Update PS Vita touchpad ID after it has changed in SDL2 v2.30.7

[Districh-ru]

fix #9305

In SDL2 v.2.30.7 the touchpads IDs for PS Vita have been changed: libsdl-org/SDL#10537
Unfortunately, the change was done not in all the code and the front touch do not work in the current release version of SDL2 yet. :(
But a fix has already been made for the future version of SDL2: libsdl-org/SDL#11442

It's good that the VitaSDK developers forced the specified fix: vitasdk/packages#324
and everything should work fine after we update the ID for touchpad.

[Districh-ru]

I've compiled a PS Vita build on my WSL using latest VitaSDK and tested on Vita3K emulator and touch (emulation) works good.

[Districh-ru]

Hi, @kazenoyouni!
Could you please test the fix on your hardware (check that front touch works and back touch activity does not affect fheroes2).
fheroes2_psv.zip

[Districh-ru]

Pønytail from Discord has successfully tested this build on his PS Vita hardware.

[Districh-ru]

Also Pønytail has reported that he has about 40 seconds hang when he wants to start a new game for the first time after fheroes2 has launched - in Russian:

"Пока что из замеченного, при нажатии на "новую игру" приложение всё так же зависает примерно на 40 секунд, после чего нажатие на эту кнопку становится адекватным до следующего перезапуска приложения. Такое поведение тоже началось с версии 1,1,3"

But I could not reproduce such hand on Vita3K emulator.

[oleg-derevenetz]

LGTM. The only change I made was to use a different device id value depending on the SDL version, so that everything works correctly with older SDL versions, if someone decides to use them.

[kazenoyouni]

Hi, @Districh-ru !
I have confirmed that the mentioned problem has been fixed on my hardware (back touch activity also does not affect fheroes2).
Thank you very much for your quick action!

[oleg-derevenetz]

Also Pønytail has reported that he has about 40 seconds hang when he wants to start a new game for the first time after fheroes2 has launched - in Russian:

"Пока что из замеченного, при нажатии на "новую игру" приложение всё так же зависает примерно на 40 секунд, после чего нажатие на эту кнопку становится адекватным до следующего перезапуска приложения. Такое поведение тоже началось с версии 1,1,3"

But I could not reproduce such hand on Vita3K emulator.

Well, it's hard to say what this is specifically related to. Apparently, I/O on a real device is much slower than on an emulator running on modern hardware, but it's hard to tell without a real device what exactly is so slow - whether getting a list of files via std::filesystem on PS Vita or some kind of map file parsing due to some changes... Those who want to, and have a real device, can try to find out exactly which change slowed down the work using git bisect. Apparently, the result of the corresponding I/O operations is cached after the first time, and then everything works faster.

And how long did it take to get a list of maps before 1.1.3?

[Districh-ru]

Well, it's hard to say what this is specifically related to. Apparently, I/O on a real device is much slower than on an emulator running on modern hardware, but it's hard to tell without a real device what exactly is so slow - whether getting a list of files via std::filesystem on PS Vita or some kind of map file parsing due to some changes... Those who want to, and have a real device, can try to find out exactly which change slowed down the work using git bisect. Apparently, the result of the corresponding I/O operations is cached after the first time, and then everything works faster.

And how long did it take to get a list of maps before 1.1.3?

I've asked Pønytail this question and he'll answer a little later.

On my emulator I've disabled all CPU optimizations (adventure map scrolling with many objects on map gives maximum 15 FPS) and did not notice any lags in opening STANDARD GAME dialog.
So probably it is a filesystem IO issue (my emulator is located on SSD).

[ihhub]

@Districh-ru , thank you very much for the fix!

[Northfear]

Also Pønytail has reported that he has about 40 seconds hang when he wants to start a new game for the first time after fheroes2 has launched - in Russian:

"Пока что из замеченного, при нажатии на "новую игру" приложение всё так же зависает примерно на 40 секунд, после чего нажатие на эту кнопку становится адекватным до следующего перезапуска приложения. Такое поведение тоже началось с версии 1,1,3"

But I could not reproduce such hand on Vita3K emulator.

I/O becomes INCREDIBLY slow on Vita if there are 100+ files in one folder (and maps folder can have hundreds of them). The main bottleneck was in the getFilesFromDirectory function. Previously Vita had a different code path there, but it looks like everything was moved to std::filesystem recently.
Native SCE_S_ISREG macro usage was a key optimization there, since it was preventing us from accessing the file once more in order to determine whether it's a file or a folder.

I can make a PR that restores the optimization (if maintainers don't mind some platform specific code there), but I'll have to dust off my Vita first..

[oleg-derevenetz]

Hi @Northfear

Native SCE_S_ISREG macro usage was a key optimization there, since it was preventing us from accessing the file once more in order to determine whether it's a file or a folder.

I don't know how exactly std::filesystem is implemented on Vita, but std::filesystem::directory_iterator returns not just the path, but the std::filesystem::directory_entry, which should also contain the file attributes, so checking these attributes should be a no-op and there should no need to access the file once more just to get its attributes. But, of course, anything can happen, and this thing may not be optimally implemented on Vita.

I can make a PR that restores the optimization (if maintainers don't mind some platform specific code there), but I'll have to dust off my Vita first.

I just did this in #9310, so, if it builds, we can try :)

[isage]

I don't know how exactly std::filesystem is implemented on Vita

Should be same as on other platforms (std::filesystem is implemented by gcc's libstdc++ and we don't touch that (yet). This, sadly, yields some issues with absolute/relative paths. This shouldn't be an issue for fheroes2, though)

[oleg-derevenetz]

Should be same as on other platforms (std::filesystem is implemented by gcc's libstdc++ and we don't touch that (yet). This, sadly, yields some issues with absolute/relative paths. This shouldn't be an issue for fheroes2, though)

It appears, however, that enumerating directory entries using sceIoDread() works faster (on a real device) than using the std::filesystem::directory_iterator given that the additional checks are performed (like calling the directory_entry::is_regular_file()). Apparently, the latter option reads the file attributes twice.