Add some option to hit an hotpath of the original router device selection
mirh opened this issue · 12 comments
See #650 (comment).
In order to play nice with Creative's official OAL router (and yes, this seems the only loophole to avoid recurrent maintenance of either system or games folders) the device name is key.
Line 919 in b8d73a2
The best way to this aim, if not any because it would be entirely addressable within openal-soft, would be for its name to always match whatever the current default device in windows is (as per K971's observations). But AFAICT this would break total mayhem in
ALC_ENUMERATE_ALL_EXT querying....unless perhaps the same knob also enabled "soundblaster-like behaviour" (i.e. only care about a single device, don't report "XX on YY")?
And so the next (and only other possible) solution would be to rely on the "partial name match" path.
I'm not 100% sure on how exactly they filter the list of the available implementations depending on the name of the current device.. but anyway long story short, if both the device and the openal driver have "X-Fi" or "Audigy" in the name they get more priority than the godawful "Generic Software" fallback.
And so, rather than just relying on some other kind of hardcoded strings, I thought that we could as well just use an environmental variable to arbitrarily change the name (so that you wouldn't have to include in the repo those names, or other awfully similar puns).
Of course the user would also have to be instructed to rename their playback device, but this is another matter.
p.s. I had originally started this feature request to just propose the second idea, but throughout the writing of the premise (and of the logical implications that would entail there aren't other ways to make OpenAL32.dll happy) the first oddball emerged too.
I'm not asking for both, or even one of them in particular.. but yeah. For the moment, after all this brainstorming I'd just like to receive a reality check if I haven't missed anything else.
EDIT: and it's ironic that they were the ones warning against "second guessing"
Spoofing this "properly" turns out to be a bit more difficult than anticipated. Adding a unique user-specified name to the enumeration lists isn't that hard, nor would it be hard to open the default device when trying to open it. However, if the spoofed name is opened, having that device report back that spoofed name as its name is a bit more involved.
Another possibility would be an option to simply exclude the "OpenAL Soft on" part of the names, and return the full device list for basic and full enumeration. Essentially treating it as if each device on the system is provided by a hardware driver. This could be a problem if there's an actual hardware OpenAL driver also providing the device with a plain un-prefixed name too, though (and I wager if someone has an X-Fi or Audigy card, that would likely be the case).
Adding a unique user-specified name to the enumeration lists isn't that hard
?
Neither of my two ideas revolved around "adding" any name.
alcDefaultName was to be replaced completely by whatever the solution would dictate.
nor would it be hard to open the default device when trying to open it. However, if the spoofed name is opened, having that device report back that spoofed name as its name is a bit more involved.
I'm getting lost here.
Another possibility would be an option to simply exclude the "OpenAL Soft on" part of the names, and return the full device list for basic and full enumeration. Essentially treating it as if each device on the system is provided by a hardware driver.
Yes, that's what I suggested in the first paragraph.
(and I wager if someone has an X-Fi or Audigy card, that would likely be the case).
Perhaps "dll names" precedence would dictate the winner. @Kappa971
Though I guess it would be certainly messy to distinguish the drivers (or could some unholy quirk just display one?).
But anyway worst case scenario would have some instructions telling the users to just delete ct_oal or sens_oal.
Adding a unique user-specified name to the enumeration lists isn't that hard
? Neither of my two ideas revolved around "adding" any name.
alcDefaultNamewas to be replaced completely by whatever the solution would dictate.
Right, but it's not that simple. alcDefaultName is the basic/driver name. For general-purpose implementations, it's the implementation name, like OpenAL Soft, Rapture3D, or Generic Software. For hardware implementations, it's the name of the hardware device it's controlling (e.g. "Sound Blaster X-Fi" or whatever Windows calls it). As long as OpenAL Soft device names are still enumerated, whatever is used in place of alcDefaultName still needs to include OpenAL Soft in the list, and similar for the ENUMERATE_ALL list. It's basically spoofing a hardware driver in soft_oal.dll, where it's still normal OpenAL Soft, just with an added device name that it pretends is a hardware device.
The alternative would be to just pretend to be one or more hardware devices, and drop all references to being OpenAL Soft (in the device names, at least).
Yes, that's actually what I was suggesting in the first scenario.
