August 92 - LIVING IN AN EXCEPTIONAL WORLD
LIVING IN AN EXCEPTIONAL WORLD
SEAN PARENT
Handling exceptions is a difficult but important part of developing Macintosh
applications. This article provides a methodology as well as a set of C tools for
handling exceptions and writing robust code. Techniques and examples are provided
for dealing with some of the Toolbox idiosyncrasies, and some interesting features of
the C preprocessor, MacsBug, and MPW are explored.
Writing software on the Macintosh can be difficult. Writing robust software on the Macintosh is
even more difficult. Every call to the Toolbox is a potential source of a bug and there are too many
cases to handle--what if there isn't enough memory, or the disk containing the code has been
ejected, or there isn't enough stack space, or the printer is unplugged, or . . . The list goes on, and a
well-written application is expected to handle every case--always recovering without loss of
information. By looking at how software is developed, this article introduces a methodology and tools
for handling the exceptional cases with minimal impact on the code that handles the task at hand.
VERSION 1: NORMAL FLOW OF CONTROL
When writing code, programmers usually begin by writing the normal flow of control--no error
handling. The code shown below is a reconstruction of the first version of a printing loop routine
that eventually went out as a Macintosh Technical Note, "A Printing Loop That Cares . . ." (#161).
Note that comments were removed to make the structure more apparent.
#include <Printing.h>
#include <Resources.h>
#include <Memory.h>
void PrintStuff(void)
{
GrafPtr oldPort;
short copies, firstPage, lastPage, numCopies,
printmgrsResFile, realNumberOfPagesInDoc,
pageNumber;
DialogPtr printingStatusDialog;
THPrint thePrRecHdl;
TPPrPort thePrPort;
TPrStatus theStatus;
GetPort(&oldPort);
UnLoadTheWorld();
thePrRecHdl = (THPrint)NewHandle(sizeof(TPrint));
PrOpen();
printmgrsResFile = CurResFile();
PrintDefault(thePrRecHdl);
if (PrStlDialog(thePrRecHdl)) {
realNumberOfPagesInDoc = DetermineNumberOfPagesInDoc(
(**thePrRecHdl).prInfo.rPage);
if (PrJobDialog(thePrRecHdl)) {
numCopies = (**thePrRecHdl).prJob.iCopies;
firstPage = (**thePrRecHdl).prJob.iFstPage;
lastPage = (**thePrRecHdl).prJob.iLstPage;
(**thePrRecHdl).prJob.iFstPage = 1;
(**thePrRecHdl).prJob.iLstPage = 9999;
if (realNumberOfPagesInDoc < lastPage) {
lastPage = realNumberOfPagesInDoc;
}
printingStatusDialog =
GetNewDialog(257, nil, (WindowPtr) -1);
for (copies = 1; copies <= numCopies; copies++) {
(**thePrRecHdl).prJob.pIdleProc =
CheckMyPrintDialogButton;
UseResFile(printmgrsResFile);
thePrPort = PrOpenDoc(thePrRecHdl, nil, nil);
pageNumber = firstPage;
while (pageNumber <= lastPage) {
PrOpenPage(thePrPort, nil);
DrawStuff((**thePrRecHdl).prInfo.rPage,
(GrafPtr)thePrPort, pageNumber);
PrClosePage(thePrPort);
++pageNumber;
}
PrCloseDoc(thePrPort);
}
if ((**thePrRecHdl).prJob.bJDocLoop == bSpoolLoop) {
PrPicFile(thePrRecHdl, nil, nil, nil, &theStatus);
}
}
}
PrClose();
DisposeHandle((Handle)thePrRecHdl);
DisposeDialog(printingStatusDialog);
SetPort(oldPort);
} /* PrintStuff */
VERSION 2: ERROR HANDLING ADDED
With code in the preliminary stage shown above, the flow of control is easy to follow. After writing
it, the programmer probably read through it and added some error-handling code. Adding "if (error
== noErr)" logic wasn't difficult, but it took some thought to determine how to handle the cleanup
