3 October 2003
Interview with Peter Hunnisett by Brian Vincent

This week Peter Hunnisett sheds a little different perspective Wine - the gaming scene. This is our 12th interview with Wine developers. Check out the Interviews page for previous ones.

Normally I write a little intro about the person being interviewed. Peter's response was pretty amusing, so we'll just jump in with that as the first question.

BV: Wanna tell us a little about yourself; what you do at TransGaming, etc?

Peter: I live in Toronto these days after stints in Ottawa and Harlow, England. And at the risk of sounding cosmopolitan, I grew up in Toronto, did all my public schooling in Toronto (including 3 schools right in a physical row) and took Engineering Science at the University of Toronto.

Well Ove Kaaven, in jest, put it like this in #winex on IRC: " peteH is in charge so he can tell me to do all the hard work" In reality I just sort of fill in the gaps where required. I did most of the development for the 3 Kohan titles we ported to Linux, and have helped out on various parts of WineX's technology as required; for example I did most of our recent pthread changes and installer improvements.

My old hobbies seem to have vanished and have been replaced with demolition - destroying a wall can be very therapeutic - and the other aspects of renovating my 3 story house. I haven't played Ultimate all summer and have only been running sparingly.

BV: How did you get involved with Wine?

Peter: Back in 1996 I bought a new computer and I didn't have a windows install for a while but I had a bunch of games, some spare time and one of these giant books that was basically not much more than a copy of every man page on the Linux system. I was flipping through looking for something on games. Lo and behold the Wine project. I downloaded the tarball and tried to run some of my games. It sucked something fierce and I couldn't quite grok the code - the lack of documentation didn't help much - so I ignored it for a while.

A couple of years later I was bored and looking for a challenge which is when I thought of that really crappy Wine program. I bought a couple of bargain bin games and started banging my head against Wine. The rest would be not very exciting bedtime reading in the form of ChangeLog history.

BV: How did you end up working for TransGaming?

Peter: I guess I provided my first patch in 1998 and I didn't start work for TransGaming until 2001. I just sort of did a bunch of Winelib and DirectX stuff; mostly DirectPlay. I started working for Gav a while after he started up TransGaming. When he formed the company he had posted a message to the development mailing list talking about it. I got in touch with him to find out a little more about the company. With both of us being busy (I was working in England at the time), we never really got around to anything serious.

Then the telecom industry and Nortel imploded and I, an expensive ex-pat in England, was sent back to Ottawa where I had no job. One severance package later, I was enjoying my summer without work. When it was getting time to start looking for a job, I was thinking about things that I enjoyed doing or found challenging; Wine was one of them. I considered CodeWeavers, but decided that games were far more interesting. Talked with Gav and not too long later the rest was history.

BV: What parts of Wine/WineX do you like to work on?

Peter: Well for me anything game related is the only reason I'm interested in Wine so I guess I'll have to fess up that DirectX would be stuff I like to do and the major reason I'm working for TransGaming. The real rush is when you fix that last bug and the game works perfectly. That's when the "testing" begins.

I did like working on the DirectPlay stuff, it's too bad that the interaction between the DirectPlayLobby and DirectPlay stuff wasn't documented. It appears that they've at least changed their ways in that regard, but there are still several layers of interaction which are not documented so one would just have to go with a monolithic black box approach rather than exactly duplicating each of their dlls.

BV: So, does that mean you're an avid game player?

Peter: Games were the reason that I got into Wine, however, work and renovating my house keep me busy enough to generally not play man games in my spare time. Besides when you play the first minute of the Battlefield 1942 a thousand times, or install Grand Theft Auto Vice City for the umpteenth time they sort of wear a little thin. However, I did recently finish off DeusEx in a couple of marathon sessions.

BV: Ah.. installation. That's always a sore topic. Have you guys solved all of the COM problems with InstallShield?

Peter: Not by a long shot. Most of the code changes tend to be "does it get this working? Good. Stop. Does it break anything? No? Done." TransGaming Technologies has certainly solved most of the pressing DCOM problems, but we don't have 100% compatibility. In general the only applications which use DCOM, that we deal with, are installers and this generation generally works. The next generation may not, but I've got my fingers crossed.

BV: There's a thread that pops up like clockwork on wine-devel: is speeding up wineserver something that needs to be done?