In the second one instead alcDefaultName would be free floating with whatever the users specifies.
E.g. this is what I had with just hex editing the second "OpenAL Soft" string that I found in the 1.22.2 dll.
All Available OpenAL Devices
Devices available with standard ALC_ENUMERATION_EXT
===================================================
OpenAL Spec/Ext Support
# Device Name Default 1.0 1.1 EFX
0 - OpenAL X-Fi X X X
1 - Generic Software X X
Devices available with ALC_ENUMERATE_ALL_EXT
============================================
OpenAL Spec/Ext Support
# Device Name Default 1.0 1.1 EFX
0 - OpenAL Soft on Speakers (X-Fi) (Realtek(R) Audio)
X X X
1 - OpenAL Soft on 27GL850 (NVIDIA High Definition Audio)
X X
2 - Generic Software on Speakers (X-Fi) (Realtek(R) Audio)
X X
3 - Generic Software on 27GL850 (NVIDIA High Definition Audio)
X X
Lmao, I just noticed that my little hack didn't affect the second query, yet for my intents and purpose it was good enough.
But this only happens if the dll is in SysWOW64. Otherwise the Generic Software on Speakers (X-Fi) coming from wrap_oal there wins (even once I managed to get OpenAL X-Fi on Speakers (X-Fi) too).
I have done some test in another discussion/issue.
If I remember correctly, the OpenAL router looks for "X-Fi" in both the OpenAL device name and the system device name. In my case it is Speakers (Creative SB X-Fi) as Windows device name and SB X-Fi Audio [0001] as OpenAL device name.
Another way is for the Windows device and the OpenAL device to have the exact same name, for example Speakers (Sound Blaster Z). In the case of the SB Z (and newer), it would "collide" with Host OpenAL which adopts the same solution. For Realtek cards I think there would be no problems.
If none of the above is valid, the router checks for Generic Hardware. If OpenAL Soft is named "Generic Hardware" in ALC_ENUMERATION_EXT, it would be set as the default device, regardless of the Windows device name.
Would there be any problem with calling OpenAL Soft Generic Hardware?
In the case of the SB Z (and newer), it would "collide" with Host OpenAL which adopts the same solution.
Yes, that's what we were wondering. But who wins eventually? And I imagine that you are getting two devices (with the same name) that you can hardly distinguish otherwise?
If none of the above is valid, the router checks for
Generic Hardware
It's not like they can check what doesn't exist...
Would there be any problem with calling OpenAL Soft like that?
This idea got me cringing.
Then it got my laughing (thinking that for instance mass effect has an awful in-game toggle that can only force this name)
And now I was kind puzzled (putting aside that well, it turns out something did indeed escape my tight logic).
...
And while trying to come up with some counterpoint (e.g. could some lazy ass game interpret "Generic Hardware" reversely to mean directsound?) something even harder hit me.
The wrap_oal.dll entries weren't winning (in my last comment) because of some randomness.
The dlls whose folder is the one where the loaded OpenAL32.dll is from, are those on top of the world (at least in cases where partial matching is being used). And this is bad for games shipping their own router.
I don't know then if the Generic Hardware couldn't have a different fate (its enumeration is somewhat of a special case of the special case), but if even that wouldn't play out I'm afraid that the only "drop once and forget" solution could be exact matching.
It's not like they can check what doesn't exist...
To know if the first two conditions are valid, I think the router actually checks if those names exist.
This idea got me cringing. Then it got my laughing (thinking that for instance mass effect has an awful in-game toggle that can only force this name) And now I was kind puzzled (putting aside that well, it turns out something did indeed escape my tight logic). ... And while trying to come up with some counterpoint (e.g. could some lazy ass game interpret "Generic Hardware" reversely to mean directsound?) something even harder hit me.
The wrap_oal.dll entries weren't winning (in my last comment) because of some randomness. The dlls whose folder is the one where the loaded OpenAL32.dll is from, are those on top of the world (at least in cases where partial matching is being used). And this is bad for games shipping their own router. I don't know then if the
Generic Hardwarecouldn't have a different fate (its enumeration is somewhat of a special case of the special case), but if even that wouldn't play out I'm afraid that the only "drop once and forget" solution could be exact matching.