and deal with the two loops. Some more error-handling code may have been added after running the
routine under stressful conditions. Perhaps it was reviewed by lots of people before it went out as a
Technical Note. Here's the new version of the code (with the added error-handling code shown in
bold):
#include <Printing.h>
#include <Resources.h>
#include <Memory.h>
void PrintStuff(void)
{
GrafPtr oldPort;
short copies, firstPage, lastPage, numCopies,
printmgrsResFile, realNumberOfPagesInDoc,
pageNumber, printError;
DialogPtr printingStatusDialog;
THPrint thePrRecHdl;
TPPrPort thePrPort;
TPrStatus theStatus;
GetPort(&oldPort);
UnLoadTheWorld();
thePrRecHdl = (THPrint)NewHandle(sizeof(TPrint));
if (MemError() == noErr && thePrRecHdl != nil) {
PrOpen();
if (PrError() == noErr) {
printmgrsResFile = CurResFile();
PrintDefault(thePrRecHdl);
if (PrError() == noErr) {
if (PrStlDialog(thePrRecHdl)) {
realNumberOfPagesInDoc =
DetermineNumberOfPagesInDoc(
(**thePrRecHdl).prInfo.rPage);
if (PrJobDialog(thePrRecHdl)) {
numCopies = (**thePrRecHdl).prJob.iCopies;
firstPage = (**thePrRecHdl).prJob.iFstPage;
lastPage = (**thePrRecHdl).prJob.iLstPage;
(**thePrRecHdl).prJob.iFstPage = 1;
(**thePrRecHdl).prJob.iLstPage = 9999;
if (realNumberOfPagesInDoc < lastPage) {
lastPage = realNumberOfPagesInDoc;
}
printingStatusDialog =
GetNewDialog(257, nil, (WindowPtr) -1);
for (copies = 1; copies <= numCopies;
copies++) {
(**thePrRecHdl).prJob.pIdleProc =
CheckMyPrintDialogButton;
UseResFile(printmgrsResFile);
thePrPort =
PrOpenDoc(thePrRecHdl, nil, nil);
if (PrError() == noErr) {
pageNumber = firstPage;
while (pageNumber <= lastPage &&
PrError() == noErr) {
PrOpenPage(thePrPort, nil);
if (PrError() == noErr) {
DrawStuff(
(**thePrRecHdl).prInfo.
rPage,
(GrafPtr)thePrPort,
pageNumber);
}
PrClosePage(thePrPort);
++pageNumber;
}
}
PrCloseDoc(thePrPort);
}
} else PrSetError(iPrAbort);
} else PrSetError(iPrAbort);
}
}
if (((**thePrRecHdl).prJob.bJDocLoop == bSpoolLoop) &&
(PrError() == noErr)) {
PrPicFile(thePrRecHdl, nil, nil, nil, &theStatus);
}
printError = PrError();
PrClose();
if (printError != noErr)
PostPrintingErrors(printError);
}
if (thePrRecHdl != nil) DisposeHandle((Handle)thePrRecHdl);
if (printingStatusDialog != nil)
DisposeDialog(printingStatusDialog);
SetPort(oldPort);
} /* PrintStuff */
Can you easily follow the normal flow of control in the second version? What if an error occurs?
Could an error ever go unreported? Could this code crash because it didn't handle an error? Does
this routine always clean up after itself? These questions are difficult to answer because the normal
flow of control of this routine is intertwined with the flow that will occur in the event of an error. If
the two could be separated, it would be much easier to tell what the routine does normally and how
things are handled when something goes wrong. Besides making the code easier to read, a
methodology that allowed such separation would make the code easier to write, debug, and, maintain.
PROGRAMMING BY CONTRACT
Programming by contract is based on the assumption that all correct routines have a contract, either
stated or implied, with their caller. The contract states that if a given set of preconditions is met, the
routine either succeeds or flags an exception and leaves the machine in a known or determinable
state. This contract is flexible enough to be applied to any correct code.
