I'm writing a database access app for storing some data and want to ask a few questions about the model/view architecture.
(Using: Qt 4.7.4, own build; PostgreSQL 9.0; Targets: WinXP, Win7 (32/64 bit)) Let me first explain what I am trying to achieve and where I am currently.
I have two pages (subclassed QWidgets inserted in a QStackedWidget) with a QTableView bound to a model. Each view is bound to a table in the PostgreSQL server. You can add/edit/delete/sort/filter items. Each page can be seen by only one type of users, lets call the roles Role1 and Role2.
The submit strategies of everything connected to the model are OnManualSubmit.
(Transaction isolation level = Serializable.) When two users want to edit(for example) the same row, I want to do a "SELECT ... FOR UPDATE" query - to make sure that when someone edits something, he will merge his changes with newer ones (if any, just like in SVN for example). But I see only a submitAll() method the QSqlTableModel. Maybe catching the signals beforeUpdate(), beforeDelete(), beforeInsert() and performing manually "SELECT ... FOR UPDATE" is one option. The other way I think is to subclass QSqlTableModel. What is the clean and nice way to achieve this?
I want to periodically update the QSqlTableView for each of the pages (one page is seen at most, Role1 users have access only to Page1 and the same for Role2 => Page2). The first thing that came to my mind is to use a QTimer and manually call select() of the QSqlTableModel, but... not sure if this is the cool way.
I also want to periodically check if the connection to the database is ok, but I think that a QTimer + QSqlDatabase::isOpen () will do.
Now, the 2 tables have the same primary keys and some columns are the same. I want when a user with Role1 changes a row in Table1 to automatically change corresponding columns of Table2 and vice versa. Should I create a trigger in Postgres?
BTW, the database is small - each of the two tables is around 3-4000 rows with ~10 columns (varchars mostly, 1 text and 2 date colunms).
Thanks for reading and Happy New Year! :)
I think you should consider doing something of the following:
Instead of using QSqlTableModel as a model I'd implement my own model as a subclass of QAbstractTableModel. This will allow you a lot of control over what you can do in terms of data manipulation.
One thing that this will require is for certain fields in the table you would need to implement subclass of QAbstractItemDelegate that will allow for modification of data in the table as I am fairly sure you don't want to allow users updating any field in the table as for example primary key is likely have to be left alone.
For question 2 I would suggest implementing a field called transaction_counter for every row so you don't have to select every row in the table just the updated ones the transaction_counter will be updated on every row update and the new one will be inserted on the new row insert. One thing that will be required is that the counter is unique across the table. For example if initial state of the table is: row1 has counter = 0 and row2 has counter = 0. If row1 is updated counter set to 1. When row1 is then updated again counter on it is set to 2. When row2 is now updated counter on it is set to 3, etc. You can certainly do the data refreshes now using QTimer and this will be much more advantageous to for example checking the data as one user may be updating the same table as another user with the same Role.
For Question 3. I don't see any reason why not custom models and especially if you decide to separate data from the model you can manipulate data separately from it's display. Sort of Data->Model->View->Controller implementation. Each one can be maintained separately as long as you have a feedback mechanism for your delegates.
For Question 4. The answer is sure or you can implement the trigger in your application.
Hope this helps. Have a great New Year!