To understand: look at AjustSegs(void) in hw_bsp.c. It reallocates the vetex_t pointers for lines as POLYVERTEX_T pointers, and of COURSE things are gonna get wacky when you're casting pointers.
I dunno how resilient the FLOAT_TO_FIXED solution is or whether it'll be netgame compatible (yayyy float precision loss) but it's not like our builds are netgame compatible with themselves
* The NiGHTS drone had a single tic of visibility when you hit the goal, which is evident stepping frame by frame through http://gfycat.com/ComplicatedComposedAoudad (the contents of which may or may not make it into 2.2).
* When completing a NiGHTS stage with a non-zero link, the link could flash up in the final few tics before the fade to black. This just checks for player->exiting to make sure it shouldn't be shown.
OpenGL planes fix
This branch removes a specific hack in the OpenGL code for detecting if a plane is for the floor or ceiling of a sector. This it turns out fixes ceiling slopes in Boinciel's SUGOI map going missing. Though, It doesn't fix that other glitch with one FOF (must be unrelated).
If you want to test these fixes out properly, make sure working in-level sectors, FOFs AND also polyobjects all still work fine, sloped or not.
See merge request !97
Fof slope crash fix
This fixes yet another software renderer crash, this time relating to sloped FOFs: sometimes the renderer thinks parts of FOFs where the top and bottom heights are the same height are actually the bottom going through the top (in other words a negative height). This causes problems when drawing normal walls around such FOFs since the game obviously doesn't expect negative heights anywhere along the FOF - before slopes, the game could just flip around the top and bottom heights automatically with no problems. This branch also should fix crashes for all genuine cases of negative FOF heights when slopes are involved, I suppose.
Hopefully this stops FuriousFox's SUGOI map crashing for the time being, all seemed fine when I tested it out myself.
See merge request !96
-warp checking for invalid map names
I've noticed a bunch of new people getting the "Cannot warp to map 0 (out of range)" error when testing their custom maps. On asking what map name they used, it was clear they didn't use a MAPxx name at all. Sadly it was not obvious to them that other kind of map names are not accepted by SRB2, so I thought it best we make that more clear in-game.
SRB2 now gives you the error "Cannot warp to map \[your input\] (invalid map name)" if you used -warp with a param that is neither a MAPxx name or a plain number.
See merge request !99
Fix R_AddSingleSpriteDef's I_Error messages
Whoops, seems I forgot about this little branch. Basically this fixes how for a character's sprites, a full sprite lump name is displayed instead of just its sprite prefix in the "R_AddSingleSpriteDef: No patches found for [sprite prefix] frame [frame character]" message (unlike when it occurs for non-character sprites), resulting in something like "R_AddSingleSpriteDef: No patches found for PLAYC2C8 frame \^" which does NOT help custom character authors at all as it just confuses them instead. (for the record, the problem would actually with the ^ frame and not the ones displayed after "PLAY")
Oh, and the same problem is also fixed for this similar message: "R_AddSingleSpriteDef: Sprite [sprite prefix] frame [frame character] is missing rotations"
See merge request !95
for some apparent reason the compiler didn't like the while loop condition edit on its own (it complained about inline failures for P_MobjReadyToTrigger for some reason), so I had to add that extra bit above the while loop... and it was happy again, huh
If you want more specifics, sloped FOFs are to blame it turns out: sometimes the bottom of an FOF wall blocking a segment of an in-level wall column can be considered ABOVE the top part of the FOF there (yikes), and then the dc_y* values go offscreen, and then BOOM
Software crashes fix... fix
Fixes a typo introduced by merge request !75 that caused upper textures to set the wrong ceiling clipping value when not visible, allowing all sorts behind the walls to be visible. This is most noticable in GFZ2 when the inside of the tunnel section is visible
...probably a good idea to make sure this one doesn't introduce any MORE visual glitches by mistake (again, compare with 2.1.15 if possible)
See merge request !93
Fix for flats with transparent pixels on slopes
This fixes how R_DrawTiltedSplat_8 unintentionally allows the cyan pixels to NOT be considered "transparent" if after being remapped, depending on sector brightness and/or linedef type 606 colormaps, the result of remapping is not palette index 247 (the cyan we typically turn transparent). That is, the original colors from the source flat graphic are not checked, but instead the __result__ of coloring the flat under the respective colormap is checked for "transparent" pixels. This is only a problem for the tilted splat drawing function, not the regular one for non-sloped planes with cyan-pixel-using flats.
I found out about this bug from the issues Ritz was having with sloped 255-alpha translucent FOFs using transparent flats and his custom COLORMAP lump (and later when he applied a linedef type 606 colormap to the FOF) for his custom map. Thankfully he has some workarounds, but this should fix the code-side issues that caused his problems in the first place.
I also fixed stuff with another splat drawing function that's not currently used atm (maybe it will be in the future, if splats themselves are ever enabled again? *shrugs*).
See merge request !92
Skybox rendering offset fix for third person/alt view camera
Fixes the issue reported in this thread: https://mb.srb2.org/showthread.php?t=41729
I dunno if this will negatively affect any existing skyboxes in SRB2's own levels, that said. I tried out THZ2 and CEZ1 with this fix at least but I forgot to compare them with how they are in 2.1.15 so _*shrugs_*
See merge request !94
OS X Makefile build setup
This merge request:
* Cleans up the OS X bundle resource location code and fixes a SIGSEGV and memory leak
* Simplifies and fixes the OS X desktop alert code, closing more leaks
* Adds the MACOSX build flag to the Makefiles, to allow building a binary (but not Mac app yet) of SRB2.
This is intended to make it easier for developers to build on Mac OS X, without having to pull in all of XCode. You can keep using CMake if you prefer.
To test, use `make -C src MACOSX=1 NONX86=1 SDL=1 NOASM=1` for a release build.
Left to do:
* Add a content bundling script to be run after building, and a flag to trigger doing that.
