qt multiple QSqlTableModels edited together in one transaction

Question

I have a window in a Qt application using PostgreSQL 9.3 database. The window is a form used do display, edit and insert new data. t looks like that:

I have data from 3 sql tables in that view. the tables are related with foreign keys:

• contractors (main table) - mapped to "personal data" section
• contacts (has foreign key to contractors.ID)
• addresses (has foreign key to contractors.ID)

So - in my window's class I have 3 main models (+ 2 proxy models to transpose tables in "personal data" an "address data" sections). I use QSqlTableModel for theese sesctions, and a QSqlRelationalTableModel for contactData section. when opening that window "normally" (to view some contractor), i simply pass contractor's ID to the constructor and store it in proper variable. Also, I call the QSqlTableModel::​setFilter(const QString & filter) method for each of the models, and set the proper filtering. When opening that window in "add new" mode i simply pass a "-1" or "0" value to the ID variable, so no data gets loaded to the model. All 3 models have QSqlTableModel::OnManualSubmit editStrategy. When saving the data (triggered by clicking a proper button), I start a transaction. And then I submit models one-by-one. personalData model gets submitted first, as I need to obtain it's PK after insert (to set in the FK fields in other models). When submitting of the model fails, I show a messageBox with the QSqlError content, rollback the transaction and return from the method. When I have an error on the first model being processed - no problem, as nothing was inserted. But when the first model is saved, but the second or third fails - there is a little problem. So I rollback the transacion as before, and return from the function. But after correcting the data and submitting it again - the first model is not trying to submit - as it doesn't know that there was a rollback, and the data needs to be inserted again. What would be a good way to notice such a model, that it needs to be submited once again? At the moment I ended up with something like that:

void kontrahenciSubWin::on_btnContractorAdd_clicked() {
    //QStringList errorList; // when error occurs in one model - whole transacion gets broken, so no need for a list
    QString error;
    QSqlDatabase db = QSqlDatabase::database();

    //backup the data - in case something fails and we have to rollback the transaction
    QSqlRecord personalDataModelrec = personalDataModel->record(0); // always one row. will get erased by SubmitAll, as no filter is set, because I don't have its ID.

    QList<QSqlRecord> contactDataModelRecList;
    for (int i = 0 ; i< contactDataModel->rowCount(); i++) {
        contactDataModelRecList.append( contactDataModel->record(i) );
    }

    QList<QSqlRecord> addressDataModelRecList;
    for (int i = 0 ; i< addressDataModel->rowCount(); i++) {
        addressDataModelRecList.append( addressDataModel->record(i) );
    }

    db.transaction();
    if ( personalDataModel->isDirty() && error.isEmpty() ) {
        if (!personalDataModel->submitAll()) //submitAll calls select() on the model, which destroys the data as the filter is invalid ("where ID = -1")
            //errorList.append( personalDataModel->lastError().databaseText() );
            error = personalDataModel->lastError().databaseText(); 
        else {
            kontrahentid = personalDataModel->query().lastInsertId().toInt(); //only here can I fetch ID
            setFilter(ALL); //and pass it to the models
        }
    }

    if ( contactDataModel->isDirty() && error.isEmpty() ) 
        if (!contactDataModel->submitAll()) //slot on_contactDataModel_beforeInsert() sets FK field
            //errorList.append( contactDataModel->lastError().databaseText() );
            error = contactDataModel->lastError().databaseText();

    if ( addressDataModel->isDirty() && error.isEmpty() )
        if (!addressDataModel->submitAll()) //slot on_addressDataModel_beforeInsert() sets FK field
            //errorList.append( addressDataModel->lastError().databaseText() );
            error = addressDataModel->lastError().databaseText();

    //if (!errorList.isEmpty()) {
    //  QMessageBox::critical(this, tr("Data was not saved!"), tr("The following errors occured:") + " \n" + errorList.join("\n"));
    if (!error.isEmpty()) {
        QMessageBox::critical(this, tr("Data was not saved!"), tr("The following errors occured:") + " \n" + error);

        db.rollback();
        personalDataModel->clear();
        contactDataModel->clear();
        addressDataModel->clear();
        initModel(ALL); //re-init models: set table and so on.

        //re-add data to the models - backup comes handy
        personalDataModel->insertRecord(-1, personalDataModelrec);

        for (QList<QSqlRecord>::iterator it = contactDataModelRecList.begin(); it != contactDataModelRecList.end(); it++) {
            contactDataModel->insertRecord(-1, *it);
        }

        for (QList<QSqlRecord>::iterator it = addressDataModelRecList.begin(); it != addressDataModelRecList.end(); it++) {
            addressDataModel->insertRecord(-1, *it);
        }

        return;
    }
    db.commit();
    isInEditMode = false;
    handleGUIOnEditModeChange();
}

Does anyone have a better idea? I doubt if it's possible to ommit backing-up the records before trying to insert them. But maybe there is a better way to "re-add" them to the model? I tried to use "setRecord", and "remoweRows" & "insertRecord" combo, but no luck. Resetting the whole model seems easiest (I only need to re-init it, as it loses table, filter, sorting and everything else when cleared)

Answer 1

I suggest you to use a function written in the language PLPGSQL. It has one transaction between BEGIN and END. If it goes wrong at a certain point of the code then will it rollback all data flawlessly.

What you are doing now is not a good design, because you handle the control over a certain functionality (rollback) to an external system with regard to the rollback (it is happening in the database). The external system is not designed to do that, while the database on the contrairy is created and designed for dealing with rollbacks and transactions. It is very good at it. Rebuilding and reinventing this functionality, which is quite complex, outside the database is asking for a lot of trouble. You will never get the same flawless rollback handling as you will have using functions within the database.

Let each system do what it can do best.

I have met your problem before and had the same line of thought to work this problem out using Hibernate in my case. Until I stepped back from my efforts and re-evaluated the situation. There are three teams working on the rollback mechanism of a database: 1. the men and women who are writing the source code of the database itself, 2. the men and women who are writing the Hibernate code, and 3. me. The first team is dedicated to the creation of a good rollback mechanism. If they fail, they have a bad product. They succeeded. The second team is dedicated to the creation of a good rollback mechanism. Their product is not failing when it is not working in very complex situations. The last team, me, is not dedicated to this problem. Who am I to write a better solution then the people of team 2 or team 1 based on the work of team 2 who were not able to get it to the level of team 1? That is when I decided to use database functions instead.