Feedback Search Top Backward

Programming the Container Control - Part 2/3

Written by Larry Salomon, Jr.

Part1 Part2 Part3

Where were we?

Last month I briefly introduced (in that oh-so-poor writing style that makes me unique) the container control and some of the programming basics that everyone should know. This month, we will continue this trek into the unknown by describing the tree view as well as some other nifty things. Also, we will write our first application using the container, on which we will build in next month's installment.

From Another Viewpoint

As a quick note, the name and text views were omitted last month because they are - from a programmer's perspective - identical to the icon view.

The Wood Nymph's Delight

The tree view should be a highly familiar one; it parallels nicely with the directory structure of a hard disk and has been used in the File Manager (from OS/2 1.x) and the Help Manager as well as many other applications. It also has three subviews that can be specified in the CM_SETCNRINFO message: tree icon view, tree name view, and tree text view (these are specified by "or-ing" the CV_ICON, CV_NAME, or CV_TEXT flags with the CV_TREE flag in the flWindowAttr field, respectively). The difference between these views is in the way the data is represented: icon view displays an icon with the text to the right and to the left is a separate bitmap indicating whether the item is expanded or collapsed (if applicable); name view is the same as icon view except that the expanded/collapsed indicator is indicated in the icon; the text view provides only the text string with the expanded/collapsed indicator as a separate bitmap to the left.

Note the difference between icon and name view. Remember when, in the beginning of the series, it was mentioned that there are a few differences between using the MINIRECORDCORE and RECORDCORE structure? This is one of those differences; since only the RECORDCORE structure has different fields for the expanded and collapsed icons, the name view cannot be used with the CCS_MINIRECORDCORE style.

When inserting records into the container, the parent record is specified in the pRecordParent field of the RECORDINSERT structure. In the icon view, we specified this as NULL, since there is not parent/child relationships in that view. However, in the tree views, this is particularly important because it specifies how the records are to be displayed. All records that have children will be displayed by adding an expanded/collapsed indicator to its left. If, by some action of the user, you switch to any other non-tree view, all but the top-level records are hidden.


In addition to the "normal" notifications, the container provides 2 notifications specific to the tree view:

This is sent to the owner after the container collapses a subtree. mpParm2 points to the record that was collapsed.
This is sent to the owner after the container expands a subtree. mpParm2 points to the record that was expanded.

CNR2 - A Sample Application

Now let us delve into the depths of our first sample application; it not only showcases the container control, but it also contains some nifty tricks that I think you'll find useful. The structure is that of a "typical" PM application and it should not be new to you. In the code, I have placed several "landmarks"; these are marked by placing a comment of the form "@n" (where `n' is the landmark number) starting at column 55. These landmarks are used to point out things of interest and are discussed in more detail later.

The application does nothing more than add a record for each month of each year for the number of years specified in NUM_YEARS. This is to show how the tree view is used. A popup menu is displayed whenever the system key/mouse sequence is pressed, allowing to you switch between icon and tree views, etc.

Landmark 1

This is simply to point out the typedefs used.

typedef struct _CLIENTDATA {                          // @1
   USHORT usSzStruct;         // Size of the structure
   HAB habAnchor;             // Anchor block of the window
   HWND hwndFrame;            // Frame of the client (== parent)
   HWND hwndCnr;              // Container window
   HPOINTER hptrYear;         // Icon for the year records
   HPOINTER hptrMonth;        // Icon for the month records
   HWND hwndWndMenu;          // Menu window

typedef struct _MYCNRREC {
   MINIRECORDCORE mrcCore;    // Base structure
   CHAR achText[64];          // Icon text

// PFNSRCH is used by searchCnr().  The parameters are:  container window
// handle, record pointer, and user-data pointer, respectively.

CLIENTDATA is the instance data for the client window. Granted, we could have used global variables, but that is poor programming practice so we avoid doing this (the only global we use is an constant array of pointers to the names of the months).

MYCNRREC is the definition for the container records. Note that we are using the MINIRECORDCORE structure.

PFNSRCH is a pointer to a function accepting three parameters and returning a boolean. It is used in a very interesting way that you'll see later.

Landmark 2

This is only to point out that while it is highly recommended that you allocate and insert as many records as possible each time sometimes it simply isn't possible. We have to allocate/insert each year separately followed by the twelve months.

Landmark 3

Since we specified FALSE in the fInvalidateRecord field of the RECORDINSERT structure, we have to send a CM_INVALIDATERECORD message to update the container.

Landmark 4

This entire procedure is useful, since (as described in a previous issue of the magazine) you cannot specify MIS_CONDITIONALCASCADE in a resource file. It should also be noted that since the cascade button takes up a bit more space, it is helpful to add a few blank spaces in the menu template to account for this. As a general rule, I use three blanks.

Landmark 5

Here is another useful procedure; it performs a recursive, post-traversal search of the container, calling a user-specified procedure at each record to see if it matches the criteria also specified by the caller. If you've ever wondered what a use for reserved parameter is, here is one. We use the reserved parameter to specify the record we are currently checking; by having the user specify NULL, we can check to see if this is the initial call.

Of course, we could have eliminated the reserved parameter and had this procedure call another procedure with a reserved parameter to remove the burden from the caller, but that's too easy. (* grin *)

Landmark 6

This is to point out the allocation and initialization of the client instance data. Note that the cleanup is done in the WM_DESTROY processing.

Landmark 7

This is to show a bug in the container - if the user uses the mouse to invoke the popup menu, the container sends us a WM_CONTROL message. Not so if the user uses the keyboard, thus we duplicate the code and check for the WM_CONTEXTMENU message.

Landmark 8

Finally, we invoke the searchCnr() function specifying a pointer to a boolean that contains the desired select state. The searchSelect() function is an interesting one.

BOOL searchSelect(HWND hwndCnr,PMYCNRREC pmcrRecord,PBOOL pbSelect)
// This function is used to select/deselect all records.  Note that it
// always returns FALSE, so that searchCnr() will traverse the entire
// record list.
// Input:  hwndCnr - handle of the container window
//         pmcrRecord - pointer to the container record
//         pbSelect - pointer to a BOOL specifying whether to select the
//                    record or not
   return FALSE;

See how it simply sets the record's select state and always returns FALSE - indicating that this record does not match - forcing searchCnr() to traverse the entire container.

That's It!

That is all there is to it! Note that while there is a lot of setup involved, the container is no more difficult to use than any of the other "standard" controls.


We have seen how tree views can be used for hierarchical structures and how they are not much different that the icon, name, and text views described last month. We have used these concepts and introduced new ones in a sample application which will be used as a scaffold to add new features in future installments of this series.

Next month we will describe the details view, selection states, direct editing, and possibly other ideas and will incorporate these concepts into our sample application.