`MACOSX_BUNDLE` maybe?
* Somehow get access to a Mac running PowerPC and figure out how to build a multi-platform binary.
* Add the proper magic to compile using gcc if requested. (Right now, compilation is done via LLVM/Clang)
See merge request !72
Gas jets, fan objects, springs and slopes
Whoop whoop whoop minor bugs that only get noticed due to weird experiments whoop whoop whoop
* For gas jets only: the object's standingslope is ALWAYS null. No drifting down the slope for you.
* For gas jets and fan objects: Now checks whether the player's top is below the bottom of the launcher, instead of its bottom. zdist calculation not affected, since it's signed and okay with being negative by about the height of the player.
* For all 3: the player's standingslope is NULL'd so the player isn't launched off at a wacky angle if they're standing on a slope then walk into the thing.
See merge request !91
Remove i_net.c
The code in i_net.c doesn't actually seem to be used in SRB2. I was able to compile a build without it, and hosting and joining netgames worked just fine (well, as fine as they can with the current state of the netcode...).
The vast majority of code in the file seems to be contained in HAVE_SDLNET ifdefs, and I'm pretty sure SRB2 has never used SDLNET in a public build. The only bit not contained in that block is I_InitNetwork(), which just prints an error and returns false.
Do we really need to keep it around? If not, I say get rid of it. It seems like useless clutter that is just going to confuse people who are trying to understand the source code.
See merge request !73
Re-factoring AA tree code from m_misc.c/m_misc.h into its own files
What the title says. The AA tree-related code now lives in the files m_aatree.c and m_aatree.h. Part of why I did this was to solve this m_misc.h/w_wad.h cyclic dependency problem (involving MAX_WADPATH and AA trees themselves) mentioned in the now-removed comments, another reason was ...only OpenGL uses AA trees at all, why include the relevent structs/functions/otherwise anywhere except where is necessary (which is very few files as it turns out)?
Otherwise, it just looked better on its own rather than mixed with all the other stuff already in the m_misc files. Not really important or anything affecting gameplay at all I guess.
See merge request !82
- Now checks whether the player's top is below the bottom of the fan/gas jet, instead of its bottom. zdist calculation not affected.
- mo->standingslope is NULL'd so the player isn't launched off at a wacky angle. (I also did this for springs, since Prime mentioned it was a problem for them too.)
Some slope improvements/fixes (plus P_GetMobjGravity)
Dear Red, I did some things.
* Made the slope flag SL_NOPHYSICS actually have an effect like we wanted to, but didn't get around to implementing yet - activated by setting the slope's linedef flags to have ML_NOSONIC.
* Made downhill slope thrusts proportional to an object's gravity and friction.
* To make the above happen - seperated out the gravity value finding code in P_CheckGravity into a seperate function, P_GetMobjGravity. (p_mobj.c, p_local.h) I also made this function available to Lua.
* Turned those PANIC n console messages (which would inevitably be followed up with a crash, since we're accessing invalid memory immediately after) into a descriptive I_Error.
* Put the SRB2CB type-shimming behind an ESLOPE_TYPESHIM ifdef.
* Removed SPRINGCLEAN-ifdef'd code.
* Cleaned up some eosteric comments.
* NEW SINCE RED +1'd THIS: The teetering code now takes slopes into account pretty well. There are edge circumstances as outlined in commit 9d221f4f3f, but this is unilaterally better behaviour in every way and the teetering code was kind of a mess anyways.
* NEW SINCE RED AND ALAM +1'd THIS: P_ReverseQuantiseMomentumToSlope. Simple function that replaces the inverse angle stuff (which also wasn't using InvAngle, just ANGLE_MAX - angle - which is inaccurate!!)
Current testing files available at /toaster/slptst3.wad and /toaster/gravitytest.lua on the ftp.
I want to do more to the branch like implement SL_ANCHORVERTEX in the near future, but this is probably safe to merge in its current state.
See merge request !77
The whole thing needs a refactor in general, but it's almost 2am here, I need my sleeb, and this fix would probably break something with 2.1 climbing if I made it any more/less (depending on viewpoint) complicated.
Sectorlist traversal
MOM GET THE CAMERA
There's a LOT of code in the source that ended up mixing m_snext (the node for the next thing in the sector's thinglist) and m_tnext (the node for the next sector in the thing's sectorlist), so I renamed the following:
* m_snext ==> m_thinglist_next
* m_sprev ==> m_thinglist_prev
* m_tnext ==> m_sectorlist_next
* m_tprev ==> m_sectorlist_prev
Then, I changed all the instances where the code was trying to go m_thinglist_next on a mobj's touching_sectorlist (which would've just gone to the node for the next thing in the same sector, instead of the node for the next sector for the same thing). Notable samples:
* FF_SHATTER blocks now disappear the moment you go into their sector. You still can't give FF_SOLID to them because in that case they will still stop you if you never enter their sector at all (ie - clip on corners), but having them nonsolid no longer allows you to phase through entirely without busting them (which was the whole downside of making them intangible in the first place).
* You can now bump into multiple Mario blocks at a time, even if you're not exclusively in their sector.
* No more getting randomly stopped on the edges of bouncy FOFs.
* Landing on polyobjects might behave a little more consistently at the edge of their host sector.
* Teetering did a SHITTON of code that basically never got executed, and then had directly-blockmap-accessing code as a backup. The code was activatable by replacing the m_thinglist_next with m_sectorlist_next, but it behaved SUPER differently from what we're used to with teetering (if the player mobj's edge was JUST off the edge of a platform, you ended up in teetering frames - even if it looked like you could stand) so I ended up removing that section entirely.
Any objections?
See merge request !85
Fixes and changes related to the act of setting up a level (in other words, setup fixes)
Changes made in this branch so far:
* a REJECT lump of zero length (NOT to be confused with a REJECT lump of non-zero length with all-zeros, just to clarify) should no longer cause problems with netgames and otherwise. The game now checks if the lump exists, has the right name, and length is not zero - if it fails any of these, the REJECT lump is not loaded and P_CheckSight won't be allowed to use it. This means ZDBSP (no reject) should be safe(r) to use now!
