On Mon, 9 Jan 2023 at 23:43, Bernhard Beschow <shen...@gmail.com> wrote: > > > > Am 9. Januar 2023 12:08:16 UTC schrieb "Philippe Mathieu-Daudé" > <phi...@linaro.org>: > >The point of a getter() function is to not expose the structure > >internal fields. Otherwise callers could simply access the > >PFlashCFI01::mem field.
"modern" QOM style somewhat relies on providing a struct definition for a device and trusting the code that uses the device not to peek into its private fields. In this case we're only passing a pointer anyway, so as you say we can just use a typedef if we want. > The getter also works with a typedef which doesn't need the structure exposed. > > >Have the callers pass a DeviceState* argument. The QOM > >type check is done in the callee. > > Performing the cast inside the getter is essentially "lying" about the > getter's real interface which requires a PFlashCFI01 type to work properly. > Weakening the typing to the super type also weakens the compiler's ability to > catch mistakes at compile time. These mistakes can now only be found by > actually running the code or by analyzing the code very, very carefully. Yes, I'm not super-opposed to these patches but I did find myself wondering whether adding all these casts in the caller is really an improvement and what we're trying to achieve. thanks -- PMM
