Native Windows PE File Loading on OS X?

ozmanjusri writes "Coders working on Wine for Mac have found that the Mac loader has gained its own undocumented ability to load and understand Windows Portable Executable (PE) files. They found PE loading capabilities in Leopard that weren't there in Tiger. Further dissection showed that Apple is masking references to 'Win' and 'PE' in the dll, which means it's not an accidental inclusion. Is Apple planning native PE execution within OS X?"

Comments

Not for Win32 compatibility

[SigILL]

I don't think this is intended for Win32 compatibility. Apple has every reason not to do that, because it will mean there will be no more native versions of high-profile applications such as Photoshop. Adobe is probably already pissed off there won't be a 64-bit version of Carbon, which requires them to rewrite the entire UI of Photoshop in Cocoa to be able to release a 64-bit version of it. Giving them an easy way out by offering Win32 binary-level compatibility isn't in Apple's best interest there.

However, consider that the PE file format is also used by Microsoft's Common Language Runtime (CLR/.NET). Therefore, I think this is a preparatory move to start offering a native implementation of the .NET platform for OS X.

Re:Not for Win32 compatibility

[SigILL]

Can you quantify this supposed gain?

Sure I can... a little.

Right now, the world's colleges and universities are churning out Java & C# programmers. Those are the popular languages, the ones for which you can almost literally open up a can of programmers for.

Not so with Objective C. It's even starting to get problematic to find competent C++ programmers.

Microsoft's seen the proverbial storm coming, and has been working on an alternative for their aging and clunky Win32 API. Remember a few years back, when the Redmontians announced that Vista (then called Longhorn) would only support .NET programs natively? Back then the world evidently wasn't ready for it. But it's slowly becoming ready, because it's getting harder and harder to find competent C++ programmers.

Meanwhile, Apple is tied to Objective C. A language few people are willing to learn (remember "Objectionable C"?). For very valid technical reasons, Apple is slowly moving its developer base over from C/C++-written Carbon apps to Cocoa apps written in Objective C. However, this makes it even harder for software vendors to find competent developers for their Mac OS X offerings.

So, enter .NET. It's reasonably fast (getting faster), it has plenty of mindshare, and most importantly: there isn't much in the way of a legacy code base for it. Supporting .NET doesn't mean hurting your own APIs, its simply an additional one.

Is that enough of a quantification for you?

Re:Not for Win32 compatibility

[iJed]

However, consider that the PE file format is also used by Microsoft's Common Language Runtime (CLR/.NET). Therefore, I think this is a preparatory move to start offering a native implementation of the .NET platform for OS X.

I was just about to post this myself... It makes a lot of sense for Apple to support .NET on Mac OS X. For a start C# is now the flagship language for Windows development and not supporting it may be the difference between getting hundreds of ported apps and not getting them at all. As a Mac user and .NET developer I think it would be a big mistake for Apple to ignore .NET.

I wonder if this is how the sandboxed iPhone SDK, which is to be available in February, will be implemented

Re:Not for Win32 compatibility

[EveLibertine]

bullhonky. Apple _was_ a hardware company. Since they started selling x86 pc's, the only way to distinguish them from any other pc out there is with their software. They are a software company when it comes down to it.

Re:Not for Win32 compatibility

[coolGuyZak]

You fail to differentiate between .Net and C#. By and large, I criticize the former. I can see where you might get the wrong idea, though, so I'll elaborate.

The .Net framework suggests that the prototype for an event is as follows: "ret_type event( Object sender, EventArgs_subclass e )". Compare with a "typical" windows callback mechanism: "ret_type function( HWND hParam, WPARAM wParam, LPARAM lParam )". Suspiciously similar, neh? HWND corresponds to "Object sender", and the W- and L- PARAM objects are wrapped into EventArgs.

EventArgs and its sub-classes encapsulate all of the data given to a particular handler, much like W- and L- params, which changed meanings depending on call context. Some EventArgs subclasses also perform odd tasks, for instance the CancelEventArgs.Cancel property. This is the downright stupidest OOP implementation I've ever seen. Cancel is not data, it's an action. I don't want to specify the "cancelness" of the data, I want to cancel an operation. A better design would be to send a message back to the sending object that says, "I can't validate this." Unfortunately, because .Net event handlers use the ambiguous "object handle." I'd need a cast before I could send my response.