* there's now a simple devmode-only message (requires "Setup" mode) if a sector is found to have no lines at all during setup. It's a far cry from the plot to I_Error the game I once had for that unsusal scenario, but such a level still works anyway so whatever.
See merge request !87
FF_REVERSEPLATFORM clipping
Ran into an issue whilst testing out one last feature (there's always one more...) for flat_alignment_revamp. This is a backported fix.
The bug was such that if you're falling through a platform with FF_REVERSEPLATFORM and not FF_PLATFORM, you'll be pushed downwards such that your head is against the bottom of the FOF. This just checks momz is greater than zero if it has FF_REVERSEPLATFORM to make sure it's okay to set ceilingz. (OR the alternative in reverse gravity.)
No test map because this was originally done for internal. Instead, test it on the other branch's test map.
See merge request !83
Software crashes fix
This branch SHOULD fix the many crashes people have reported lately that all point to the software renderer. Simply put, the software renderer allowed stuff to be drawn out of the screen even though that wasn't safe, and even the existing checks to prevent that didn't work.
If you saw me worrying about the sky HOMs I discovered in AGZ earlier in the commits for this branch, don't worry - it turns out that issue already existed in 2.1.15's srb2win.exe (and probably srb2dd.exe too) anyway, the changes in this branch didn't cause them. Hopefully nothing else broke then.
See merge request !75
Also, the teetering angle on slopes is now FRACUNIT/2 because there's literally no way to stand still on a slope that steep unless it doesn't have physics.
This means the current skewing-by-default effect isn't changed, and OpenGL's equivalent code doesn't have to be touched since apparently it was already like that.
This should mean that maps built with ZBSDP (no reject) should have less or no problems in netgames compared to the standard ZenNode maps now, hopefully. =)
note: once this is merged into internal, you should probably remove me from "programming assistance" so i'm not duplicated for no clear reason. unless you want me to slowly take over every section in the credits >:3c
"frontsector" in this part of the code isn't actually the polyobject's sector for back-side polyobject segs, it's the in-level sector the polyobject as a whole is being rendered in it turns out.
Red apparently left in code for single-sided linedefs to NOT skew their midtextures ...but it doesn't work because it doesn't stop the skewing code from running instead, regardless of whether Effect 1 is on or not. If it's decided single-sided line midtextures shouldn't do this though, the non-skew code could just as well be thrown out lol (or something else I guess?)
When I first wrote this, I thought the .h file that contained a function declaration needed to have the same name as the .c file the function was in. Now I know that's not the case, off to p_local.h with you.
Two interesting points of note:
* The touchspecial sector flag seems to actually do its job now.
* Detection of sectors with polyobjects in seems to have done this incorrectly, but this doesn't mess with anything about touching the polies themselves so it seems to really only handle edge cases where the polyobject was too close to the border of another sector (which would've likely made rendering glitches anyways).
* There was a whole swathe of teetering code that was basically never run properly because of this mistake. I did a simple fix at first, but you started teetering whenever you were slightly less than your radius away from a sector's edge, which was completely different and undesirable behaviour. Instead, I cut out the code that was never running, and just left the hacky method in instead since it was more accurate to what we want in general.
Issue was caused by attempting to traverse the sector's thing-touching-list across all the things in the sector (which would inevitably have the same sector as the first node in mobj->touching_sectorlist) instead of traversing the thing's sector-touching-list (which has the same thing but different sector references).
I wonder how many times AJ copypasted this code with absolutely no idea why it wasn't working properly. I'll figure that out tomorrow, maybe set up some compiler macros so this mistake is never made again. For now, I must sleeb.
Behaves ALMOST as you'd expect. It gets the z position of the slope at the player coordinates when it comes to the sectorlist check (which is first), though, so there's a few oddities that are amplified with steep slopes:
* If the slope's sloping away from you at a steep angle, you might not be able to step down onto it, but you won't teeter (because it's at a step-down-able height if it extended to directly beneath you)
* If the slope's sloping towards you at a steep angle, you might end up in teetering frames when you're able to step down onto it (because it's NOT at a step-down-able height if it extended to directly beneath you)
HOWEVER, it would be pretty obnoxious to hold back code which is functionally superior in every way otherwise, and it doesn't really seem like there's a good way to get that checked tbph
Also updated any relevant project files that I can think of to include the new files, as well as the makefile of course. Some of the other project files haven't been touched in years so I'll leave those alone ...unless someone objects
Basically this makes sure numwadfiles is updated before loading the SOC/Lua scripts, so if a Lua script calls COM_BufInsertText with the contents "addfile scr_mysticrealm.wad" it can't overwrite the last written wadfile slot! Not that COM_BufInsertText really should be used like that to begin with
*Didn't take into account object scale
*Doubled force when on the ground (ignore what the comment of the line I moved says, it was relevant for slopes...)
This also led to a mistake with slopes, where I was double-multiplying by the gravity constant to get half (because of a quirk of numbers...)
Also took the opportunity to nuke or otherwise neuter a bunch of Kalaron's bizzare ramblings (most are questions which have long-been answered by Red's efforts) at the same time.
*The No Physics flag now works (Red, you might want to doublecheck this to see whether I haven't missed any eosteric stuff out). Going downhill is a little bumpy, and I'm not sure whether that's good or not. Someone help me out here?
*The SRB2CB typeshims are now behind #ifdef ESLOPE_TYPESHIM instead of #if 1 for easier disabling.
*Slopes' downhill thrusts are now scaled with regards to object gravity. This is actually untested in gravities other than normal and reverse normal but it's one line which can be easily reverted in that circumstance. I also checked with MI to make sure this is how it's calculated elsewhere, so fingers crossed this doesn't cause any edge cases.
