We now use a hashtable for lookups of specific paths, and organize the
entries into a directory tree. The end result is fast lookup and fast
enumeration without having to search a sorted array or tapdance with
substrings...which means the rare, mysterious bug where we failed
to find an existing file should be gone now, too.
Hopefully this will work on Win10 as well... Only time will tell!
Here's what's changed:
1. Added a platform_winrt.cpp file. It's based on platform_windows.c but has some WinRT-specific changes.
2. Changed physfs_platforms.h to define PHYSFS_PLATFORM_WINRT when running on WinRT. PHYSFS_PLATFORM_WINDOWS should also be defined, as WinRT behaves a lot like "normal" Windows and this flag is tested against in multiple parts of the PhysFS source code.
3. Changed platform_windows.c to not be used when PHYSFS_PLATFORM_WINRT is defined.
4. In order to work on Windows ARM devices, I had to change in physfs_internal.h, to include _M_ARM as a flag for running on ARM as "ARM" and "__arm__" aren't defined when running on a Windows ARM device.
What I've tested:
* Setting up a "sane folder configuration"
* Reading and writing files
* Mounting zip files
* Using seek & tell
* Checking file length
All of these work fine. I couldn't get the actual test software to run on WinRT, but it seems like eveyrthing's working. I've tested these things on my Windows 8.1 computer, as well as my Lumia 920 which is currently running Windows 8.1.
This lets an application supply its own archivers, where they will work like
any built-in archiver. This allows abstract directory interfaces the same
way that PHYSFS_Io allows stream implementations.
This is a work in progress still. The API is still changing, and will remain
at version 0 until it is finalized (a theoretical future version 1 will be
for when the final public interface changes, not when we evolve the initial
API design).
This would try to build something under the basedir for Windows 95, OS/2, etc,
but those targets are gone now. Modern systems provide this. If a given
system can't in the future, they can pull this code out of revision control
and use it in their implementation of __PHYSFS_platformCalcUserDir().
This change let me remove __PHYSFS_platformGetUserName(), too.
It's only used by archive_dir.c, and it's trivial to make it into
platform-independent code for that one module.
Bonus: less malloc() pressure...now every access to the native filesystem
doesn't require a temporary heap allocation.
It was an ugly piece of code, didn't work on The Hurd (apparently), likely
dangerous in its incomplete understanding of Windows paths, and only
used in places that should have given us normalized, absolute
paths in the first place anyhow.
Ryan C. Gordon
Martin Ahrnbom
Aleksi Nurmi