* FInterpolator depended on external references to prevent its content from getting GC'd.
* none of the pointers in the interpolation objects were declared to the GC.
The result of these issues was that changing anything about the life cycle of interpolation objects caused corrupted memory crashes when a level was changed.
- Looking over the code again, I see that discarded SysEx messages can
cause the same issue as unhandled meta events, so generalize the
returning of a NOP for everything.
Note about A_SkelMissile: The direct change of the missile's position was changed to use SetOrigin. If this is later supposed to be portal-aware, such direct coordinate changes are a no-go to ensure that everything is properly maintained.
This contains some advance work for handling line-to-line portals in A_PainShootSkull.
This function is special because it performs a map check itself instead of using one of the common functions from p_map.cpp, like most of the rest of the game code.
This commit is just preparation for upcoming changes to completely encapsulate the coordinate info in AActor because I'm going to have to work with an altered version of actor.h that cannot be committed without breaking the engine.
With this file present in the repo before work is started the changes can be committed piece by piece.
There is no need to close (and thus deallocate) console window explicitly
This will be done by autorelease pool in application controller event loop
OS X with GC and/or ARC was not affected by this issue
Older versions like 10.4 or 10.5 crashed because of double deallocation
Because frac variable has fixed_t type it must be compared with fixed one
Test case: try to break stained glass windows or open door at the beginning of Hexen's MAP01
* since we can trivially decide whether a line crosses the trace behind the end point from checking the return value of P_InterceptVector, there is no need to add those to the list of intercepts as they get discarded anyway in t
* maxfrac is always FRACUNIT so there's no need to waste some variable for a constant value.
* the sight checking code needs to be as precise as possible and should not depend on some old semi-broken routines. (This is more a precision issue of these routines - P_PointOnDivlineSide removes the lower 8 bits of each value - than having an issue with returning the wrong side in some cases.)
* for slope creations it is flat out wrong to use the old routines at all.
* also ignore this in the modern (box-shaped) case of FPathTraverse::AddLineIntercepts. This functionality is new to ZDoom and therefore not subject to compatibility concerns.
* the line-to-line teleporter. It seems the hideous fudging code was just there to work around the design issues of these functions, so let's better not ever call them here in the first place.
* A_PainShootSkull: Its usage here does not depend on these issues.
* P_ExplodeMissile: New code exclusive to ZDoom.
* FPolyObj::CheckMobjBlocking
All occurences in p_map.cpp have been left alone although most of them probably won't need the compatibility option either.
- Envelope data needed to be converted to SF2 values.
- Fine tuning was ignored which made pretty much every instrument off tune.
- Despite this, it still sounds like shit compared to FMOD or Microsoft's
wavetable synth. There are lots of missing notes and some instruments
are still off tune. I'm not sure it's worth trying to salvage it. It'd
probably be better to scrap it, since Timidity is very much oriented
toward GF1 patches, which it handles perfectly fine.
- The missing comma on the first line of dBm_pan_volume's definition
looked suspicious, so I checked the MIDI specification at
http://www.midi.org/techspecs/rp36.php and found the equations
there gave different dB values than were in the table. So I
rebuilt it using the equation given there:
20*log (sin (Pi /2* max(0,CC#10 – 1)/126))
'ceilingterrain' is needed because the top of 3D-floors refers to the model sector's ceiling, so in order to give a 3D floor a terrain it must be assignable to the sector's ceiling.
Note that although it is basically the same property, its actual function bears no relevance to its use in Eternity.
Issues with the old code:
* when calculating friction for a 3D-floor - swimmable or not - the 3D floor's top texture must be used. The previous version always checked the sector's floor texture, even though the top might as well come from the sector's ceiling.
* 3D floors never checked for the SECF_FRICTION flag at all. According to Boom specs, sector-based friction must be ignored if this is the case.
* Terrain based friction had a higher priority than the sector's own value.
Changed the rules as follows:
* if the sector's SECF_FRICTION flag is set (i.e. something explicitly changed the sector's friction), this value is used regardless of terrain settings.
* if this flag is not set, the terrain's friction is used, if defined, using the proper plane for 3D-floors.
* otherwise the default is used.
- Disconnect scripts were previously run at some point after the player
left. Now they are run immediately before destroying the player. Since
the player hasn't actually been destroyed yet, the player also gets to
be the script's activator. This gives you a chance to scrape whatever data
you want from the player before they're history. Note that if you do
anything to make the script wait, the script's activator will become the
world, as it was before.