*As a consequence of the above point, there's now a function in p_mobj.c/h that returns an object's internal gravity - seperated out from the logic of P_CheckGravity, which really didn't need to be so monolithic. Multiply by global gravity to get the thrust. This should probably be available to Lua somehow, but I have absolutely no idea where to start with that. Wolfs, maybe?
Non-comprehensive test file available at /toaster/slptst3.wad on the ftp.
Do we really need to keep it around? If not, I say get rid of it. It seems like useless clutter that is just going to confuse people who are trying to understand the source code.
* -lSDL2_mixer is already added to SDL_LDFLAGS by default, unless NOMIXER=1 is set
* -DSDLMAIN should also be added to OPTS by default for MINGW=1 builds, unless NOSDLMAIN=1 is set
NiGHTS hotfix
Fixes the following issues relating to playing as NiGHTS Super Sonic that apparently popped up between 2.1.14 and next (mostly due to the changes to SRB2's trig stuff it seems):
* Super Sonic drifts to the side at some angles around an axis, and is unable to go directly upwards or downwards as a result
* Drilling to the side when on the ground causes the drill sound to constantly restart
* CEZS's start not actually being lined up properly with the first axis means the player is not able to go backwards along the track (because the player is not actually aligned with the track properly, preventing you from touching the attached line transfer)
* trying to hug some walls such as the tall wall before the library section of CEZS allows Super Sonic to go through them
These fixes needs proper testing before this branch can be merged in, in case they accidentally break other things as a result or something.
See merge request !71
Basically I kind of worked around any potential trig inaccuracies by not using the player position directly for setting momx/momy. This way, if player->angle_pos == player->old_angle_pos, momx/momy are zero
Demo replay fixes
Changes made/bugs fixed in this branch:
* Replay camera is now controllable when climbing (https://mb.srb2.org/showthread.php?t=38668), and in waterslides
* localangle (read: the angle between you and the camera, I think) now doesn't change during demo replays in most situations, unless the player is in analog mode. Exceptions include zoomtubes and NiGHTS super
* Replay camera now doesn't act silly if the player is in analog mode (assuming you also recorded it in that mode to begin with, anyway)
See merge request !66
DO NOT MERGE THIS INTO THE INTERNAL REPO. This is a temporary 2.1.15 only fix. This commit allows an optional boolean for tan(), which when true will automatically shift angles by ANGLE_90.
the most common condition (correct drawing) shouldn't be last, however it can't be first without making the conditions longer anyway.
it's a nitpicky thing, but this is the renderer we're talking about here.
i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs i hate FOFs
Previous overflow fix resulted in extra tall FOFs disappearing up close (see: ERZ1's elevators at start)
This works "better" in that only some lighting bugs and really really finicky visual glitches show now. I give up trying to totally fix this stuff dfsdfgdgf
One more name in the credits
I probably messed up by making toast_credits based on master instead of next. If that's blocking, just delete this branch and I'll re-do it. It's a single line, though - shouldn't exactly matter so much.
Could probably stand to be merged into Internal as well, since I hadn't actually worked on any textures when we'd updated the credits there.
See merge request !69
Fix portal and plane/sky interaction
More portal-related fixes:
* Fixes rendering issue 4 from issue #21 (Slope planes render with wrong height values when visportals are visible on-screen)
* Fixes sky rendering through portals, so that the sky you see through each portal is what you'd expect to see if you were actually there. Easiest way to see what I mean is through sky 22's planet, in a map with portals at 90 degrees to the other sides respectively (the example map on the wiki for ld40, for instance).
See merge request !65
The idea is for the layman Lua user to understand better what range of values to use for mobj types, states, sfxs, player #s etc. Additionally, mobjinfo/states/sfxinfo/hudinfo tables all now have actual bound checks when accessing/editing them. Yikes, why didn't they have any before?!
Slope fixes
This branch fixes the following slope-related physics and rendering bugs:
* Rings in multiplayer stages respawning inside slopes (even despite being able to spawn ABOVE them on map load)
* Player starts spawning players inside slopes
* Elemental flame trails not appearing if a player spindashes UP a slope; see issue #21
* Dying players "jumping" off slopes to the side if they were previously standing on one
* Some issues with FOF slope rendering
* Various issues with sprites displaying through walls adjacent to slopes
Other features added:
* Objectplace now supports slopes (this is Inuyasha's doing)
* Automap in DEVMODE now supports slopes (my doing)
Just making this merge request now rather than later, ~~in case I decide not to make any more fixes for the time being~~ (this branch doesn't seem to want to die lol), and so we can get these merged in as soon as the code's all been checked over.
See merge request !50
xorshift* PRNG
This needs testing to ensure I didn't mess anything up switching function names around.
Our PRNG sucks. This is probably obvious. I wish I had known better at the time I implemented it, but oh well.
The replacement is an xorshift* PRNG variant with period 2^32 - 1 (meaning that the PRNG state will loop after four billion calls ... that's not likely to happen), versus the old PRNG's period of about 2^22 (?). The output is also much more random and less predictable; the old PRNG would fall into a predictable loop of output after about 4000 numbers were generated, which isn't much.
The PRNG here also outputs numbers as fixed point from [0,1) (that's 0 to FRACUNIT-1, in other words) instead of single bytes at a time. This makes it much easier to calculate things for, say, P_RandomRange and P_RandomKey. A new macro, P_RandomChance(p), is now in use that returns true _p_ percent of the time, where _p_ is a fixed_t probability from 0 (0%) to FRACUNIT (100%).
This doesn't affect netgames at all; the code for seed saving and restoring is identical (aside from a check to prevent seed being set to 0, which breaks xorshift PRNGs). Demos break, but A: _duh_ and B: they're already broken by all the changes to physics to accommodate slopes.
P_Random is deprecated in Lua, as the function was renamed to P_RandomByte. Aside from that, nothing special.
