Volume Number: 16 (2000)
Issue Number: 10
Column Tag: Carbon Development
by Daniel Jalkut
A feature frequently overlooked by new Mac OS users is the ability to select items from a list by typing part of its name. Apple calls this functionality "type selection." Seasoned Mac users are accustomed to type selection and navigate rapidly through Finder hierarchies and StandardFile dialogs without ever letting their hands leave the keyboard. When a user discovers this feature, it is probably with some disappointment that they find support for type selection varies widely amongst third-party applications. Every application that uses StandardFile and Navigation Services dialogs gets type selection for free in those portions of their program. For custom application dialogs and views, developers have been required to write their own, custom type selection code. The result is that some applications don't support type selection at all, some support it to varying extents, and almost none support it in a way that mimics every last nuance of Apple's own type selection algorithm. The opportunity to end these inconsistencies is before us, for Apple's Type Selection APIs have made their public debut.
Without fanfare, Apple included their time-tested internal Type Selection APIs in the first public releases of the CarbonLib extension for Mac OS. As of this writing, the latest version is CarbonLib 1.0.4, and it is available from Apple's web site at http://asu.info.apple.com/swupdates.nsf/artnum/n11673. CarbonLib is the library that enables applications ported to Mac OS X's Carbon APIs to continue running on Mac OS 9 and earlier (as early as 8.1) systems. There is no shortage of compelling reasons to port your application to Carbon. Apple's decision to support two platforms with a nearly identical API set provides an irresistibly easy way of adding Mac OS X to the list of your supported platforms, without losing your existing customer base. Access to standardized type selection is by no means the greatest benefit of porting to Carbon, but it has the potential of causing a vast improvement in the user's experience on the Mac.
You might be wondering why something as seemingly straightforward as selecting list items with the keyboard should require a standardized API. The user hits a key, you select a matching item, and we're done - right? Actually, the apparent simplicity of type selection is a testament to its internal complexity. You may not have even considered much of the functionality Apple's TypeSelect APIs provide unless you have spent a great deal of time focusing only on the dynamics of type selection. Most developers do not have the time to become type selection experts, so they implement something that makes sense to them, yet lacks the subtle elegance of Apple's code. Some of the features of Apple's API are:
International Support
Support for international scripts, which Apple has always promoted, is becoming more and more important as the customer base for Apple outside of the United States continues to grow. As a developer, anything that provides international support for free is a big win for your product, because it is one fewer thing you need to worry about when pushing to make a localized release available. Apple's type selection APIs use script-aware string comparison functions, which in turn guarantee that every selection made by the user causes a match as expected for the appropriate script.
Multi-Character Matching
A major shortcoming of some third-party implementations of type selection is that only the last key pressed is ever used as a criterion for matching against items. If you type characters in a Finder view, you will notice that the active selection changes as the characters you type grow into a more specific match with the item you're seeking. For instance, if there are three icons in a view, "Belize", "Biafra", and "Burundi", typing just "B" will select the first, "Bi" the second, and "Bu" the third. In an implementation that didn't support multi-character matching, the user would be forced to type "B" to locate the first item, and then navigate with arrow keys to the desired item. No amount of typing would select either "Biafra," or "Burundi" without moving a hand from the keyboard to the mouse.
Time-Out and Cancellation
One unexpectedly complicated aspect of a proper type selection implementation is elegantly guessing the train of thought a user has embarked on: knowing when to stop one selection and begin another. This task is impossible to do correctly every time, but Apple comes close with comfortable and consistent behavior that gives the user a great deal of control.
The type selection APIs keep a running tab of the keys that have been pressed, and when no keys have been pressed for a duration of time, that buffer is flushed, with the assumption that the user has finished typing the fragment they were seeking. To accommodate the user who knows they have mistyped, or who has quickly changed their mind, the APIs also respond to the escape key, which forces type selection to clear its buffers and begin matching keys as a new string.
To demonstrate the use of the type selection APIs, I have included with this article a sample application, "ListSample", which simply displays a list of selectable items in a dialog. The applications responds to keyDown events by passing them to the Type Selection API, "TypeSelectNewKey", which determines whether the key pressed justifies searching for a new match. If it does, another API, "TypeSelectFindItem" is called, which uses the state of the key buffer and a list of items provided by the client to determine the most appropriate selection.
Application Prerequisites
Before I could implement type selection in ListSample, I first had to ensure that the application had been carbonized and compiled with Apple's latest Universal Headers. If you have already carbonized your application, then you're ready to go. If you haven't, then you will need to do so before you can call any of the Type Selection APIs described in this article.