The secret to writing robust code is to understand what the preconditions of a given routine are,
when an exception can be flagged, and how to handle the exception. Separating the logic that checks
conditions and handles exceptions from the algorithm of the routine allows code to be written in a
straightforward way with the flow of control seen as easily as in our first version.
PRECONDITIONS
The preconditions of a routine specify the state the machine must be in for the routine to execute
without failure (where failure implies a crash--not flagging an exception). A routine may require in
its precondition items such as
- a previously called initialization routine
- valid ranges for value parameters
- available memory
- initialization of global state
- a specific software version
For some routines the preconditions may be readily apparent either in the interface or in the
documentation. Sometimes it's necessary to experiment to discover the preconditions of a routine.
When writing a routine, the "strength" of the precondition can be set according to the use of the
routine. For example, a routine named DivideLong is written with a description that states:
Given two numbers, numer and denom, DivideLong will divide numer by denom, set numer to
the result, and return noErr. If denom is zero, DivideLong will return divideByZeroErr and
leave numer unchanged.
With this description, numer and denom can be any numbers of the proper type. This is a weak
precondition. Another description might read:
Given two numbers, numer and denom, DivideLong will divide numer by denom and return the
result. If denom is zero, DivideLong will fail.
With the second description, it would be the caller's responsibility to ensure that denom isn't zero.
This is a strong precondition.
In general, it's better to have a strong precondition in a routine that is used within a sequence of
related routines or shares conditions with other routines, because it generates more efficient code byeliminating error checking. It's better to have a weak precondition in routines that are called only
once or at the start of a sequence of related routines. Routines with weak preconditions free the caller
from ensuring the state of the machine before making the call.
A precondition can be strengthened by the caller but must not be weakened. Strengthening is useful
when you're making a sequence of related calls where being sure additional conditions are met
guarantees that no routine flags an exception. For example, given the first description of DivideLong
it would be valid for a caller to do the following:
if (denom != 0) {
(void)DivideLong(&numer_a, denom); /* Ignore return. */
(void)DivideLong(&numer_b, denom); /* Ignore return. */
} else HandleError();
This may be more desirable than checking for a result of divideByZeroErr after each call. An
example of weakening a precondition would be to call DivideLong as described in the second
description without ensuring that denom isn't zero. This would constitute a bug.
POST-CONDITIONS
Post-conditions specify the state of the machine on the return of a routine. They include side effects
and changes to global state as well as function results and variable parameters. The post-conditions of
a routine must be determinable for the routine to be correct. They don't vary in strength and, if not
met, the routine has a bug. A thorough understanding of the post-conditions of a routine is required
to ensure that the routine is being called correctly and that cleanup can occur when the routine flags
an exception.
Sometimes it's necessary to rephrase the preconditions and post-conditions of a routine to use it
correctly. For example, a common misconception is that the only preconditions for calling TEKey
are that it has passed a valid TEHandle and the appropriate Managers have been initialized. Since
there's no mechanism for TEKey to flag an exception, the assumption is that it can't fail. But TEKey
may need to grow the hText handle if the character isn't replacing others and isn't a backspace.
Growing a handle requires memory--something there may not be enough of. Since TEKey can fail
without flagging an exception with these preconditions, it appears to be incorrect and contain a bug.
However, by strengthening the preconditions to require that hText must be able to grow by the size
of a character, the routine is once again correct. Strengthening preconditions is an easy fix often used
in system software. (See the section "Preflighting Calls" for tips on how to ensure preconditions.)
HOW TO WRITE CHECKS
The checkmacro is used to ensure that static preconditions and post-conditions are being met
during development. It also documents conditions for you, making it a very useful tool that adds to
the maintainability of the code. Unfortunately, these conditions cannot be expressed directly in the
interface so as to be more apparent to the caller. The syntax forcheckis
check(assertion );
To use thecheck macro, include Exceptions.h (provided on theDeveloper CD Series disc). For
MacsBug, use ResEdit to add Exceptions.rsrc to the DebuggerPrefs file in the System Folder.
