In the Delphi RTL and VCL, a method called LoadFromFile or the like will raise an exception if the file doesn’t exist or is invalid. Similarly, if an object property getter implements a lazy-loading pattern (meaning, the value of the property will only be initialised when first needed), it will raise an exception if the situation is such that nothing valid can be loaded or initialised. This is pretty basic exception programming – when an error arises, the flow of the calling code is forcibly interupted with a clear account of what call was invalid. Imagine if (say) TBitmap.LoadFromFile didn’t raise an exception when the file name passed to it didn’t refer to a valid bitmap file – calling code could go happily on its way assuming a graphic has been loaded when it hasn’t. Likewise, what if the Items property getter for TList<T> just returned nil (or equivalent) with an out of bounds index? If T is a class type, then off-by-one errors in the calling code will end up with cryptic access violations later on that might be nowhere near where the erroneous call was actually made!
Alas, but such basic understanding of exception programming is not grasped by at least one developer on the FMX team. Repeatedly across the XE2, XE3 and now (I learn) XE4 releases, methods that an experienced Delphi programmer would expect to raise an exception when nothing valid can be returned do not. As soon as one case is fixed – e.g., the TFmxObject.Children property getter was done so in XE3, albeit implicitly – another one or two are added, and frankly, it needs to stop now. The latest cases I learn are the HScrollBar and VScrollBar property getters on TScrollBox – done properly in XE3, the anti-pattern has however been introduced in XE4, causing ‘random’ access violations in TMemo (the FMX TMemo inherits from TScrollBox).