See merge request !64
* add skidtime, which we forgot before 2.1 release apparently
* change tics from INT16 to INT32
* change eflags from UINT8 to UINT16
* change actionspd/mindash/maxdash from INT32 to fixed_t
Until we use something besides Native MIDI to play
back MIDI music, MIDI volume changing is disabled
since it causes way too much of a damn headache.
(It's not even our fault, it's fucking MS.)
Fix for teetering on PolyObjects
So... somebody goofed and didn't realise PolyObject sectors could be added to the sector node list for each object (which is referenced by mobj->touching_sectorlist), via their linedefs if they are nearby the player (yes, PolyObject linedefs are special and get to move about the level). As it turns out, this allows even INTANGIBLE PolyObjects to make you teeter in a seemingly inexplicable way.
What is happening is that PolyObject sectors, when they are added to the mentioned lists, are then checked under normal sector teetering conditions - if the player is above the floorheight by 24 FUs, you're officially teetering unless stated otherwise. The actual PolyObject teetering code can't help you here if the conditions are right, especially if they're taller than 24 FU in height.
There are a number of things wrong with the teetering code in general that I'd like to sort eventually, but at least now teetering on PolyObjects is fixed at last ...right? Please check the branch out if you can to check this, obviously.
See merge request !54
Fixes for patch scaling and V_FLIP usage
The following issues are fixed by this branch:
* a patch using both scaling and V_FLIP does not appear in the "correct" place on the screen. Thanks to LJSonic for pointing this out to me on 'fun
* Scaled and/or V_FLIPed patches wrap at the left/right screen edges even though they're not supposed to. V_FLIP patches at these edges may also crop the wrong side of the patch if they're at the right edge.
See merge request !60
Some fixes for portals
Specifically the following things are fixed in this branch:
* a memory leak resulting from not clearing away clipping-related arrays each tic you view a portal
* a very specific crash to do with portals sometimes (unintentionally) using a hack on drawsegs that don't actually belong to them ...which results in a crash if the drawsegs in question have midtextures. I reported it on the MB a year ago, with a test map included there: https://mb.srb2.org/showthread.php?t=38199&page=4#79
* another specific crash to do with mirrored (horizontally flipped) sprites that are scaled, particularly when you cover up the left edge of them via portals at the least. Needs more testing to be absolutely sure this is fixed, and is also reproducable in the test map linked in the post above
May be fine to merge changes into master too, I don't really know exactly
See merge request !42
Camera fix
This fixes the third person camera being silly around intangible PolyObjects, particularly the fact they can affect the camera's floorz and ceilingz even though I see no reason why they should do so.
Also a good reminder that POF_SOLID is the same as POF_CLIPLINES and POF_CLIPPLANES combined, which is probably how this issue came about to begin with. Can't say that with certainty of course.
See merge request !57
P_RandomChance is now a macro for something that should happen a
certain percentage of time.
P_SignedRandom was moved to a macro. Nobody cared.
# Conflicts:
# src/p_inter.c
This actually passes most diehard tests, as opposed to the old RNG.
It's also similarly fast.
Internally the PRNG generates a fixed point number from [0,1) now,
which makes P_RandomKey and P_RandomRange much easier to
calculate. P_Random is just a simple shift, as well.
Also, the lack of floating point math in P_RandomKey and
P_RandomRange now is probably for the best.
Also makes comptime.bat work with git if able.
Development builds will now show the branch and the SHA1 hash of the revision. Also been tested to work with subversion, where it displays "Subversion r####". You know, just in case.
There is a caveat to this: The first time EvalMath is used, a
deprecated function warning will be shown to the user that tells
them to use _G[] instead.
This reverts commit 9d36cf37bd.
This appears to fix a few holes that have been known to appear with FOF slopes sometimes, dunno how they actually came about exactly but this apparently sorts them out
Lua sector lines
Adds support for "sector.lines" in Lua, an array containing all the linedefs in a particular sector variable:
#sector.lines returns the number of lines in the sector.
sector.lines\[ _i_ \] (e.g. sector.lines[0], sector.lines[1], sector.lines[2], etc ....) gives you individual linedefs in the sector. If the number you supply is equal to or greater than #sector.lines, this returns nil.
Test script for your benefit: (see comments)
Type "luatest" in console in any level and you'll get a few print messages that tests these features for sectors[0]
See merge request !32
SOC_**** lump name support
Exactly what it says on the tin: lumps with "SOC_" prefix now are read as SOC lumps like with MAINCFG/OBJCTCFG. Go nuts.
As a bonus, I've changed things with SOC lump detection so MAINCFG, OBJCTCFG and the new SOC_**** lumps are loaded in the order you find them in WAD files (rather than an arbitrary load-MAINCFG-then-load-OBJCTCFG thing as before). All of these are still loaded after Lua scripts though, mind.
See merge request !38
Lua video lib expansion
New video/drawer library functions for hud.add to use:
v.width() and v.height() return the screen width and height respectively (in other words you get the screen resolution), while v.renderer() returns the string for the renderer used ("software", "opengl", or "none").
Possibly add more stuff later, but for now these things at the least can be merged in next
See merge request !40
Add PlayerSpawn hook to Lua
I don't know how I did this but I did. Something about staring at code from 3AM till 5AM...
Here's a test script too:
```Lua
addHook("PlayerSpawn", function(player) player.health = 99 end)
```
See merge request !48
Version constants for Lua
Pretty simple thing. Allows scripters to use VERSION, SUBVERSION, and VERSIONSTRING to check the game's version and potentially futureproof scripts. Have tested and confirmed working on my end.
See merge request !55
BACKPORT: removal of music slots
Relevant commits cherry-picked. Basically everything except the internal music track name switches.
See merge request !43
The funny thing is you really can't see ANY change here unless you have a sloped FOF intersecting a sector floor/ceiling (and a second FOF on the other side), which has other problems anyway lol
BACKPORT: FF_ANIMATE simplistic state animations
this is a lot more complex due to the need to remove the dispoffset related code as well as a lot of the redefinitions; combined with the code changes due to the sprite2 system in internal master.
