Microsoft, please support (at least a tiny bit of) C99 | Hacker News Reader

Microsoft, please support (at least a tiny bit of) C99

blog.reverberate.org

51 pointsbluemoon34 comments

Article author here. If you have similar pain, you can show your support for this here: http://visualstudio.uservoice.com/forums/121579-visual-studi...

I know this is probably a losing battle, but I feel better at least being able to express my displeasure about the issue.

Voted. Thanks for doing this!

It's not that there are no alternatives..

http://gcc.gnu.org/

http://clang.llvm.org/

http://software.intel.com/en-us/intel-sdp-home/

http://www.pathscale.com/ekopath-compiler-suite

The reason given when this same article was discussed on reddit was that library authors need to use the same compiler as their customers. In other words, a GCC-compiled Windows DLL is not a drop-in replacement for a MSVC-compiled DLL.

I'm sure this is true for C++ which has a crazily complex ABI for things like classes and exceptions. But is this really the case for plain C? (not in the Linux world AFAIK)

Just as a comparison, I've heard that Oracle will most likely adopt GCC's C11 ABI when C11 support is added so that one actually will be able to mix-and-match objects from both compilers.

x264 has worked fine in many many MSVC applications, even compiled in minGW with gcc.

There are some minor catches:

1. If you want a .lib instead of a .dll, you have to make one yourself separately using a tool. Not a big deal, it's just that gcc does do it for you.

2. No debug symbols, because MSVC doesn't support DWARF/etc.

3. gcc guarantees 16-byte stack alignment on x86_32, but only with respect to the caller; MSVC does not, so the stack may not be aligned as gcc expects it. You can use -mpreferred-stack-boundary to make gcc expect it, or you can just explicitly align the stack, either with the gcc intrinsic (requires ~gcc 4.2) or a small assembly function.

There are probably a few others, but overall, it does work; cdecl is cdecl.

C works fine, I even toyed with a kernel module build using MinGW instead of the WDK. One caveat is that GCC doesn't generate PDB information, the debugging format used by Microsoft tools.

Only C++ doesn't work.

Pathscale compiler doesn't work on Windows, even though they claimed a beta will be relesed at some point. Clang+LLVM on Windows is very immature, I like to play with it and put it to the test once in a while but it has many problems under Windows I wouldn't trust it in production yet.

Gcc works and produces good code, unfortunately you have to deal with the disgrace that is MinGW and it doesn't integrate in Visual Studio. Most Windows developers I know care about Visual Studio integration.

The Intel compiler produces excellent code, it is C99 and Microsoft compatible, and it seaminglessly integrates with Visual Studio. Unfortunately, it's very expensive, I have never seen it used in production.

When discussing this topic, people like to say that C is more commonly used that C++. But I think that the reality is that -Microsoft's customers- use C++ much more than C. I am such a customer and in my field 95% or more of programming is C++.

I still wish they would support a few very basic C99 features though.

I think there might be a bit more than people realise - for example, the last company I worked for did primarily C++ Windows applications, but we had several third-party C libraries built in as well (Python, GDAL, proj, etc) which we needed to be able to build in VC++. Presumably this kind of use case is exactly the sort of thing which would prevent libraries like Python from moving on to use C99 features.

I definitely agree that it would be much better if MS just supported those few features, which frankly don't seem that hard in the scheme of things. Makes me think that they see some advantage to them in not supporting it, although I'm not sure what that'd actually be.

Yes, a few of the libraries I've been involved with do not want to allow C99 features into the codebase because of MSVC. This, IMO, is the main reason for wanting some C99 in MSVC -- to stop Windows from holding the rest of OSS hostage when it comes to more modern language features.

The stdint.h header comes with MSVC 2010.

Do not beg. Do not use it.

Yes. And do the same with everything you want to use but doesn't have a feature you can't add yourself* --see how far that will get you.

(*) e.g because you don't know how to do so, you're not good enough to program it, the main maintainers don't agree or because the source is closed.

Surely Windows must use tons of C. I wonder what compiler internal devs are using. Archaic support of C seems more like a tactical choice than a laziness one in that case

It's probably all C++ -clean, so they can compile everything as C++.

No, you use whatever you want, in practice C is more used than C++ and C code is compiled with the C compiler, not C++ compiler.

The tools used to build Windows, including compilers, are available as the free Windows Driver Kit (WDK): http://msdn.microsoft.com/en-us/windows/hardware/gg487428

Sometimes the compiler versions match those of the current Visual Studio release, but not always.