- The old algorithm is something I threw together that produced decent,
but not spectacular results since it had a tendency to waste space by
forcing everything onto "shelves".
The new packer is the Skyline-MinWaste-WasteMap-BestFirstFit algorithm
described by Jukka Jylanki in his paper *A Thousand Ways to Pack the Bin - A
Practical Approach to Two-Dimensional Rectangle Bin Packing*, which can
currently be read at http://clb.demon.fi/files/RectangleBinPack.pdf
This is minus the optimization to rotate rectangles to make better fits.
The old method had a problem with missing order numbers and aborted the level load and made many assumptions that no longer apply with BSP based polyobject rendering.
- NODISTANCE: Disables distance checking.
- CHECKSIGHT: The qualifying actor must be in sight in order to count.
- SET<TARGET/MASTER/TRACER>: Gets the first qualifying actor and sets the calling actor's specified pointer to it.
- SETONPTR: If the function is being aimed at another actor other than the caller, sets that actor's pointers instead. Requires a SET* flag to work.
- FARTHEST: The actor farthest from the checking actor is set as the pointer. Requires a SET* flag to work.
- CLOSEST: The closest qualifying actor is set as the pointer. Requires a SET* flag to work.
- major overhaul of the static sector damage system:
* consolidated special based damage, Sector_SetDamage and UDMF properties into one set of damage properties. The parallel handling that could lead to double damage infliction was removed. This also means that damage through sector specials can be retroactively changed through Sector_SetDamage.
* all special cases were turned into flags. The new system can switch between Strife's delayed damage and regular damage, and it can also set whether terrain splashes are used or not. It also has access to the special properties of the end-level type (i.e. switching off god mode and ending the level.)
* the damage related flags are accessible through Sector_ChangeFlags, not the damage functions themselves.
Since Eternity got this it's a good candidate for a potential Super-Boom standard, and it's also useful for silencing a sector temporarily without removing the sound sequence.
- moved sector secret information from sector_t::special and secretsector to two flag bits in sector_t::Flags.
This is to get rid of the bit masking madness in the floor/ceiling thinkers which need to preserve this bit when they change a sector's type.
- fixed: Sector_SetPortal's Eternity translation was not correct.
The ZDoom variant allows specifying the plane as part of the portal, Eternity does not. Added a new 'plane' type 3 which just means 'any'.
Native OS X backed didn't center mouse cursor in fullscreen mode with Retina/HiDPI support enabled
Incorrect size of content view led to placement of cursor in upper right corner of the screen upon releasing of mouse capture
When some action is assigned to this corner using system Hot Corners feature, the given action was triggered on acquiring mouse capture
* OPL: specify the core to use for playing this song
* FluidSynth: specify a soundfont that should be used for playing the song.
* WildMidi: specify a config file that should be used for playing the song.
* Timidity++: specify an executable that should be used for playing the song. At least under Windows this allows using Timidity++ with different configs if the executable and each single config are placed in different directories.
* GUS: currently not operational, but should later also specify the config. This will need some work, because right now this is initialized only when the sound system is initialized.
* all other: no function.
These options should mainly be for end users who want to fine-tune how to play the music.
- increased the valid range of patch values for MUS. According to the original MIDI2MUS code it can handle numbers up to 188, not 181, and at least one track from Eternal Doom uses #183.
According to blzut3, it looks like it is a byte followed by a variable length field. It can be any value 0-15 and will be followed by that many bytes one for each bank used. If the bank count is 0 then it is shorthand for using one bank (bank 0).
Based on evidence from several songs in Eternal Doom the description in all known documents is wrong. The instruments are not stored in a 16-bit word but in an 8-bit byte, followed by some variable size data.
Known variations are:
* second byte is 0 - no additional data follows
* second byte is 1 - a third byte for the 'bank' value follows.
The resize indicator overlapped with progress bar and Quit/Cancel button on 10.6 and older versions
OS X 10.7 and newer don't have this indicator at all
- Did anybody actually use this? Use WildMidi instead if you want
something that sounds more like Timidity++ without actually being
Timidity++, since not even the old Timidity manages that.
Christoph Oelckers
coelckers
MajorCooke
Randy Heit
Edward Richardson
Chris
Braden Obrzut
Kyle Evans
alexey.lysiuk
Edoardo Prezioso
Christoph Oelckers
Benjamin Moir