What checkdoes depends on the setting of the compile-time variable DEBUGLEVEL.
DEBUGLEVEL can be set to one of the following values:
- DEBUGOFF or DEBUGWARN:checkdoes nothing andassertion is not
evaluated.
- DEBUGMIN or DEBUGSYM:assertion is evaluated and, if it's false (zero), a
debugger break is executed. (The debugger break is Debugger ( ) for DEBUGMIN
and SysBreak ( ) for DEBUGSYM. The first is useful for low-level debuggers like
MacsBug or TMON, the second for symbolic debuggers like SourceBug, SADE,
or THINK C.)
- DEBUGON or DEBUGFULL:assertion is evaluated and, if it's false (zero),
MacsBug is entered and the dprintf dcmd is invoked to display more information.
If DEBUGON,assertion is displayed and if DEBUGFULL, the source code file
and line number are also displayed (see "Wonders of MacsBug and dprintf" for
more information about dprintf).
Normally,checkis used at the start and end of a routine. At the start it's used to ensure that
parameters are within a given range and are not specific values (such as nil). At the end it's used to
ensure that allocations succeeded and results are as desired.
REQUIREMENTS FOR BETTER LIVING
Although
checkcan ensure that preconditions and post-conditions are being met during
development,
checkis of limited value in situations where it cannot be determined whether the
conditions are being met statically, because
- it disappears when DEBUGLEVEL is set to DEBUGOFF
- it doesn't provide sufficient support for handling exceptions to return the machine
to a known state
What's needed is a mechanism that does not compile out and provides the ability to invoke a handler
whenassertion fails.
WHAT WE REQUIRE
The requiremacro was created to make handling exceptions simpler. The syntax forrequireis
require(assertion , exception );
Ifassertion evaluates to false (zero), execution continues at the handlerexception . (Theexception parameter, by convention, shares the name of the routine that failed, but this isn't mandatory.)
Handlers are typically written as shells with control falling from one to the next, cleaning up after
prior calls along the way. The extent of the cleanup needed gets deeper as more of the routine
succeeds. Figure 1 shows an extended form ofrequirecalled require_action. The extended form
executes a statement whenassertion fails before executing the handler. This is most useful for setting
an error variable. The syntax forrequire_actionis
require_action(assertion , exception , action );
Like check, requirebreaks into MacsBug and displays pertinent information depending on the
settings of DEBUGLEVEL. Unlikecheck, requiredoes not compile out when DEBUGLEVEL is
set to DEBUGWARN or DEBUGOFF. It evaluatesassertion and invokes the handler (andaction ),
but no break occurs.
The nrequiremacro is equivalent torequire(! assertion , exception ). However, under rare
circumstances it generates more efficient code, and when debugging is on, it displays the value ofassertion . It's also easier to read. As a general rule, userequirewith handles and pointers andnrequirewith errors.
VERSION 3: IMPROVED WITH REQUIRE
A close look at the code in version 2 reveals some problems:
- No error handling is done after PrCloseDoc, though any errors will get caught
either after the next PrOpenDoc or on exit.
- If the NewHandle at the start of the code fails, it won't print and the user is never
notified why.
- If an error occurs in the copies loop, the loop isn't terminated.
If printing is well behaved and does nothing once PrError has been set, none of these problems poses
much of a threat to the actual stability of the code (with the exception of GetNewDialog). However,
the use ofrequirewhen writing the code could have avoided the problems and the code would be
easier to understand and maintain. This is shown in the code below--version 3. The structure of the
code in version 3 is almost identical to version 1 with the addition of therequirestatements and the
handlers at the end. Writing code like this is straightforward. When a routine is called that can flag
an exception, arequirestatement is added with a handler. The statements executed in a handler
typically clean up after the routines called before the routine flagging the exception. (See the section
"When To Clean Up" for more discussion.) Although PrClose should never cause an error, acheck statement was added during development.