Maybe they've switched to using a C++ compiler and C linkage specifications..

Doesn't Microsoft have a disincentive to support it here? It would encourage people to use open-source alternatives to The Microsoft Way.

The same thing could be described as a disincentive for supporting more-recent C++, and yet Microsoft is implementing those features at a pace which is unusually fast (for VC). I'm sure the real issue is much more prosaic: it costs money to implement, and it's not something that most of VC's customers are asking for. The expense here probably isn't in changing the compiler itself; it's in the implementation of the test suites. And while there may be open source test suites available under an acceptable license, the company's willingness to make use of outside source is changing at a glacially-slow pace -- particularly for core products.

Disclaimer: I work at Microsoft and know some people on the C++ compiler team. These comments are purely speculative on my part.

I don't think there's a disincentive on the C++ side - the best Windows apps use C++ so improving C++ helps make Windows a strong platform. The article also points out many of the most missed features of C99 are straightforward, and hopefully the tests would be too.

I suspect that C11 support is closer than C99 support (and it would be good too). Things like variable length arrays are now optional.

There are exactly 2 features which were mandatory in C99 and are optional in C11: variable-length arrays and complex types.

C11 adds some non-optional features whch needs compiler support (alignment specifiers, type-generic selections, unicode literals, ...).

Do you really think vendors which don' care about C99 will suddenly jump to implement all mandatory C11 features?

Since it means not having to support VLAs, yes.

One is not obligated to support all features of a standard in order to support some, a reality Microsoft has consistently availed itself of during its existence.

So by your standards, Microsoft is doing quite well. After all, they do support parts of C99 such as stdint.h.

You really have no idea what my standards are. All I did was say they didn't have to support VLAs to support other parts of C99. That tells you absolutely nothing, stop assuming so much.

I don't think VLAs would be that difficult to support as the underlying functionality (alloca) is already implemented by the compiler.

It's a matter of sanity, not difficulty. VLAs are not well-loved by many of the same people who write the compilers.

Perl Philosphy is simply to torment the implementors on behalf of the user -- Larry Wall

Look at the C99 rationale:

The inability to declare arrays whose size is known only at execution time was often cited as a primary deterrent to using C as a numerical computing language.

That's also the main reason for the introduction of complex types and type-generic math functions.

It's not as if the C committee got together and thought "Hey, what can we do to piss off compiler writers?" - there was actual demand for these features!

Sure, it made language semantics messier (runtime evaluation of sizeof, introduction of variably-modified types restricted to block or prototype scope, magical macros for math functions - which has been fixed with C11, btw), but C99 is indeed a superior language than C90 for doing numerics.

I don't dispute for a moment that some people wanted VLAs or that they're useful in some contexts. On the other hand, your cited field is relatively small, and the use case could have been satisfied with a mechanism much more in line with the existing language.

In fact, we've de-facto had such a mechanism for decades, but it hasn't been codified anywhere, actually for some of the same reasons people don't appreciate VLAs, but it at least rests much more comfortably with the rest of the language.

At some point, you go from enhancing C to breaking it in the interests of a small minority (who apparently don't want to use Fortran?). VLAs are, if not over the line, right on it.

On the other hand [...] the use case could have been satisfied with a mechanism much more in line with the existing language.

In fact, we've de-facto had such a mechanism for decades

Unfortunately, that's not the case. There are actually two primary use cases for VLAs:

1. Variable-length multi-dimensional array parameters:

    // C99
    double trace(size_t n, double mat[][n])
    {
        double sum = 0;
        for(size_t i = 0; i < n; ++i)
            sum += mat[i][i];
        return sum;
    }


    /* C90 */
    double trace(size_t n, double mat[])
    {
        double sum = 0;
        size_t i = 0;
        for(; i < n; ++i)
            sum += mat[i * n + i];
        return sum;
    }

While this may not look like much of an improvement in simple cases, not having to manually emulate array subscription is quite convenient in more complex ones.

2. Allocation of variably-sized objects with automatic storage duration. Some libc implementations provide alloca() for that purpose -- unfortunately, it has issues (see eg http://c-faq.com/malloc/alloca.glb.html ).

What are "for loop initial decorations"?

  for (int i = 0; i < n; i++) { ... }

instead of

  int i;
  for (i = 0; i < n; i++) { ... }

Reply on news.ycombinator.com