I can't give you a technical answer, but from what I understand, Creative's OpenAL router does certain checks:
- If it finds in the device name in ALC_ENUMERATION_EXT and in the Windows device name "X-Fi", it will use this device otherwise go to step 2
- If it finds the exact same name in the device name in ALC_ENUMERATION_EXT and in the Windows device name, it will use this device otherwise go to step 3
- If it finds "Generic Hardware" in the device name in ALC_ENUMERATION_EXT, it will use this device otherwise go to step 4
- If it doesn't find any of the above, use "Generic Software" which I think is wrap_oal.dll
If the name "Generic Hardware" causes no conflict, it would solve the problem. Otherwise uninstall Creative's OpenAL router and rename soft_oal.dll to OpenAL32.dll and paste it into System32/SysWOW64.
I guess "Generic Hardware" is the name of the device in wrap_oal.dll using DirectSound3D. It is probably something dating back to Windows XP and with Sound Blaster cards not supporting OpenAL. On modern versions of Windows it would probably never be detected (perhaps forcing it with ALchemy and removing ct_oal/sens_oal or any other driver, but it wouldn't make sense).
We have the source code of the router.. there isn't much to guess, only to be smart enough to comprehend it well.
And I just realized how big of a change a small difference in implementation between ALC_ENUMERATION_EXT and ALC_ENUMERATE_ALL_EXT can make.
Here you can see the former (notice how "playback device name" based entries will always strictly win over the generic renderers)
Lines 591 to 620 in 3bedb2b
And here's the later (set aside the first priority, and just focus on what that "all devices list" would mean for partial matching)
Lines 689 to 706 in 3bedb2b
Of course every "XX device driver on YY playback device" is always gonna match the YY playback device name!
This includes openal-soft unsurprisingly, but this time "Generics" are in here too.
And the "Find Device" function is just a stupid for loop, that will break off on the first match.
This is what makes the "load order" count, insofar as you'll slyly get in front of others in the device list.
If in my previous EnumerateWin32 example, openal-soft had been 1 instead of 0 (something that can also happen in the same local folder if you just rename it to zoft_oal.dll) it would still have been screwed.
So.. uh, no. The "Generic Hardware" idea wouldn't help the slightest for our last problem of ALC_ENUMERATE_ALL_EXT partial matching. And there's nothing that gets checked before the OpenAL32.dll folder (even if we somehow weaponized windows or something, some names their router BF2OpenAL.dll or DefOpenAL32.dll) so the only way out seems the exact one.
If not any if you don't want to tinker with the folders of every single game you install (full disclaimer: I don't really know of any one that employs this ~2007 extension as opposed to the old one, your opinion is welcome on whether we should even care then as a matter of principle or not)
Of course every "XX device driver on YY playback device" is always gonna match the YY playback device name!
This includes openal-soft unsurprisingly, but this time "Generics" are in here too.And the "Find Device" function is just a stupid for loop, that will break off on the first match. This is what makes the "load order" count, insofar as you'll slyly get in front of others in the device list. If in my previous EnumerateWin32 example, openal-soft had been 1 instead of 0 (something that can also happen in the same local folder if you just rename it to
zoft_oal.dll) it would still have been screwed.So.. uh, no. The "Generic Hardware" idea wouldn't help the slightest for our last problem of
ALC_ENUMERATE_ALL_EXTpartial matching. And there's nothing that gets checked before theOpenAL32.dllfolder (even if we somehow weaponized windows or something, some names their routerBF2OpenAL.dllorDefOpenAL32.dll) so the only way out seems the exact one.If not any if you don't want to tinker with the folders of every single game you install (full disclaimer: I don't really know of any one that employs this ~2007 extension as opposed to the old one, your opinion is welcome on whether we should even care then as a matter of principle or not)
What is the point of this useless and long discussion? Neither is a programmer and arguing by making others seem to understand what you are talking about is ridiculous.
What I wrote I tried, I understand "Generic Hardware" is just a text string that Creative's router accepts as the default device if nothing else is found.
I only knew Unreal 2 to have an OpenAL router (is it a router? I don't know) which can still be disabled from the config file or renamed or deleted (so it uses OpenAL32.dll in the system folder like other games, I find it would be easier than changing the Windows device name from the registry).
