GitLab

This reverts commit 81f7c71b.

clipmove() really needs concrete indication of a moving point's starting floor height for this optimization to work correctly, and inferring that ourselves without it being explicitly passed in isn't something that would work.

SW: fix buffer overflow preventing the game from starting, exposed by switching to the new memory allocation systems

r_pr_vbos 0 was broken and r_pr_vbos 1 became useless once it was clear r_pr_vbos 2 wasn't causing issues.

engine: add GL context version to glinfo, allow in 8 bpp mode (it's still backed by a GL surface, after all...)

Makefile: don't bother trying to fill VC_REV, VC_HASH, and VC_BRANCH if we don't even have a .git directory

Duke3d: fix timing issue with logo.anm playback that could occur if an actual hard video mode change is required at startup (e.g fullscreen in a non-native resolution)

This isn't really ideal, but neither is hitting the limit due to maps with an excessive number of sprites that use clipshapes.

I remembered

Duke3d: fix compatibility with savegames made before 4fdd8dc8 and e8470bd2

engine: implement buildvfs_exists() using GetFileAttributes() on Windows and stat() elsewhere, FORCE_INLINE some stuff in vfs.h

The rationale behind this change is that I was profiling CPU use during loads of particularly heavy .def files (10000+ items) and discovered an inappropriate amount of CPU time was being spent inside access(). So, I did some research and found several cases where people on stackoverflow etc had asked what the fastest method of determining whether or not a file exists was, with the results of several benchmarks being given as answers. GetFileAttributes() seemed to win on Windows and stat() seemed to win on Linux, and I did not see any results where access() was a serious contender when included in the spread of tested functions, with the best I saw being 1100-something nanoseconds for access() and 800-something for stat().

This whole thing would be faster if we just read in the entire directory index and matched files against it ourselves with just the few API calls involved with retrieving the index, but access to uncooked data littered across a directory tree isn't really something I give a shit about optimizing further at the moment.

Duke3d: don't print extended data size and offset to console when saving and loading games in release builds

MIRROR is only ever on .overpicnum for valid mirrors.

engine: tweak a couple fix16_t functions to use the versions of min/max/clamp etc that we use elsewhere instead of their own inline implementations

This never did what I was trying to do and it wasn't really clear until recently how much of a perf hit it could be. Byeeeeee.

Required for key remapping to work in the editor after the commit that added the "working directory only" checkbox.

Duke3d: in EVENT_LOADGAME, give RETURN a default value equaling the userbytever field from the savegame header and relax sv_loadheader() check

This was a cool hack for its time, but it's not useful to us anymore.