Fix graphical vga mode scaling on loading states

[JoeOsborn]

I encountered an issue similar to the one described in PR 1123. While using VGA text mode, if I made a save state during text mode and started a fullscreen game that was at a different resolution, then when I loaded the state the canvas size (notably its actual size, not its style) would still be set to a smaller value. This is because the number of rows and columns did not actually change, so the text mode display was not refreshed. I fixed this by adding a force parameter to the involved functions to give a way to ensure that the settings were refreshed even if it seemed locally to the screen code like a no-op.

[copy]

Threading a boolean flag through a bunch of functions doesn't look like a very readable solution. Can you try to find something nicer?

[JoeOsborn]

Would you prefer having restore first call set_size_text/ set_size_graphical with zeroes or ones, and then call them shortly afterwards again with the real values? Or some explicit reset method on Screen to be called at the beginning of a restore? Another idea is to remove the optimizations where Screen checks that the new values differ from the old ones.

[JoeOsborn]

I just filed a different bug about graphical text mode, #1579 , and if it is related to initialization somehow maybe the solution is connected to this issue also (but the patch in this PR does not fix it on its own). @chschnell any thoughts? Sorry for the double-ping but I realized you weren't addressed on this PR.

[copy]

Would you prefer having restore first call set_size_text/ set_size_graphical with zeroes or ones, and then call them shortly afterwards again with the real values? Or some explicit reset method on Screen to be called at the beginning of a restore? Another idea is to remove the optimizations where Screen checks that the new values differ from the old ones.

It seems that set_size_graphical_text and set_size_text do things other than just changing the size, so I'd move their body to another function that does all the actual invalidation, and then the set_size functions only implement the check, change the size and call the invalidation function.

[JoeOsborn]

So I did something slightly different, instead adding a method to invalidate the screen's internal state. I tried to do as you suggested first but it looked like I'd need to duplicate some logic around what exact calls needed to be made during set_state. Instead this roughly matches how the graphical state is invalidated during a state load in graphical mode.

[JoeOsborn]

Oh, and I'm also happy to combine these into one commit and force push a cleaner history after review but I wanted to give you a chance to review commit-by-commit first.

[copy]

Thanks! Squashed and merged.