Peter: Well the question of speed, as with anything, depends on your perspective. For a random application it probably doesn't matter much especially when it doesn't work. But for applications which tend to make a lot of calls doing file I/O or synchronization the slowdown can become obvious. The paper that we produced, so the shared memory wineserver would have some documentation, gave some good examples. Quake 3, for instance, saw no improvement, however Alice, which is based on the same graphics engine, but a different sound engine, got a 50% improvement in frames per second. So the old adage of "Your Mileage May Vary" applies here in spades. One thing to note, however, is that even office applications can be sped up; this isn't a multimedia specific optimization.

The basic approaches that have been discussed for reducing the expense of calling the wineserver are a kernel approach and a user space approach.

The kernel approach moves the functionality of the wineserver into a kernel module. This approach is obviously going to be fast but requires a rewrite of most of the wineserver and is by definition not particularly portable. A misbehaving application, or bug in the kernel module, could toast the whole machine.

The shared memory approach is entirely user space, shouldn't require many particular unusual software features and is reasonably portable like the rest of the code. A misbehaving application, or bug in the wineserver, could toast all windows applications running.

BV: Is WineX now using the shared memory approach?

Peter: WineX is not yet using the shared memory approach. TransGaming was hoping that more people would be interested in helping, either through code or testing, resolving some of the problem areas with the prototype. We never got that once Alexandre gave it the thumbs down for Wine; however, I keep holding out hope every time the thread comes up. We still believe that it's the right approach for WineX and we will explore it in the future, perhaps for our WineX 4.0 release.

BV: Alexandre has pretty much been against a shared memory wineserver because it doesn't isolate processes. Do you think that's a valid argument?

Peter: All that was donated was a proof of concept design. One thing that it was capable of doing was to only allow certain calls to be implemented through the shared memory mechanism. This is a good thing to be able to get the implementation up and running quickly with ease of extension - however, it also allows for individual process, or for only certain "safe" server calls, to be allowed to use the shared memory mechanism.

The concern over a particular process being able to crash other processes is a valid concern, but I'm not sure it's something that's going to be a problem in practice. The common reasons for such a crash would be either accidental or intentional scribbling. This can be avoided by simply protecting the shared memory against writes, except for the shared memory lock owner, as mentioned in the paper.

BV: One thing I've never understood about that argument is it seems to assume people are running a lot of separate independent programs. Yet, I don't really know anyone who runs more than one Windows program under Wine at a time. Am I wrong on that?

Peter: For games, it's certainly the case that only 1 visible program is really ever going to be running at once; it's intended to be an immersive experience and they're often just written to suck up 100% of the CPU. However, I believe that the ultimate goal, for Wine, is to provide something like windows where multiple windows programs could be running at the same time. I guess the ultimate question that needs to be answered is: "will Wine be the technology that fills in the application gap or will it be the end solution?"

BV: Gazing into your crystal ball, what's the answer to that question?

Peter: In my opinion, and that of my rather broken and hazy cubic zirconium crystal ball, Wine is going to be used to fill the gap. Given time, the Linux side of things will develop suitable clones for any of the interesting products and leave Wine working with the niche products.

Video games are a different story. Newer AAA titles cost between 7 and 10 million US dollars to make. If people keep supporting the nascent Linux gaming market it will eventually be worth the notice of AAA title publishers. Unfortunately, I don't see this happening any time soon so I think that Wine is going to continue to play a major role in Linux gaming.

BV: Do you think there will ever be a general purpose installation of Wine that will work for everything?

Peter: Given time there will be, but in general it's not really required. If something runs the majority of the programs that people want, it'll be good enough.

BV: Did you work on the SDL driver at all? How could it be useful?

Peter: I only did a small amount of work with the sdldrv as it was mostly hacked out by Ove Kaaven one day. The sdldrv is useful in that it is a driver capable of graphics, unlike the ttydrv. Also SDL has the ability to use several different graphical backends like X11 or a framebuffer which means that it's useful in more situations than just machines with X. In fact TransGaming has made use of the SDL driver in a few situations where the X11 driver hasn't made sense.

BV: The new threading model coming out with new distributions gave a lot of people headaches a few months ago. How did you guys end up solving it? Was your approach different than the one Alexandre came up with?

Peter: Headaches and sleepless nights :( TransGaming solved the issue in the same way that was done for the NPTL - embrace the fact that we couldn't override errno and use pthreads. TransGaming dusted off our old pthreads implementation that we had donated to WineHQ, and basically extended it so that it would work on both LinuxThreads - Wine does a runtime binary patch rather than using pthreads in this situation - and with NPTL without the user having to configure differently. They are, however, in essence the same.