ListSample Implementation
The easiest way to demonstrate the use of the Type Selection APIs is to explain the three pieces of code in ListSample which interact with the APIs:
First, initialize a TypeSelectRecord. This is done by calling TypeSelectClear() with an empty TypeSelectRecord as its parameter. In ListSample, this is done in main()
Listing 1: main
Main
Setup the menu bar, initialize TypeSelection, create our sample dialog and run the event loop until we are quit.
main()
{
// Setup the menu bar
SetMenuBar(GetNewMBar(rAppMenuBarID));
DrawMenuBar();
// Initialize the type selection record
TypeSelectClear(&gTypeSelectState);
// Prepare the sample dialog
gTheDialog = SetupSampleDialog();
if (gTheDialog != NULL)
{
// Handle Events!
while (gQuitApplication == false)
{
ApplicationEventLoop();
}
DisposeDialog(gTheDialog);
}
ExitToShell();
return 0;
}
Next, for each keyDown event that pertains to the list, ask if a new item needs to be matched. This is done by calling TypeSelectNewKey() - if it returns true, then it is time to find a new match for the list by calling TypeSelectFindItem. I have implemented this functionality in the ListSample routine that handles all keyDown events:
Listing 2: HandleKeyDownEvent
HandleKeyDownEvent
Handle key down events not handled by the Dialog Manager. If the key is not a menu-shortcut, then ask the TypeSelection APIs whether the key might change the state of the current selection, and if so, ask it to determine the new selection
void HandleKeyDownEvent(EventRecord* theEvent)
{
Boolean eventHandled = false;
static IndexToStringUPP myStringGetter = NULL;
// Handle the cmd-key menu case first
if(theEvent->modifiers & cmdKey)
{
long menuKeyResult;
menuKeyResult = MenuKey(theEvent->message &charCodeMask);
DoMenuSelection(HiWord(menuKeyResult),
LoWord(menuKeyResult));
}
else if (gTheDialogList != NULL)
{
// Only allocate the IndexToStringUPP once
if (myStringGetter == NULL)
{
// This call-back function is used by Type Selection to
// fetch elements of the list the user is navigating.
myStringGetter =
NewIndexToStringUPP(GetStringFromIndex);
}
// Ask Type Select APIs whether we need to re-check
// the selection
if (TypeSelectNewKey(theEvent, &gTypeSelectState))
{
short listCount;
short newListIndex = 0;
Cell newSelection;
// How many items are we dealing with?
listCount = (**gTheDialogList).dataBounds.bottom;
// Ask Type Select for a new index, based
// on the current state of typing
newListIndex = TypeSelectFindItem(&gTypeSelectState,
listCount, tsNormalSelectMode,
myStringGetter, NULL);
// Unset any selected items before choosing
// a new selection
SetPt(&newSelection, 0, 0);
// Starting at the beginning, get a selected cell
while (LGetSelect(true, &newSelection, gTheDialogList))
{
// Unselect it
LSetSelect(false, newSelection, gTheDialogList);
}
// Set the new item to selected
SetPt(&newSelection, 0, newListIndex - 1);
LSetSelect(true, newSelection, gTheDialogList);
// Auto scroll to make sure the selection is visible
LAutoScroll(gTheDialogList);
// Workaround to List Box control behavior - LsetSelect
// causes an immediate redraw into the list's port,
// which in the ListBox control case is an offscreen
// buffer allocated by the control manager, and doesn't
// cause an update event to be generated for the parent
// window, so we need to force a redraw.
DrawOneDialogItem(gTheDialog, rDialogListBoxItem);
}
}
}
One of the parameters to the TypeSelectFindItem call is a callback routine that you provide to allow TypeSelection to determine the elements of the list it is choosing from. Depending on the circumstances of your implementation, the items you are selecting from might not be a straightforward list. In our case the routine is quite simple. It simply looks up the desired index in the list control we are searching in:
Listing 3: GetStringFromIndex
GetStringFromIndex
Callback routine for the type selection routine TypeSelectFindItem(). This routine is called to fetch the items in a list, and determines which of the items is the best choice considering the key strokes that have been pressed.
pascal Boolean
GetStringFromIndex(short theIndex, ScriptCode *whichScript,
StringPtr *whichString, void *ignored)
{
#pragma unused (ignored)
Boolean returnValue;
static Str255 thisString; // static because we return a
// pointer to this string
// Set the script - in this sample, we know that
// all the list items are in Roman script.
