Input validation in dzLib

In a recent Google+ post Andrea Raimondi was mentioning the JVCL’s JvValidators components and asked whether we use/know them or not. Daniela Osterhagen mentioned that she preferred the input validation from my dzlib which led to a short discussion about how I implemented it.

I added input validation functionality to dzlib because I didn’t like the way JvValidators works. I’ll not go into the reasons but the most important one was performance, the programs using JvValidators felt like running on an old 286 computer. Instead I’m going to show here, how it currently works. Note that I am not satisfied with this yet, see below for the reasons.

First thing you do, is get an instance of the IdzInputValidator interface by calling the InputValidator function from unit u_dzInputValidator. That function takes two optional TColor parameters OKColor and ErrColor which default to clWindow and clYellow respectively.

procedure TMyForm.CheckInput;
  iv: IdzInputValidator;
  iv := InputValidator; // we just take the default colours

This interface has a number of overloaded Check methods that take different types of controls, I have implemented Check methods for T(Custom)Edit, T(Custom)ComboBox, T(Custom)Checkbox and TJvDateEdit (but see below). These methods return specialized interfaces that are meant to validate the data in theses controls. I’ll show that using a TEdit for entering an integer number between 5 and 10 (including these limits) as an example.

   // continued from above
   iv.Check(ed_NumberInput).AsInteger.IsBetween(5, 10);

If it is allowed for the TEdit to be left empty, there is this alternative:

   // continued from above
   iv.Check(ed_OptionalNumberInput).AsInteger.IsBetweenOrEmpty(5, 10);

These are actually several function calls wrapped into one line:
Check, returns an IdzEditValidator interface, AsInteger returns a IdzEditIntegerValidator interface and IsBetween / IsBetweenOrEmtpy returns a boolean. I am using interfaces because I am a lazy bastard™. I don’t want the hassle of freeing the classes that are being created on the fly.

After you have validated the input of all these controls, you ask the IdzInputValidator interface for the result. There are two overloaded GetResult methods, one simply returns a boolean, the other also returns an error message. I am assuming here that there is a TStatusBar on the form which will display this error message and an OK button that should be enabled only if the input is valid.

  // continued from above
  b_Ok.Enabled := iv.GetResult(ErrMsg); // note: Declare ErrMsg: string
  TheStatusBar.SimpleText := ErrMsg;

To make CheckInput do anything at all, you need to call it when the data in the form changes, e.g. in the OnChange events of the controls.

procedure TMyForm.ed_NumberInputChange(Sender: TObject);

Behind the scene IdzInputValidator and friends change the background colour of the input controls based on whether the input is valid or not. If it is valid, the colour is set to the OKColor, if not, it is set to the ErrColor. These colours were passed to the call to the InputValidator function right at the start and default to clWindow and clYellow. It’s a matter of (bad?) taste which colours you prefer. I like yellow because it still allows to read the input easily in contrast to e.g. red. Instead of colouring the controls it is also possible to show a blinking icon with an exclamation mark like JvValidators does but I am not going into that right now.

This implementation is already the second rewrite of the code. In the beginning IdzEditValidator used to have lots of methods like ValidateFloat, ValidateFloatBetween, ValidateInteger, ValidateIntegerBetween, ValidateDate etc. I changed it to have these As<someType> methods that return more specialized interfaces because it became unwieldy. I was pretty satisfied with this implementation until there was a requirement to validate additional controls.

The first one was TJvDateEdit. This required JvToolEdit to be added to the uses clause of u_dzInputValidator. Not every program uses the JVCL, so I added the conditional define NO_JVCL to exclude this code. Not pretty, but I could live with it.

The next one was TSigFilenameEdit (which is an internally developed control at work that is a bit similar to TJvFilenameEdit). It required an ugly hack because I could not simply add the unit that defines it to the uses clause of u_dzInputValidator.

It became worse when there was a requirement to not only validate the controls on a form but the validation was to span several frames which were put on this form.

Generally this problem is referred to as tight a coupling, in this case between the validator interface an the controls. There are a few ideas to resolve this issue:

1. I could create descendants of IdzInputValidator that know additional controls, so there would be additional Check methods that take a TSigFilenameEdit or even a Tfr_MyFrame parameter. This breaks down as soon as there are multiple, mutually exclusive controls to support. e.g. Support for JVCL controls and internal controls, because there is no multiple inheritance support in Delphi. (Hm, I’m not even sure whether mutliple inheritance could solve that problem. Maybe generics or rather my pseudo templates could? I’ll have to thing about this one.)

2. I could create overloaded functions ControlValidator that take a control and an IdzInputValidator interface like this:

function ControlValidator(_ed: TEdit; _iv: IdzInputValidator): IdzEditValidator; overload;
function ControlValidator(_ed: TJvDateEdit; _iv: IdzInputValidator): IdzJvDateEditValidator; overload;

These functions, in contrast to the methods of an interface, could be distributed over multiple units so I could include only those units that support the controls I am currently using. (This would be similar to the different IDatasetHelper implementations for BDE, ADO, ZEOS, tdbf etc. in dzlib.) This would sacrifice some convenience but would pretty much work. I don’t like the additional iv parameter, it just clutters the code. Also, the programmer would be responsible to add the unit containing the function declaration, e.g. for supporting TJvDateEdit add u_dzInputValidatorJVCL.

3. I could come up with some kind of registration mechanism for control validators that register validators for various kinds of controls with a central factory. That factory would then return the correct validator interface for a given control. That would add more complexity to the currently simple way of the validation mechanism. Also, how do I then call the actual validation methods? Have a generic interface that takes e.g. a string describing the check I want to do? No, too ugly.

4. I could revert to checking strings rather than control content. Maybe all that is required is a way to get the control’s content as a string, so TCheckbox.Checked would be converted to ‘Y’ and TCombobox.ItemIndex with IntToStr(…), so solution 3. could be simplified to supply an interface that returns a string which then gets checked. That could be even taken further to using regular expressions. (Remember that joke: You have a problem, you decide to solve it with regular expressions, now you have two problems.)

I am currently leaning towards solution 2. It allows the most flexibility without adding too much overhead. But I am going to publish this blog post on G+ to ask for feedback. Maybe somebody else has a brilliant idea?
Here is the G+ post.