If you can find a game that has problems with "Generic Hardware", then the idea is to be discarded.
@kcat could give us a more technical answer to this than we ever could, let's focus on the field tests.
I can notice that in ALC_ENUMERATE_ALL_EXT OpenAL Soft is already set as default device so I don't think there is anything to do here.
The point, which I already stated, is just for people not having to manage every single game.
You don't really need to touch the registry to change device name, but yes "Generic Hardware" could certainly avoid any extra bother once you assume Vista+ (my vagueness was just due to fact that I cannot hex edit such a long name, and thus I couldn't test it directly).
That would still still be achieved through the tentatively named ALSOFT_NAME idea, so there wasn't really too much to discern here.
But if ALC_ENUMERATE_ALL_EXT was to matter somewhere, then that still wouldn't be enough.
My results above were made from a clean folder, that didn't have the openal router together with the executable. So loading of just about every dll comes from the Windows folder, where "soft" comes alphabetically before "wrap".
But this lucky factor isn't there to save you from the precedence of folders order, and so system-only openal-soft gets behind the local-folder wrapper.
I don't know then if the
Generic Hardwarecouldn't have a different fate
Well, well, well.. I may have unduly blamed the router beforehand, but while fishing for ALC_DEFAULT_ALL_DEVICES_SPECIFIER quirks I found out discrimination can indeed happen on the side of applications.
https://github.com/OpenXRay/xray-16/blob/1144-december-2021-rc1/src/xrSound/OpenALDeviceList.cpp#L94
https://github.com/ioquake/ioq3/blob/main/code/client/snd_openal.c#L2573
In order to avoid bugs with DS3D, some engines forcefully switch you to Generic Software if the alternative is "Generic Hardware".
A selected number of developers (I reckon those with new enough of an interest to look into openal after 2007) also did actually switch to using ALC_ENUMERATE_ALL_EXT when detected.. but they didn't really re-check the logic they previously had, let alone re-thinking the implications of the difference between devices and endpoints.
By all means, this is only going to affect a minuscule fraction of applications (albeit this quick github search may have been quite the skewed sample) but nonetheless few is not zero.
https://github.com/OpenMW/openmw/blob/openmw-0.47.0/apps/openmw/mwsound/openal_output.cpp
And now that I think to it, even those who did thoroughly take care of letting the user in charge.. Still sounds at the mercy of whatever the list order is, in order to pick up one to start with the first time (I'll grant this is really a trifle though)
But with all of this said then, not all is doom and gloom eventually.
Lines 1860 to 1861 in 3bedb2b
The router itself is still relying on the default device from
ALC_ENUMERATION_EXT for alcOpenDevice(NULL) calls.And IMHO games fancy enough to have incorporated the newer extension should also be fancy enough to still have somebody somewhere supporting them somewhat (hell, some recent ones even outright decided to just come built-in with openal-soft and screw everything else).
So not all is lost for the
ALSOFT_NAME camp, which I think is the best shoot at trying to eat one's cake of 100% simple compatibility and have it too with "Enumerate All" features (just remember for the string to be a strict superset of "Generic Hardware", to avoid the consequences of XP-drivers hate).
But the tentatively named SOUNDBLASTER/XP/COMPATIBILITY__MODE still sounds the safest bet for the time being.
So, after much pondering I think both proposal should be implemented.
Perhaps.. would it be possible for soft_oal to detect if it's getting called directly by some executable, or if instead OpenAL32.dll is involved? This way you could minimize the scope of the latest hack.
Of note that (of course, duh) older OpenAL routers would only have Audigy available for partial matching.
https://github.com/rpavlik/openal-svn-mirror/blob/653d01b45fad8abccc41741b55318140f9a8a1da/win/Router/alc.cpp#L1858L1861 No wonder that sometimes even newer OG soundblaster users had to bang their heads.
Btw "Generic Hardware" may only be accepted if it's perfectly matching with the device name here (if it even works at all.. though again I cannot really be certain without actual testing). Meaning that we are kinda back to square one about any one solution being strictly better than the others.
EDIT: actually, I was reading a bit more into the S.T.A.L.K.E.R. source code, and they had an eax_unwanted property to blacklist all "Generic" devices from using eax