Here's version 3 of the code (with changes from version 1 shown in bold):
#include <Printing.h>
#include <Resources.h>
#include <Memory.h>
#include <Errors.h>
#include "Exceptions.h"
void PrintStuff(void)
{
GrafPtr oldPort;
short copies, firstPage, lastPage, numCopies,
printmgrsResFile, realNumberOfPagesInDoc,
pageNumber;
DialogPtr printingStatusDialog;
OSErr theError;
THPrint thePrRecHdl;
TPPrPort thePrPort;
TPrStatus theStatus;
long contig, total;
enum { dialogSlop = 8192 };
GetPort(&oldPort);
UnLoadTheWorld();
thePrRecHdl = (THPrint)NewHandle(sizeof(TPrint));
require_action(thePrRecHdl, NewHandle,
theError = MemError(););
PrOpen();
nrequire(theError = PrError(), PrOpen);
printmgrsResFile = CurResFile();
PrintDefault(thePrRecHdl);
nrequire(theError = PrError(), PrintDefault);
if (PrStlDialog(thePrRecHdl)) {
realNumberOfPagesInDoc = DetermineNumberOfPagesInDoc(
(**thePrRecHdl).prInfo.rPage);
if (PrJobDialog(thePrRecHdl)) {
numCopies = (**thePrRecHdl).prJob.iCopies;
firstPage = (**thePrRecHdl).prJob.iFstPage;
lastPage = (**thePrRecHdl).prJob.iLstPage;
(**thePrRecHdl).prJob.iFstPage = 1;
(**thePrRecHdl).prJob.iLstPage = 9999;
if (realNumberOfPagesInDoc < lastPage) {
lastPage = realNumberOfPagesInDoc;
}
PurgeSpace(&total, &contig);
require_action(contig >= dialogSlop, PurgeSpace,
theError = memFullErr;);
printingStatusDialog =
GetNewDialog(257, nil, (WindowPtr) -1);
require_action(printingStatusDialog, GetNewDialog,
theError = memFullErr;);
for (copies = 1; copies <= numCopies; copies++) {
(**thePrRecHdl).prJob.pIdleProc =
CheckMyPrintDialogButton;
UseResFile(printmgrsResFile);
thePrPort = PrOpenDoc(thePrRecHdl, nil, nil);
nrequire(theError = PrError(), PrOpenDoc);
pageNumber = firstPage;
while (pageNumber <= lastPage) {
PrOpenPage(thePrPort, nil);
nrequire(theError = PrError(), PrOpenPage);
DrawStuff((**thePrRecHdl).prInfo.rPage,
(GrafPtr)thePrPort, pageNumber);
PrClosePage(thePrPort);
nrequire(theError = PrError(),
PrClosePage);
++pageNumber;
}
PrCloseDoc(thePrPort);
nrequire(theError = PrError(), PrCloseDoc);
}
if ((**thePrRecHdl).prJob.bJDocLoop == bSpoolLoop) {
PrPicFile(thePrRecHdl, nil, nil, nil, &theStatus);
nrequire(theError = PrError(), PrPicFile);
}
}
}
PrClose();
ncheck(PrError());
DisposeHandle((Handle)thePrRecHdl);
DisposeDialog(printingStatusDialog);
SetPort(oldPort);
return;
PrOpenPage:
PrClosePage(thePrPort);
PrClosePage:
PrOpenDoc:
PrCloseDoc(thePrPort);
PrPicFile:
PrCloseDoc:
DisposeDialog(printingStatusDialog);
GetNewDialog:
PurgeSpace:
PrintDefault:
PrOpen:
PrClose();
DisposeHandle((Handle)thePrRecHdl);
NewHandle:
SetPort(oldPort);
PostPrintingErrors(theError);
} /* PrintStuff */
PREFLIGHTING CALLS
Preflighting a call is the process of ensuring that the preconditions are met. Usually this isn't
necessary since the preconditions will be satisfied by handling the exceptions of prior calls or will be
implicit in the caller's preconditions. For example, there's no need to ensure that the TEHandle
being passed to TEKey isn't nil if the exceptional case of the previous TENew returning nil was
handled. In case preconditions haven't been satisfied by handling the exceptions of previous calls,
requirecan
be used to check the precondition and invoke a handler if it's not being met. This is especially useful
for routines that have strong preconditions or preconditions that are difficult to determine. Earlier,
TEKey was used as an example of a routine with strong preconditions. To ensure the preconditions
for TEKey,
requirecould be used as follows:
OSErr SafeTEKey(short key, TEHandle hTE) {