The complexity of implementing a cancel message is likely greater than CancelEventArgs, but the solution is far more intuitive. We don't even need to go as far as sending a message, though. Provide a real type to the sender argument (for instance, ICancelableControl, or just Control), and provide a Cancel method, and I'd be happy.

Performance is a shoddy argument for the lack of a message passing system, because .Net treats the event system as a messaging network anyway. AFAIAC, use events, but add more formality to the event system. Call your EventArgs what they are -- a message--and type the sending object appropriately. Finally, differentiate functioanlly between events and multicast delegates. Events should manage their subscription list; if an object subscribing to an event is garbage collected, fail silently. If the event lacks subscribers, then succeed.

And now for something completely different.

Everyone knows MSDN is a steaming pile of crap. What's worse, Microsoft seems to be doing very little to correct that image. IMHO, this is a mistake. As a developer, my first exposure to .Net is through MSDN--fundamentally, it's marketing for techies. It should be thorough, describing how components interact, typical real-world use cases for code, the history or motivation of a particular interface, etc. MSDN should serve the same function as an O'Reilly book--set a mood and mindset for development.

MS can spend as much money developing the perfect language as they want, but without the proper supporting tools--and don't get me started on the woes of VS--their efforts piss people off. This is precisely the motivation for the GGP's note that Objective-C developers are so "happy" and my "bullshit" post.

Re:noooo FP

[onefriedrice]

> please not - i don't need every windows malware able to run on my mac...

Except windows malware is just that: malware written for Windows. While it could potentially run, malware wouldn't automatically become a problem. You'd have a much easier time accidently running OS X malware than Windows malware. Think of it as WINE for OS X (which is apparently exactly what it is or will be except Jaguar can load the binaries itself). People running Windows binaries via WINE on Linux don't experience the same problems with malware because the expected security flaws in the underlying OS and/or applications aren't there.

In short, if Apple plans to implement a built-in WINE-like ability to run some Windows binaries in OS X, there is no reason to suspect it will cause a breeding ground for Windows malware. Malware only has the opportunity to run if it can somehow get installed.

PE Loading on x86 is easy,,,

[Jeff_at_RAD]

Our game middleware products load PE format binaries on GNU/Linux and MacOS. It's like 200 lines of code to load and fix up a DLL - not difficult at all.

The two wins are that we have one binary on all x86 platforms (which usually means testing on one platform is sufficient) and that the MS compiler generates faster binaries. The downside is that when you *do* have to debug on the non-native platforms, you must resort to printf-style debugging.

You also have to replace the default MS crt, because they make a lot of Win32 calls that you don't want to have to emulate. We just have a tiny OS layer for memory, file IO, threading, audio, and video that is native (about 30 functions).

Running Windows apps natively would make no sense

[LKM]

I think Gruber had it right when he said that Apple wants its users to think of Windows as the new Classic, i.e. if Windows apps run inside Mac OS X, they should do so the way Mac OS 9 apps used to run inside OS X: With distinctly different windows, in a separate environment, and a bit glitchy. Users need to be reminded that running Windows apps is not the preferred choice, but merely a last resort.

The idea is to tell users "Yeah, you can run your Windows apps using Parallels or VMWare if you really have to, but if you can, we'd much rather you ran real Mac applications." Running Windows apps quasi-natively by implementing the APIs would send the wrong message; it would put Windows apps on the same or a similar level as Mac apps. That's a bad thing: The Mac relies on Mac-only or "better on Macs" applications; the high quality of software is one of the Mac's selling point. If developers could write Windows apps and they would run on Macs just fine, hey, why not write Windows apps and have five or ten times the market at no additional cost?

Of course, I'd personally love to see something like this; Office for Macs is about to lose support for Macros, so I'll probably have to run Office in Parallels, soon. Come to think of it... Maybe that's Apple's way of fixing Microsoft's Macro Mistake? Maybe the idea is to let Windows Office run natively on Macs?

Anyway, Apple's actions have been extremely hard to predict recently, so I'm not ruling out anything. Maybe they are indeed going to give the Windows APIs the Carbon treatment...