precision values to extended is being done to maintain accuracy. It's
entirely possible to generate values of extended precision while doing the intermediate math
with single-precision values, and the MPW compilers do this conversion to preserve accuracy.
If your application doesn't need the accuracy, you can declare the parameters to be of type
LONGINT, and typecast them as necessary within your procedure or function. There's no way
to tell the MPW compilers not to do the conversion if the parameters are declared as SINGLE.
Q What is the difference between RelString and EqualString? What should Macintosh developers use when
sorting? Do you suggest having an option for the international sort?
A RelString and EqualString are mainly intended for the File Manager. The File Manager uses
them for quick-and-dirty string comparison so that it knows how to return files ordered
alphabetically when you use indexed File Manager routines, and so that it can detect file name
collisions. Beyond that, RelString and EqualString aren't localizable or extensible.
The International Utilities string comparison routines are localizable and extensible. They use
information in the active 'itl2' resource to determine how the characters are sorted. Most
localized systems come with their own 'itl2' resource, so string comparisons are done correctly
for the region for which the system is localized. Because RelString and EqualString stay the
same for all these regions, you'll probably find some cases in which strings are compared
incorrectly by these routines.
One important place where RelString and EqualString don't work very well is with the new
characters in the extended Macintosh character set. When the LaserWriter was introduced, the
LaserWriter fonts used the extended Macintosh character set, which added many new
characters, including several new uppercase characters with diacriticals. In system software
version 6.0.4, the International Utilities were updated to take advantage of these new
characters. For example, the uppercase "E" with a grave accent first appeared in the extended
Macintosh character set. With 6.0.4, the lowercase and uppercase "E" with a grave accent were
considered to be equal in primary ordering, and unequal in secondary ordering, which is
correct. Even today, RelString and EqualString think in the old Macintosh character set,
erroneously believing that lowercase and uppercase "E" with a grave accent have nothing to do
with each other.
Q The Macintosh IIsi and LC computers don't come with programmer's switches. How do I get reset and
NMI on these machines?
A Both machines have an ADB (Apple Desktop Bus) I/O processor which incorporates the reset
and NMI functions through the keyboard. The functions are accessed as follows:
- Use MacsBug 6.2 or later to enable the NMI on the Macintosh IIsi or LC. Both machines
start up with NMI disabled because nontechnical users might get into trouble with it, so the
ADB controller must be programmed to enable it. Once MacsBug is installed, you can
activate NMI by holding down the Command key while pressing the Power button.
- Hold down the Command and Control keys while pressing the Power button, with or
without MacsBug installed, to reset these Macintosh systems.
Under some circumstances you may have to hold the reset and NMI key combinations down
for a little while (but no more than a second) to make sure the ADB processor sees them.
You'll find the latest MacsBug on the Developer Essentials disc and in AppleLink's Developer
Support folder. The MacsBug reference manual is very helpful. A package containing both
MacsBug 6.2 and the manual is available from APDA for $35.00 (#M7034/B).
Q What's wrong with having VBL (Vertical Blanking) tasks make calls to the Macintosh Memory
Manager, either directly or indirectly?
A The problem is that the Memory Manager could be moving memory around when an interrupt
occurs. If the VBL task also moves memory, the heap could be destroyed. VBL tasks are intended to
provide a method of time-syncing to the video beam of the display.
(On slotted Macintosh models you'd use SlotVInstall.) They're also used to get periodic time
for short tasks, although the Time Manager is better for this. VBL tasks should minimize
execution time. The best use of a VBL task is to do a short condition check and set a flag for
the main process to indicate that it's now a good time to do something.
Q Can I still write to James Brown at his work release address?
A The Godfather of Soul was released from prison on parole last March. His new address is
Universal Attractions
218 West 57 Street
New York, NY 10019
Kudos to our readers who care enough to ask us terrific and well thought-out questions. The answers are supplied by our
teams of technical gurus; our thanks to all. Special thanks to Pete "Luke" Alexander, Mark Baumwell, Rich Collyer, Marcie
"MG" Griffin, C. K. Haun, Dennis Hescox, Kevin Mellander, Guillermo Ortiz, Keith Rollin, Kent Sandvik, Gordon
Sheridan, Paul Snively, Bryan Stearns, Forrest Tanaka, Vince Tapia, and Scott "Zz" Zimmerman for the material in this Q
& A column.*
Have more questions? Need more answers? Take a look at the developer technical library on AppleLink (updated weekly)
or the Q & A stack on the Developer Essentials disc.*