~~BEFORE ACCEPTING THIS: get sryder to look at and fix any possible brokenness with OpenGL MD2s~~
See merge request !45
Fof flag change hotfix
This fixes rendering issues caused by changing FOF flags mid-level via Lua or linedef type 445. Everything else that toggles FF_EXISTS already had this fix it turns out.
Not sure if this is safe for master too, could someone check that for me?
See merge request !53
This fixes how removing certain flags (such as FF_EXISTS) from FOFs can cause HOMs on walls bordering their target sectors. Also fixes how the force-FOF-to-vanish linedef special can leave behind the FOF's shadow
Fix chain launching
Basically you're not allowed to launch off a chain the frame you touch it anymore.
Coincidentally the changes here allow you to actually use PF_*STASIS in a Lua script now and have it work as you'd expect it to (it lasts for a tic).
See merge request !49
cpuaffinity/sdl_mixer changes
Hopefully this will alleviate SDL2 sound issues.
If not, hopefully this will give us info on what the hell is going on.
See merge request !51
(No, game, you can't allocate a map header for the credits, why were you doing that before and how did it not break anything??)
(cherry picked from commit e1f9a01229)
Use whatever names you want for your music. So long as you prefix the lumps with O_ or D_, it doesn't matter anymore.
DISCLAIMER: Linedef type 413 (change music) and Lua scripting is not tested.
(cherry picked from commit 025ca413a2)
# Conflicts:
# src/p_user.c
Travis-CI builds
Merge support to build SRB2 for Linux system via Travis-Ci
This will build SRB2 with GCC and Clang to make sure we do not break Linux/GNU builds
See merge request !41
Monster Iestyn's Miscellaneous (netplay-compatible) changes
Just cleaning up some unused or unnecessary things left in the source code, see the commits for exact changes made if you like.
May add more stuff to this branch later, there's no rush really.
See merge request !39
Diagonal spring rings tweak
If you didn't know before, those special placement ring lines for diagonal springs only use multiples of 45 for their angles in-game. In other words, they only face any of the 8 basic cardinal directions (N, S, E, W, NE, NW, etc). Considering that springs themselves don't follow the above behaviour, you can probably work out that's a bad thing.
This branch changes that of course, if you couldn't guess from context. Diagonal spring rings can now be placed with any angles like most thing types already could!
See merge request !37
some of the mess in here really bothers me
(cherry-picking this commit of mine from next since it only fixes a small oversight with compiling and adds a comment)
the screen texture does not need an alpha channel.
the fact that it had one made OGL copy the topmost pixel of the screen texture's alpha channel.
which, naturally results in the screen becoming partially transparent and letting you see the working texture in the background.
g_game.c: In function 'G_CheckDemoStatus':
g_game.c:5588:22: warning: logical not is only applied to the left hand side of comparison [-Wlogical-not-parentheses]
if (!modeattacking == ATTACKING_RECORD)
^
This is not how you use pushlstring! This is actually sending uninitialized memory to Lua, which is making scripts have inconsistent results (duh?)
c/o JTE: "Tell Red they're a doofus."
Admittedly I knew of this particular method from the start but wanted to avoid it in favour of a less-hacky looking method of getting sector.lines' size ...but there was none to be found at all.
Coloropposite hotfix
Quick fix to prevent ColorOpposite(MAXSKINCOLORS) or higher input from giving results out of the actual array's bounds. In other words, preventing it from giving you nonsense values or something.
I created the function for Lua to begin with, so clearly this is 100% my fault once again. Welp.
See merge request !31
Objectplace reallocates the mapthings list to add one more mapthing. By itself there's no problem with this.
But, mobj->spawnpoint is a pointer to the mapthing's location in the mapthings list.
So by reallocating the mapthings list, all references to mobj->spawnpoints point to freed memory.
... Oops.
Now when objectplace reallocates the mapthings list it actually corrects the locations of all mobj's spawnpoints to point to the new list.
Hooray, you can use NiGHTS objectplace again if you really want to.
The point here is ColorOpposite(MAXSKINCOLORS) would have given an actual result of its own since MAXSKINCOLORS & MAXSKINCOLORS is still MAXSKINCOLORS. This shouldn't happen though, as both Color_Opposite[MAXSKINCOLORS*2] and Color_Opposite[MAXSKINCOLOR*2+1] aren't defined.
Note: polyobj_t's "translucency" is apparently a SIGNED integer, so in theory it's possible to get polyobj flats to use the "spanfunc = splatfunc" line using negative values. If this is not meant to happen, this should probably be fixed asap
Conflicts:
src/f_wipe.c
Also makes comptime.bat work with git if able.
Development builds will now show the branch and the SHA1 hash of the revision. Also been tested to work with subversion, where it displays "Subversion r####". You know, just in case.
more detailed description: vissprites now store dispoffset in a separate variable from (y)scale, and uses it to influence order between sprites without it affecting the actual drawing of the sprites themselves
Fix md2s
shoutouts to MI for breaking them accidentally
I was about to just commit this straight to next but it's the perfect reason of why code review is beneficial and I'd be a hypocrite to point that out and then skip the process
See merge request !28
Apparently all parts of the source code that require GLPatch_t are themselves used only if HWRENDER is defined. Do I need to say more?
Not sure if this will fix Wolfy's latest problem or not though
Basically, Wolfy's linux (non-CMake) compiling apparently fails here, and config.in.h actually lives outside of the sdl folder. Blame a particular someone for blindly copy+pasting these includes in this file without considering the consequences when adding support for CMake everywhere.
Apparently all parts of the source code that require GLPatch_t are themselves used only if HWRENDER is defined. Do I need to say more?
Not sure if this will fix Wolfy's latest problem or not though
Basically, Wolfy's linux (non-CMake) compiling apparently fails here, and config.in.h actually lives outside of the sdl folder. Blame a particular someone for blindly copy+pasting these includes in this file without considering the consequences when adding support for CMake everywhere.
