changelog

小型游戏引擎

            so we will go back to the correct resolution, and the mouse
            will ungrab in case of a segfault. A C version (mostly working?)
            of qinterpolatedown16short() has been added to pragmas.c, and
            the ASM version had a fix that allows ceilings to show (though they
            render incorrectly in the correct place, and now certain sprites
            segfault. Ugh. PageUp/PageDown scancodes now work, so you can
            raise and lower ceilings and floors. Changed caption in titlebar
            of windowed mode.
08292000 - Matt Helsley contributed some ASM -> C conversions, and some test
            code to help with further porting. Cleaned up some cruft in the
            sdl_driver. Dan nailed another clobber list issue in pragmas.c.
            Thanks to Nicholas Vining, we now have a copy of Watcom C. Ryan
            sat down and stepped through the ASM in gdb (NOT fun at all),
            while Nicholas stepped through the DOS version in parallel.
            Eventually we stumbled upon some misassembled opcodes; we ended
            up filling some registers with a pointer, and not what the pointer
            pointed at!  Ryan went through a_linux.asm and fixed all these,
            and now 3D mode is almost entirely FLAWLESS.  This is so cool.
            Bedtime; I hope you all enjoy this.  :)  Thanks Nicholas!
08302000 - GAME now works. There was something awful in multi.c, probably
            memory corruption with a static initializer, so SDL_Init() would
            die in XOpenDisplay(NULL). Yikes. I've stubbed out what game.c
            needs for multiplayer and put it in multi_tcpip.c (guess what
            THAT'S for.  :)  ). The original multiplayer code is completely
            modem and COMit based, and is useless anywhere but DOS, so this
            isn't a big loss. Added some support for building under Cygwin,
            but not much. Note that distributing binaries of BUILD built with
            Cygwin violates the GPL, due to newlib's license, so don't do it.
            But you can try compiling your own local copies. Where the hell
            did CONTRIB go? Quick rewrite of that file. Minor glitches and
            sound (and networking?) are the only things pending.
08302000 - Temporary fix; either nsqrtasm() or its tables are not correct
            (our C rewrite gives exactly the same results as the ASM version,
            too), so we've replaced it with a call to the C runtime's sqrt()
            function, which seems to work fine. I hacked at getting better
            calls to timerhandler(); still not happy with the results.
            I also fixed some stuff in sdl_driver.c, so you can have
            resolutions > 320x200 in 3D mode. Use F4 in the game to change
            modes. Dunno how you change it in the editor, but you'll note that
            the 3D mode defaults to 640x480 now...this is where the editor
            SHOULD be, but I had 320, 320 hardcoded in _setgamemode() for some
            strange reason. Oh, and CONTRIB was lowercased. That's where it
            went. URL for this project is now listed on Ken's homepage.
09012000 - Happy September. Code tree is now available via CVS, thanks to
            Matt and "witten". Dan fixed another keysym. port i/o macros were
            being called when changing to 3D mode in the editor; #ifdef'd to
            PLATFORM_DOS, now. Misassembled opcode in msqrtasm() is now fixed,
            so the function operates correctly. Removed the #if 0, so that
            ASM now gets used.
09032000 - Dan fixed some printext16 calls in 2D mode. Added a BUILD_NOPENTIUM
            environment variable: if set, it disables PentiumPro/PentiumII/MMX
            optimizations. The code runs at more than 10fps less without
            the Pentium overlays on my box (62fps to 50fps in the start of
            nukeland.map at 640x480). I tried adding all the crazy compiler
            optimizations I could think of. First, anything above -O2 causes
            engine.c's compile to fail with a duplicate label from setgotpic.
            This is an optimizer bug in inline asm. It cuts and pastes it
            to inline it into each function, despite __volatile__ and the
            lack of an inline keyword on the function where the ASM exists,
            and you get a duplicate label. It's dumb. All the other
            optimizations I threw at it worked, including pentiumpro
            compilation, -fexpensive-optimizations, and -funroll-loops, but
            there was absolutely NO increase in frame rate. I imagine most of
            the time in 3D mode is being spent in a_linux.asm anyhow. If we
            ever get this converted to C, those optimizations will be more
            handy, I imagine, but for now, I've not committed the Makefile.