enum { teSlop = 1024 };
OSErr error;
TEPtr w = *hTE;
Handle hText = w->hText;
short teLength = w->teLength;
SetHandleSize(hText, teLength + teSlop);
nrequire(error = MemError(), SetHandleSize);
SetHandleSize(hText, teLength);
TEKey(key, hTE);
return(noErr);
SetHandleSize:
return error;
}
The constant teSlop is used instead of 1 just to be safe. Adding some slop for routines with implied,
rather than stated, preconditions is always a good idea.
For some routines the preconditions are too complex or subject to change to accurately state as an
assertion. This is the case for GetNewDialog, as shown in version 3. GetNewDialog can fail when
there isn't enough memory for one of the numerous QuickDraw elements to be allocated, to load the
WDEF, or, if the dialog contains TextEdit items, to create the TEHandle. About all that can be
done to guarantee that GetNewDialog succeeds is to ensure that there's a reasonable amount of
memory available. It's fairly safe to rely on the Process Manager in System 7 to make sure there's
space in the system heap for the WDEF. This is what's done in version 3. The assertion is based on
contiguous memory instead of total memory in case the heap is too fragmented to allocate some of
the larger blocks required. Sometimes all that can be done is to increase the chances of survival.
WHEN TO CLEAN UP
Just as preconditions can sometimes be tricky to determine, post-conditions can be hazardous as well.
It's important to understand the post-conditions of the routines being called, so that the machine can
be returned to a known state, ensuring valid post-conditions for the calling routine. Normally, if an
exception is being raised, a routine should dispose of everything it
successfully allocated, close
everything it
successfully opened, and release everything it
successfully locked. So, if NewHandle is
called successfully, DisposeHandle is called in the handler. If OpenFile is called successfully,
CloseFile is called in the handler.
But this rule isn't always true. One counterexample is the Printing Manager. Even if PrOpen flags an
exception PrClosemust be called. The same is true for PrOpenDocument and PrOpenPage.
Shared resources present another potential problem. If GetResource is successfully called on a system
resource, it's a bad idea to release it, because it may also be in use by another routine.
SetResLoad(false) and GetResource can be called to determine whether the resource is already in
memory before loading it, and then it can be released only if it was loaded. This, however, is taking
things to an extreme. It may be better to document that these resources may be loaded even if the
routine flags an exception. Since this is determinable by the caller, it suffices as a valid post-
condition.
FUTURE DIRECTIONS
The routines and macros provided in Exceptions.h lay the foundation for writing robust software.
There are more sophisticated exception-handling mechanisms, such as the proposed
catch and
throw implementation for C++. Ada has a reasonable exception-handling mechanism, as does CLU
and Eiffel. However, these mechanisms don't lend themselves to dealing with exceptions from
routines that were not written using the same mechanism and so are difficult to use on the Macintosh
when dealing with the OS and Toolbox. The
checkand
requiremacros are flexible enough to be
useful in most situations and are implemented in C, so they can be of value for many (if not most)
existing projects. They are also C++ friendly and can be of great use to C++ programmers as well.