Compiling errors fixed in this commit:
* Various cases of mixed declaration and statement code
* Implicit declaration of slope functions (read: you forgot to put "include "p_slopes.h" in MORE than a few places)
* an odd case of a bad fixed_t to float typecase, cause by using P_GetZAt directly inside FIXED_TO_FLOAT
* a few minor cases of bad unsigned-signed comparisons
* no prototypes for some of the new slope functions. For goodness sake Red, this is basic stuff!
Actions tweaks
This just rewrites the action A_SetTargetsTarget currently, dunno if I'll bother to tweak any more actions in the near future \*shrugs\*
That said, PLEASE check whether the action works properly before accepting the merge. It's been a while since I last did anything with this branch, so I forget entirely whether I tested it or not.
See merge request !27
Slopes and stuff
Adds support for slopes, slopes on FOFs, slopes on translucent FOFs, slopes on FOFs with holes in the flat, slope physics, dynamic slopes, vertex slopes, dynamic vertex slopes, and a ham sandwich to the game. Only for software mode right now, though. (OGL still gets the physics and the sandwich.) Some things still need to be done, but for now this can be merged in to be finished later.
Please make sure nothing in the vanilla game breaks before giving the thumbs up for this merge.
Since this doesn't merge automatically, if the code review turns out positive and nobody else has done it, I'll handle the merging.
See merge request !22
To be specific: when a sector had a sloped ceiling and a colormap was
placed above it, the colormap wouldn't fill anything above where the
ceiling height is at the sector's midpoint. This is fixed.
Tweaks to R_PointToAngle and R_PointToAngle2
Exactly what it says in the title! See commit description for more information on what I did, since I'm too lazy to write it all up a second time. =P
Could someone check that these changes don't cause anything else in particular to go wrong in the game or source code? I haven't checked myself that much yet, mostly because I totally forgot about this thing until now, lol.
See merge request !19
Note: Before this change, North and West directions would be returned as ANGLE_90-1 and ANGLE_180-1. This caused the pusher polyobjects in THZ2 to slowly move sideways as a side-effect (and probably caused similar bugs in the past too, these functions have barely been touched in a decade it turns out.)
Goo Water (THZ Goop) adjustements
At Nev3r's request:
Adjusted goop so that you spend less time bouncing around in it. The goop will become a walkable surface with a higher velocity threshold.
The actual goop physics haven't been altered as far as the initial entrance and underwater time is concerned, only leaving goop and subsequent bounces has been dampened significantly.
See merge request !7
finesine table
I pasted in finesine from #11 and made a merge request.
This has been lightly tested to ensure the renderer doesn't immediately break. No ill effects have been observed so far.
See merge request !17
A_SetObjectFlags tweak
Only reset the sector/blockmap links on an object calling A_SetObjectFlags if the MF_NOSECTOR|MF_NOBLOCKMAP flags change. Fixes a freeze related to LD442 demonstrated in MascaraSnake's example WAD at https://dl.dropboxusercontent.com/u/27962790/statetest.wad .
See merge request !11
Re-add/fix broken platform momz mobj code.
The changes in this branch re-add the platform's momentum to players and mobjs which leave the platform (eg. by jumping) so that they move with relative velocity as expected. This behavior was unintentionally broken in SRB2 2.0, which adds a lot of artificial difficulty to certain segments of the levels, where you have to jump between high velocity moving platforms which seemingly cut your jump height to nothing.
Not only has the behavior been fixed, but it has now been enhanced to move the camera while free-falling between platforms as well, completing the illusion of full relative velocity with minimal hiccups. [Observe.](http://i.imgur.com/zmSfUyp.gifv)
See merge request !14
Miscellanous fixes to merge
These are the commits from the "miscellanous-fixes" branch that are okay to merge in.
Bugs fixed include the following:
* CTF flags respawning incorrectly: they cannot z position themselves correctly, and they cannot flip themselves.
* The weird "jumping" spring/monitor effect: this is the result of an internal mobj_t pointer (tmfloorthing, specifically) not resetting itself to NULL for the next object's thinker, resulting in Z movement code thrusting the object vertically at tmfloorthing->momz.
See merge request !18
... because some things (Lua. custom header entries) move it.
https://mb.srb2.org/showthread.php?t=40580
(Technically breaks netgame compatibility for Lua-heavy mods, so in next.)
That's supposed to be run once a frame, not once per hook
per mobj per frame you moron. If you just run it seven
thousand times a frame, of course your framerate will drop.
Dummy, what do you think you're doing? If you
just push mobjs and players into Lua all willy-
nilly everywhere, you'll wind up generating
tons of metatables and stuff you arne't even
gonna use!
Oh. Thanks me, I'm really smart.
This _should_ solve some significant performance
issues Lua experiences. If not, I will be very
upset for having wasted so much time and effort.
There will be bugs, this kind of thing needs to
be thuroughly tested and this is just the first
iteration of it.
Assume that every frame the player is on the ground, their pmomz will
be re-set properly if the floor is moving, therefore if the platform
STOPS, we need this to set it to 0.
This fixes the issue with upside-down springs shooting downwards if you touch another of its kind. Also fixes one of the issues with monitors in Icicle Falls (after you phase inside the East-most float-bob FOF's monitor via the other bug and jump up to break it while there, the NEXT monitor moves upwards too)
Here's how it works: When a player walks off the
moving platform, it applies their pmomz once, and
then _keeps pmomz set_ so that the camera still
adds pmomz to its movements until they hit another
floor. This way, the camera doesn't jerk around.
Also gain velocity from walking off an "up" elevator normally?
This _looks_ incorrect because the camera stops matching
the platform movement the moment you step off, but I
assure you it is a correct and accurate movement.
(Try it with chasecam off.)