            Cleaned up the README. After some tinkering, I managed to get
            build to compile and RUN on Windows 98 with Cygwin. I can't
            distribute binaries of this, as it would violate the GPL.
            However, I will be merging the changes back into the main source
            once I figure out the best way how. Mostly, the inline ASM needed
            some hacks, but nothing I can cleanly #ifdef. Amazingly,
            mprotect() worked identically on Linux and Cygwin. The mind
            boggles. Changes that have made it into the CVS repository for the
            Cygwin work follow: Removed dependency on the SDL event thread
            (SDL_Init() fails on Win32 if you request an event thread), and
            put SDL_PumpEvents() calls into _idle() and _nextpage(). This
            seems to work fine, but there can be some breakage if there's a
            keyboard polling loop somewhere that doesn't call either of these
            functions. Added some %define macros to a_linux.asm to handle
            C compilers that append an underscore to C identifiers. Updated the
            Makefile for cygwin and other improvements.
09052000 - Checked if Duke3d maps can be edited. Apparently they can be,
            but only sorta. Quitting from 2D mode only prompts you to save
            if you've changed something, now. Fixed the CVS info in README.
            Added Duke3D and Shadow Warrior data file info to README. Minor
            cleanups in the Makefile. Moved VIDEOBASE defines to display.h.
            _platform_init() now takes icon and titlebar titles, and has been
            updated in all drivers.
02102001 - Someone on the 3DRealms Web Forums finds icculus.org:
            http://www.3drealms.com/cgi-bin/ultimatebb.cgi?ubb=get_topic&f=5&t=001318
02132001 - (Five months later...ugh.) Dan noticed that my webpage is being
            discussed on the 3DRealms forums, so I went back into the code
            and finished the work to get this running under Cygwin. Here's
            the list of that work: Makefile is now more robust, added a
            non-__attribute__ version of asm_prohlineasm4() to a.h, moved some
            #includes in cache1d.c to prevent a double-definition of O_BINARY.
            Fixed the inline asm in engine.c and pragmas.c to handle C
            compilers that add underscores to identifiers. Removed the
            SDL_HWSURFACE flag in sdl_driver.c (didn't realize this is
            actually slower on most hardware, and it also breaks fullscreen
            mode in Win32 on my box). Had to #ifndef CYGWIN the fnmatch stuff
            in build.c until I find a portable, non-GPL solution.
02142001 - Dan Olson nailed an assembly bug that caused segfaults in the
            2nd overhead view in Ken's game. After that fix was applied, I
            tagged the current CVS as "final-non-win32", as I start the effort
            to make this thing compile as a native win32 executable with
            (gasp!) Watcom C. Lots of stuff is getting torn around, as
            assumptions get challenged: PLATFORM_UNIX vs. PLATFORM_DOS? Now
            there's a PLATFORM_WIN32. unix_compat.h's display stuff has been
            moved to display.h, and all the files include "platform.h" instead
            of unix_compat.h, which allows us to choose the correct
            compatibility header at compile time (win32_compat.h,
            unix_compat.h, or doscmpat.h). All the platform-specific includes
            (io.h, and such) have been moved into their respective compat
            header, which cleaned up a lot of mess. Just to keep it real with
            Cygwin, the Makefile now autodetects Cygwin, and sets the build
            properties accordingly. Using a non-Cygwin GCC on Linux doesn't
            set the Cygwin-related properties. SDL_INC_DIR and SDL_LIB_DIR can
            be set as environment variables, so you never need to touch the
            Makefile directly with Cygwin. "make clean" now cleans up various
            Watcom C output, and vi's annoying "~" backup files. Tons of
            patches and enhancements and little cleanups just to get this
            to build on Watcom. Thing builds now, but has lots of problems
            running, which will be resolved, hopefully, tonight. The Linux
            version still works fine, and presumably so will the Cygwin
            version. Overall, the quality of the port's codebase is improving
            through this effort. I wonder if the DOS version will build now...