After you read the code in version 3 that uses these macros it should be fairly simple to answer the
questions asked about version 2 at the beginning of the article. This is left as an exercise.
Turn the page if you want even more detail . .
MORE DETAIL THAN MOST FOLKS NEED
The
require and
check macro implementation is shown in Figure 2. To ensure that there aren't any side
effects, any macro that's larger than a single statement is enclosed in
do { } while(false). This ensures that the
macro behaves as a simple statement and can be used anyplace a simple statement would be (such as after
an
if). The
do { } while(false) does not generate any object code. In some of the macros,
if statements
appear in the form
if (assertion) ; /* Do nothing. */
else { /* Do something. */ }
Under some conditions, this will generate more efficient code than
if (!assertion) { /* Do something. */ }
(This was especially true back in the days of the MPW 3.1 compiler.) There are variables declared within the
scope of the macros when debugging is on. This avoids side effects caused by evaluating assertion multiple
times (once in the condition and once to display it). For example:
nrequire(ReadCharacters(), Fail);
If ReadCharacters returned a value other than nil, MacsBug would be invoked to display the result before
executing the handler Fail. Without the local variable, ReadCharacters would be executed a second time to
display the value. The second execution may cause side effects like increasing a file pointer as well as
reading in a different set of characters.
When assertion is an error code returned by a function, it can be assigned to a variable to preserve the
error. This also keeps the exception-handling code enclosed within the requirestatement. For example:
nrequire(error = GetError(), Fail);
However, with warnings set to full, this invokes a warning because the assignment takes place as part of anif statement. Using
error = GetError();
nrequire(error, Fail);
generates identical code (at least with MPW 3.2) and doesn't cause any warnings.
A macro, resume, is provided for recovering from exceptions. It's used within a handler and takes the form
resume(exception );
where exception corresponds to exception used in a require statement. The resume macro simply transfers
control to the point immediately following the require statement. Because of the resume feature, multiplerequire statements cannot share the same exception handler. Sometimes sharing a handler is convenient, soresume can be disabled with a statement:
#define resumeLabel(exception)
To reenable resume, use
#define resumeLabel(exception)\
resume_ ## exception:
There's also a check_actionmacro which, like require_action, allows a statement to be executed whenassertion fails. The check_action macro compiles out like all check macros and should be viewed as a
development-time tool only. Being able to execute a statement allows for the exit of a routine if the
preconditions aren't met.
#define require(assertion, exception) \
do { \
if (assertion) ; \
else { \
dprintf(notrace, \
"Assertion \"%s\" Failed\n" \
"Exception \"%s\" Raised", \
#assertion, #exception); \
goto exception; \
resumeLabel(exception); \
} \
} while (false)
#define check(assertion) \
do { \
if (assertion) ; \
else { \
dprintf(notrace, \
"Assertion \"%s\" Failed", \
#assertion); \
} \
} while (false)
Figure 2Implementing require and check
WONDERS OF MACSBUG AND DPRINTF
The MacsBug dcmd, dprintf, is used by the
requireand
check macros to display useful debugging information. The dprintf command is also a powerful tool that
provides all the features of the standard printf but uses MacsBug as the console. The dprintf command
assumes MPW parameter-passing conventions. The syntax for dprintf is
dprintf([no]trace, formatString , ...);
where trace and notrace are used to specify whether
or not to continue after displaying the information in MacsBug. The variable formatString is a printf style-
format string with some extensions (see the comment in the Exceptions.h file on the Developer CD Series disc).
Following formatString are the parameters to display. This can be a very useful tool for viewing complex
structures or difficult-to-read values like floating- or fixed-point numbers.The implementation of the dprintf dcmd is shown in the DPrintf.c file on the CD. It's fairly straightforward and
can be extended easily to add any special data types required (for example, a t format character that would
take a pointer to text and an integer length and display the text). The dcmd is invoked from C using the
inline declaration for dprintf. The inline declaration invokes the DebugStr trap and pushes a long on the
stack. The push
is required because DebugStr uses Pascal calling conventions and so pops the [ no]trace string from the stack.