*whichScript = smRoman;
if (gTheDialogList != NULL)
{
Cell desiredCell;
short stringLength = 255;
// Fetch the item from the list,
// and get the cell data (a string) into the result
SetPt(&desiredCell, 0, theIndex - 1);
LGetCell(thisString + 1, &stringLength, desiredCell,
gTheDialogList);
thisString[0] = stringLength;
*whichString = thisString;
returnValue = true;
}
else
{
*whichString = NULL;
returnValue = false;
}
return returnValue;
}
Type selection is certainly not the top selling point of Carbon. The APIs I have described would have been appreciated if they were publicized years ago in the classic Mac OS arena, but at least they have finally arrived in Carbon. I hope this article has demonstrated that it's not difficult at all to implement this functionality, and that the benefits to the customer are overwhelming. I hope to see all of your applications sporting fancy new type selection capabilities in their next releases. Happy typing!
Thanks to Darren Litzinger for reviewing this article.
The type selection APIs consist of only four routines and one callback routine. Here is a short description of their functionality:
TypeSelectClear
Used to initialize a TypeSelectRecord, which is the data structure that holds the state of type selection at any given time. Call this routine at least once, when your application is starting up. After you launch, only call this routine if you want to intentionally void the typing the user has made at a given point.
void TypeSelectClear(TypeSelectRecord *tsr); tsr Points to a TypeSelectRecord, which requires initialization.
TypeSelectNewKey
Every time your application receives a keyDown event that might pertain to selection of a list item, you should pass the event to this routine. It examines the current buffer of characters and the value of the key event it is receiving, and returns true if the new keystroke has warranted the need to update the active list selection.
Boolean TypeSelectNewKey(const EventRecord *theEvent,
TypeSelectRecord *tsr);
theEvent A pointer to an event record containing a keyDown event.
tsr Points to a TypeSelectRecord previously initialized with TypeSelectClear.
TypeSelectFindItem
When TypeSelectNewKey returns true, use this routine to actually determine the most appropriate list item to receive the new selection. This routine takes as a parameter a callback function that you use to supply the algorithm with the contents of your list.
short TypeSelectFindItem(const TypeSelectRecord *tsr,
short listSize, TSCode selectMode,
IndexToStringUPP getStringProc,
void *yourDataPtr);
tsr Points to a TypeSelectRecord previously initialized with TypeSelectClear.
listSize The number of items in the list you are selecting from. If the number of items is unknown, pass 0x7FFF, and be sure that your callback function returns false when a requested index is not found.
selectMode Specify one of "tsPreviousSelectMode", "tsNormalSelectMode", or "tsNextSelectMode" to request the item before the matched item, the matched item itself, or the item after the matched item. The previous and next modes are what the Finder uses to respond to the Tab and Cmd-Tab keys. Typically you will pass tsNormalSelectMode.
getStringProc Pass a UPP to a routine that will fetch strings by index from the list of items you are selecting from. yourDataPtr Pass any value you would like to have passed back to your callback function.
TypeSelectCompare
TypeSelectCompare is used to compare specific items from a list, using the exact same comparison that TypeSelect would use in a call to TypeSelectFindItem. This is useful if you have a sorted list, and want to optimize item selection based on knowledge about the sorting in the list. TypeSelectFindItem compares each and every item in a list, so for very long lists pre-sorting and using TypeSelectCompare might result in a performance gain.
short TypeSelectCompare(const TypeSelectRecord * tsr,
ScriptCode testStringScript,
StringPtr testStringPtr);
tsr Points to a TypeSelectRecord previously initialized with TypeSelectClear.
testStringScript Indicates the script of the string that is being compared.
testStringPtr Points to the string that is being compared.
MyIndexToStringProc
This routine is defined by your code, and serves as an access point for TypeSelection to determine the items of the list that you are selecting from. When called, you will need to fetch and return a string from your list that is indexed by the given item number. You must return both a pointer to the string and the script code for that item.
Boolean MyIndexToStringProc(short item,
ScriptCode *itemsScript,
StringPtr *itemsStringPtr,
void *yourDataPtr);
item The index of the list item being requested.
itemsScript You return the script of the requested string here.
itemsStringPtr You return a pointer to the requested string here.
yourDataPtr A reference pointer for whatever you choose.
Daniel Jalkut is a software engineer in the Mac OS X Carbon API group at Apple Computer. In his spare time, Daniel works on his guitar playing and book reading. You can contact him at jalkut@red-sweater.com, or view his home page at http://www.red-sweater.com.