02152001 - Did initial work to get a protected mode DOS binary running again.
            Hey, why not? Not complete (compiles clean, no link). Lots of
            cleanups, and again, this is improving the quality of the port's
            codebase. As usual, the Linux, and presumably Cygwin, versions
            still compile clean and run. Haven't tried Watcom/Win32 tonight.
            Renamed dos_driver.c to dos_drvr.c, so you can build the DOS bits
            on a regular FAT filesystem. Added a Makefile.dos for use with
            Watcom C's wmake (based on Ken's original). Lots of other files
            touched. Added a "make package" rule to the Makefile, to improve
            the quality of my own life. Added (har) a FILEID.DIZ file, which
            is really used for the package's zipfile comment, but I always
            wanted to have one of those.  :)
02182001 - Removed the ENDLINE_CHAR stuff in build.c; that textfile reading
            code is now able to handle '\n', '\r', and "\r\n" endlines.
            The findfirst/findnext/whatnot code now works on Linux, Cygwin,
            and Watcom C (dos, and win32). It's still kludgy, though. More
            compat header cleanups. Figured out how to get SDL working fully
            with Watcom C (-ei command line to wcc386.exe, and declare your
            SDL callbacks as __cdecl, etc.).
02192001 - More Watcom fixes; build.exe's 3D mode is now fully operational.
            One graphical glitch in 2D mode, still, but otherwise appears
            fully functional. Put safety checks in a.h. Put a fix in build.c's
            overheadeditor(), so it doesn't try to reference element -1 in an
            array when clicking the left mouse button. Other ASM cleanups.
            Fixed an annoying function declaration in bstub.c. Fixed an
            incorrect assumption in sdl_driver.c's VBE_setPalette() that was
            exposed by game.c under Watcom/win32. One quick addition of the
            keyword "extern" to game.c, and we have a working game.exe under
            Windows that isn't GPL encumbered. Woohoo! Added a Watcom/win32
            Makefile, updated the README, and released win32 binaries.
02202001 - Turned on more optimizations in the Watcom makefile, and cleaned
            it up/abstracted it for public use. Added an nt_win option to that
            makefile to make it not pop up a console window when started from
            an icon. Added README-win32bins.txt to the CVS repository. Fixme
            note to self added in engine.c, and an incorrect use of ASM in
            pragmas.h was repaired. Now the drawpixel16 function is working,
            closing the last bug specific to the Watcom/win32 port. Yay.
02212001 - Fixed the timer bug! It was a one line change in game.c, pointed
            out by Matt Saettler. Amen. This led to a better abstraction:
            PLATFORM_TIMER_HZ, defined in platform.h ... Removed the wedged-in
            call to SDL_PumpEvents() in game.c...not needed anymore.
02222001 - Ported the codebase back to DOS. Point wmake.exe at Makefile.dos.
02262001 - Added code to enable write access to self-modified ASM memory pages
            so this will work on WinNT and Win2K. Discovered it wasn't needed,
            and the real problem is that Watcom produces an incorrect PE
            header in the generated EXE files. Steven Fuller dug up a program
            that fixes this, and did a ton of debugging. Added him to the
            CONTRIB file. Fixed Makefile.w32 and changed the default builds
            to optimized and not debug to make binary distribution easier.
            Automated the packaging of binary releases in Makefile.
04262001 - EDuke 2.1 goes into private beta. Will it be the last release based
            on Ken Silverman's original BUILD object code for Duke Nukem 3D?
05102001 - Embedded a Perl interpreter that runs a function in game.pl once
            per frame into the KenBuild game. This is just a test for a future
            project, and probably shouldn't be compiled by the average person.
            Changes to the Makefile, game.c, game.pl, and buildperl.[ch] were
            added. Changed all occurances of my email address from
            lokigames.com to linuxgames.com. One or two comment tweaks
            elsewhere. Updated README with SDL12 CVS instructions.