Re: [hatari-devel] Replacing direct file system access with an abstraction layer?

[Nicolas Pomarède <npomarede@xxxxxxxxxxxx>]

Le 31/08/2023 à 20:38, Brad Smith a écrit :
> I recently created a new adaptation of Hatari as a Libretro core (an 
> alternative to the existing libretro/hatari), and in an Atari Forum 
> discussion Eero Tamminen suggested it might be worth incorporating some 
> of the adaptations I had to make into the main project.
>
> Atari Forum thread: 
> https://www.atari-forum.com/viewtopic.php?p=450412#p450412 
> <https://www.atari-forum.com/viewtopic.php?p=450412#p450412>
> My adaptation: https://github.com/bbbradsmith/hatariB 
> <https://github.com/bbbradsmith/hatariB>
>
> In my adaptation I had to find and replace (or remove) every direct use 
> of the filesystem. I replaced fopen, fseek, ftell, fseeko, ftello, 
> scandir, stat, etc. with an abstract replacement interface provided by 
> my core, which itself was a wrapper around Libretro's virtual file 
> system interface)
>
> Would it be worthwhile and/or desirable to do this for main Hatari? 
> Essentially the idea is to create an abstraction interface in Hatari's 
> file.h, and use it to replace all existing direct filesystem calls. 
> File_fopen, File_ftell, File_stat, etc.
>
> I hadn't initially considered it, since I don't see the direct file 
> system use as a problem for the stand-alone application, but maybe it 
> would make it easier for other adaptations like mine where the standard 
> file system is not available? It may also just be useful to consolidate 
> all the file system use in one place like this for purposes of debugging 
> and diagnostics.
>
> So, that's the proposal. I think there's some potential benefits, but I 
> didn't want to do the work of preparing a PR for this unless I first 
> understood that it was wanted.
>
> In my own adaptation I already had to do the work of finding and 
> replacing them all, so for that alone it's probably just extra work 
> rather than a benefit... though at least it could remove ~100 points of 
> change vs. the main source, which would be nice. However, I'd be happy 
> to do it if others think it would be a good change to have in the main 
> project.
>
> So, is that worth doing? Would anyone here want to see this change?

Hi

thanks for your work on hatari + retroarch, well done !

regarding the change of fopen/fread/... becoming an intermediate/wrapper 
function such as Fopen/Fread/... or any other prefix, I think I have a 
similar opinion to Eero's one (that he posted in atari-forum) :
it could be handled with some define such as :

#define File_ftell ftell
#define File_fopen fopen

In fact I had the same issue when merging WinUAE's CPU core to use it in 
Hatari : it uses many Windows' system call to access file, instead of 
the unix/linux equivalent. So instead of replacing all thoses calls in 
WinUae's source (which makes merging patch more complicated later), I 
added a list of define to replace each windows' call with the posix 
equivalent.

I think you could do the same in your retro version, just adding a few 
define and keeping all fopen/fread/... as they are in Hatari sources.

Nicolas