Now players will apply platform movement when jumping,
but only if the platform is moving the same direction
as their jump is, and all other objects will have an
appropriate pmomz in reverse gravity FOF situations.
These functions were already here before, and I /swear/ the slope
physics became slightly less glitchy after switching to them...
Only issue is the slope plane mapping code hasn't been properly
converted yet, so they don't render properly for now.
tmthing must not be set outside of P_MapStart and P_MapEnd
or the game will fail a sanity check which ensures that
mobj references are not persistent across frames and crash.
Angles now go from 0 to 0xFFFF (360 degrees == FRACUNIT) instead
of using a full UINT32. Lua only has one number type, so signedness
gets in the way of using angle_t directly. This handling of angles
matches up with how ZDoom ACS scripting and the like does it.
I also changed all the integer casts and pushes of fixed_t to
their own macro in preperation for possible future seperation.
EvalMath is for SOC only.
It spawns an entirely seperate instance of Lua and requires
uppercase-only strings, and it's ability to parse strings to
enums is redundant to Lua's _G table (try using
_G["MT_BLUECRAWLA"] for instance)
Your acceleration vector parallel to the slope is reduced based on slope
angle if it's going up the slope. The pull physics' momentum increase was
toned down a bit to go along with this. Also, I removed the ifdefs for
OLD_MOVEMENT_CODE because why should that be kept around?
Actual blockmap fix
MI's "fix" was a reversion of something that allowed 2.0 maps to use the entire blockmap. This MR reverts that fix and adds a proper fix to the issue of west/south edges of the blockmap not working as they should. Tested with a thokbarrier-less square map (all sides were solid) and with AGZ (objects are tangible all around the map, like they are in 2.1.14).
See merge request !10
Re-fix the server global variable in Lua
I screwed up the conditions on my first attempt to fix this, since I only tested one scenario. Tested this in SP, at the main menu, and both clientside and (dedicated)serverside in MP. Everything works as intended.
See merge request !9
Polyobject more fixes
Extra fixes related to polyobjects; actually properly putting their flats alongside them in the draw list, and making them able to use single-waypoint zoom tube sequences. Also threw in a smoothness fix for swinging chains while I was there.
See merge request !8
Polyobject scroll hotfix
Things fixed:
* Polyobjects should now carry the same thing types as conveyors (notable example; they'll now carry Crawlas when they wouldn't before)
* The drifting issue with players on spinning polyobjects should be fixed. (I swapped in the old bad hack for a new hack that should work like it's supposed to)
See merge request !6
Now it will only bounce ONCE when you jump on it, and Knuckles
just barely gets enough velocity out of his jump for even that.
This will leave players vulnerable and annoyed for a lot less
time while they wait to finish bouncing.
...
(It's that infamous intangible West/South linedefs bug, which was really a blockmap-related bug all along. AND IT WAS SO SIMPLE TO FIX!)
git-svn-id: https://code.orospakr.ca/svn/srb2/trunk@9048 6de4a73c-47e2-0310-b8c1-93d6ecd3f8cd
Core code has too many #define dependencies on interface-specific
defines. This means that it's currently not possible to safely
separate the core and interface code into different contexts. The
core code should be refactored to accomadate for this because we
should not have any interface-specific code in core in the first
place.
This reverts the static library SRB2Core from a7135094 and instead
adds the core sources to the SRB2SDL2 target directly.
So frustrating...
Core and SDL2 are two separate targets now. Core is a static library
that is linked into SRB2SDL2. The sources for both are separated.
When using an IDE like Visual Studio or Xcode, the source code
organized into groups that explain what that group of sources does.
In the future, "Main" could be split into a few more groups based on
file prefixes, but I think the way it is set up works for now.
Makefile targets are not affected by source_groups and typing `make`
will automatically compile both the "Core" library and SRB2SDL2
itself.
Access the benefits of -debug, console devmode, and a complete gamedata.dat (all secrets unlocked) all in one go.
Moved "#if 0" and "#if 1" to "#ifdef DEVMODE" so the existance of this cheat and MD5 validation and all that can be toggled in one place too.
git-svn-id: https://code.orospakr.ca/svn/srb2/trunk@8998 6de4a73c-47e2-0310-b8c1-93d6ecd3f8cd
Todo: Stop the redundancy that currently goes on with adding MD2s on game start-up (note that HWR_AddSpriteMD2/HWR_AddPlayerMD2 are run for all sprites/skins BEFORE HWR_InitMD2 is called)
git-svn-id: https://code.orospakr.ca/svn/srb2/trunk@8988 6de4a73c-47e2-0310-b8c1-93d6ecd3f8cd
It is really messy at the moment. There is no support for copying the
necessary frameworks and dylibs out to the bundle for distribution, and
it is a frankenstein of manual find_library and find_package which can
sometimes pick up Homebrew dylibs.
Apparently the mistake also caused myargc to be cleared in Linux64 SDL2 (as pointed out by that ilag11111 guy on GitHub), so that should be fixed now too.
In some cases, the warp back to center was being detected as a mouse motion, causing all sorts of silliness with the mouse. The workaround is by only using the first motion event and ignoring every event after that, until the next call to I_GetEvent.
Fixes https://mb.srb2.org/showthread.php?t=39921 where the issue before was that when you transformed your momentum would still have you moving slowly, which could cancel out the animation.
Previously in Lua, dedicatedserver would only return true if the player running the script was the one hosting the dedicated server. In all other cases, it would return nil. This fixes it to point to the server host no matter what.
GL Initialization needs to happen before window creation,
otherwise the GL library will get reloaded while in use and
Windows will have a heart attack. This is bad, obviously.
This is a massive performance boost on slow processors, because
before, the intermediary buffer had to be swizzled to ABGR8888
before being uploaded -- for large resolutions this was an
enormous performance penalty.
MD2's can be translucent again.
MD2's can use sprites instead of another random texture if they have no
texture.
Patches are drawn in the correct place on non aspect correct
resolutions.
Cropped Patches are drawn.