BV: Since Wine and WineX diverged last year there's been some patches that have floated back and forth between the two, notably some of the ALSA stuff and Ove's DCOM work. Are the two codebases quite different now?

Peter: The portions which represent our core technologies are incredibly different. Other areas depend on the volume of change on both sides. With every WineHQ and WineX patch they diverge a little more since relatively few people license their code for the ReWind project which is how we donate and receive patches. Another issue with ReWind patches is that often X11 patches depend on functionality of LGPL patches, so ReWind is not able to use them until the patch is available with an X11 license or a different patch with equivalent functionality is available. Another problem is that many LGPL patches are made to code we contributed so those improvements never come back. This just ends up making it more difficult for ReWind to be useful to Wine.

BV: One thing that's been quite necessary in WineX is support for all the copy protection mechanisms game manufacturers use. What problems have you run into there?

Peter: There are a million and one ways to protect a game and all the solutions out there do it a little differently. The common thing that they all do is to make sure that no debuggers are present - which could figure out what they're checking - and then to attempt to verify that the CD-ROM in the drive was the specially manufactured one. This means that often the CD-ROMs have special information encoded on them that a standard reading of the disk will not pickup. For instance PS2 games are only manufactured by Sony so that they can do all sorts of special manufacturing tricks such as printing a logo onto one of the layers of the CD-ROM.

We have the problem of needing to support not only a wide variety of different copy protection systems, but also trying to support users with a number of different brands of CD-ROM drive, interfaced through the kernel in several different ways: straight IDE, IDE-SCSI, and SCSI. We have to deal with subtle kernel and hardware bugs to get these working correctly.

BV: Is all of your copy protection code proprietary?

Peter: Yes. The companies which protect their games would not like it if their techniques were disclosed publicly in an obvious way like source code.

BV: Microsoft released DirectX 9 a while ago, are there any significant additions in it?

Peter: We've only started looking at and implementing DirectX 9 recently since the games that use the new technology tend to hit the streets about a year after release. In fact the first couple are just arriving on the shelves now - the first one actually didn't even use DX9 it just had that as a requirement listed on the box. The things that present the largest challenge are those features which are not easily mappable into OpenGL features. Some of the features of the latest versions of vertex shaders and pixel shaders, for instance, in DX9 cannot be mapped into any functionality in OpenGL 1.4 - in fact their functionality may not even be in OpenGL 2.0 which is a ways from ratification. Fortunately, this functionality is presently in few graphics cards, so they are unlikely to be required for a game.

BV: For the longest time OpenGL seemed to languish with outdated features. It seems like the high-end PC video cards being released the last few years have changed that though. How well does OpenGL map to the capabilities of graphics cards?

Peter: Well I'm not really a graphics guru, but given the number of OpenGL extensions out there, it would seem obvious that the graphics card technology drives OpenGL. In general if something is missing it can be emulated in software, however, most times it's just way too computationally expensive. One exception can be vertex shaders which can be run effectively in software if the hardware is doing most of the hard graphics work. However, in general, these GPUs are pretty damm powerful, so they're almost always going to be faster.

BV: Distributions have included Wine from time to time. I've always been fairly disappointed that none have stepped up to provide even usable integration with their product (at least not with the free version of Wine.) Has TransGaming done anything special to integrate with distributions?

Peter: Well there's not that much that really can be done other than putting some intelligence into the configuration of the package and ease into using the product. There are so many different flavours of distributions. each constantly changing their technology, that you don't really want to integrate too much otherwise you spend all your time keeping up to date. About the only thing that we do that might seem a little different is that we provide rpm, deb and slack tgz packages rather than using something like the loki installer; that's generally enough package integration for TransGamers.

One thing that distributions really should do is provide a more reasonable automounter type of functionality. Some distributions do have reasonable automount functionality, and others have just bizarre ones. And the fact that you can't eject a CD-ROM while it's mounted still bothers me, but that's a *nix mentality problem. Why it can't just return an IO error when attempting to access something on an eject CD-ROM I'll never quite understand; I seem to remember NFS effectively doing that.

The big thing that TransGaming has done is to develop a graphical user interface, called Point2Play, which resolves the issues of start menu icons and package management. Basically it'll let you keep track of all your versions of WineX, manage all your games and any particular options you like to run them with.

BV: If you could add one feature to Wine, what would it be?

Peter: I'd like to have time to finish off the shared memory wineserver - let's hope that the TransGamers vote it up!