Since dprintf is a C-based function, the stack is fixed, so the string isn't popped twice. Both traceand notrace are macro Pascal strings containing ";dprintf;doTrace" and ";dprintf". Since the strings begin
with a semicolon, MacsBug interprets them as commands and executes them. The dcmd then fetches the
parameters from the stack according to formatString and displays them. The MacsBug macro doTraceevaluates to "g" or "". It's used to switch tracing between trace and break by entering either traceGo ortraceBreak in MacsBug.
When developing software, it's useful to insert dprintf statements to display information in sections of code
that are executed only in unusual circumstances. If dprintf is bracketed with #if debugon / #endif directives,
it compiles out when DEBUGLEVEL is set to DEBUGWARN or DEBUGOFF. With trace the information is
displayed without seriously interrupting the execution of the code. The trace macro is also useful for logging
timing
statistics by displaying Ticks. Since formatString is interpreted in MacsBug with interrupts disabled, even a
complex formatString has minimal impact on timing results.
MACSBUG POWER USER TIP
If you have more than one monitor, you can use the swap command to make MacsBug always visible and
use dprintf with trace to continually log information. You can set which screen MacsBug uses by opening the
Monitors control panel, holding down the Option key, and dragging the "Happy Macintosh" to the monitor
on which you want to display MacsBug (you have to restart for it to take effect).
MPW POWER USER TIP
At the end of the comment for dprintf in Exceptions.h is a section that uses Echo to pipe code to the
assembler.
/*********************************************
Echo " ðnð
PRINT OFF,NOHDR ðnð
INCLUDE 'Traps.a' ðnð
PRINT ON ðnð
PROC ðnð
_DebugStr ðnð
SUBQ #4,SP ; Fix the stack ðnð
ENDPROC ðnð
END ðnð
" | Asm -l
*********************************************/
If you select this section and press Enter, it generates a listing with hex output. This is a handy way to
generate and document inline functions..
RELATED READING
- Macintosh Technical Note "A Printing Loop That Cares . . ." (formerly #161).
- Object-Oriented Software Construction by Bertrand Meyer (Prentice-Hall, 1988). Contains more
information on programming by contract.
- Debugging Macintosh Software with MacsBug by Konstantin Othmer and Jim Straus (Addison-Wesley,
1991). Contains additional MacsBug tips.
SEAN PARENT (AppleLink PARENT, Internet parent@apple.com) is a parent, but Parent is his last name, not his title. He
grew up in Renton, Washington, with his parents (you know, the people who produced him), who are also Parents. Sean
came to Apple to pursue his lifelong interest in reference manuals. He enjoys a good ANSI standards document duringbreakfast, and likes catchy punch lines such as, "No, no! I said 'ANSI,' not 'ASCII'!" Sean also likes to write a good
hack, and consistently comes in next-to-second-best at the annual MacHack MacHax Hack Contest. Unable to hide his
prowess, he gave in to the inevitable job at Apple, and now he wants to change the world, one programming paradigm
at a time.*
CouldDialog, which was intended as a preflight tool for GetNewDialog, is a no-op in System 7 (it's been broken since the
Macintosh II was introduced).*
Seriously insane cycle counters take note that the MPW 3.2 C compiler doesn't reuse a register to store a variable in a
local scope when the register was used in a prior scope containing a goto statement (require generates a goto). This can
lead to code that isn't as efficient as it should be but can usually be coded around (it's difficult to generate in the first
place). Hopefully this will be fixed in a future compiler. *
THANKS TO OUR TECHNICAL REVIEWERS Scott Boyd, Konstantin Othmer, Sam Weiss. Also, special thanks to everyone in the Print Shop (present and former
members) for using this stuff and suggesting numerous improvements during the past few years. *