Why no C++ in Windows itself?

It is a long time now that I am using "Dependency Walker for Win32" and
navigating through different Dlls in "System32" folder to find my target API
functions. The thing I have noticed is that, 'almost' all system .dll and
..exe
files are written in C and not C++. If C++ is a powerful OO language and
it's much greater than C for large application development, then why haven't
Microsoft used it in its operating system? Is there a compatibility issue
here that I am not aware of? Or something else?
---
Thanks for comments

re: Why no C++ in Windows itself?

Herr Lucifer wrote:[color=blue]
> It is a long time now that I am using "Dependency Walker for Win32" and
> navigating through different Dlls in "System32" folder to find my target API
> functions. The thing I have noticed is that, 'almost' all system .dll and
> .exe
> files are written in C and not C++. If C++ is a powerful OO language and
> it's much greater than C for large application development, then why haven't
> Microsoft used it in its operating system? Is there a compatibility issue
> here that I am not aware of? Or something else?[/color]

How can you tell that they are written in C? Have you examined the
assembler to look for the presence of C++ startup code (such as the
running of global object constructors)? Remember that C++ code need not
dynamically link with the C++ runtime, and it need not present a C++
interface to the outside world.

Tom

re: Why no C++ in Windows itself?

The GDI system is written in C++, at least.
[color=blue]
> Herr Lucifer wrote:[color=green]
> > It is a long time now that I am using "Dependency Walker for Win32" and
> > navigating through different Dlls in "System32" folder to find my target[/color][/color]
API[color=blue][color=green]
> > functions. The thing I have noticed is that, 'almost' all system .dll[/color][/color]
and[color=blue][color=green]
> > .exe
> > files are written in C and not C++. If C++ is a powerful OO language and
> > it's much greater than C for large application development, then why[/color][/color]
haven't[color=blue][color=green]
> > Microsoft used it in its operating system? Is there a compatibility[/color][/color]
issue[color=blue][color=green]
> > here that I am not aware of? Or something else?[/color]
>
> How can you tell that they are written in C? Have you examined the
> assembler to look for the presence of C++ startup code (such as the
> running of global object constructors)? Remember that C++ code need not
> dynamically link with the C++ runtime, and it need not present a C++
> interface to the outside world.
>
> Tom[/color]

re: Why no C++ in Windows itself?

I don't know what you think of C++, but in its earliest definition, C++ was
called "C with classes". So if there is no class in there, there is no C++
code; it is just pure C....
PS: dependency walker (version 2.1.36668) is able to show functions inside
their classes if their are C++. Take a look if you have VS.NET 2003
-----------------------
"Tom Widmer" <tom_usenet@hotmail.com> wrote in message
[color=blue]
> How can you tell that they are written in C? Have you examined the
> assembler to look for the presence of C++ startup code (such as the
> running of global object constructors)? Remember that C++ code need not
> dynamically link with the C++ runtime, and it need not present a C++
> interface to the outside world.
>
> Tom[/color]

re: Why no C++ in Windows itself?

"Herr Lucifer" <"\n"HerrLucifer\n@microsoft.com> wrote in message
news:elPJIoqDFHA.4016@TK2MSFTNGP10.phx.gbl...
[color=blue]
>I don't know what you think of C++, but in its earliest definition,
>C++ was called "C with classes". So if there is no class in there,
>there is no C++ code; it is just pure C....
> PS: dependency walker (version 2.1.36668) is able to show
> functions inside their classes if their are C++. Take a look if
> you have VS.NET 2003[/color]

If the fundamental Windows API exported classes then surely it would
only be accessible to programs written in C++.

There *is* a version which exports classes - it is called MFC most
of which is a very thin wrapper around the Windows API. [And it
isn't particularly object-oriented.]

Dave
--
David Webber
Author MOZART the music processor for Windows -
http://www.mozart.co.uk
For discussion/support see
http://www.mozart.co.uk/mzusers/mailinglist.htm

re: Why no C++ in Windows itself?

Herr Lucifer" <"\n wrote:[color=blue]
> It is a long time now that I am using "Dependency Walker for Win32"
> and navigating through different Dlls in "System32" folder to find my
> target API functions. The thing I have noticed is that, 'almost' all
> system .dll and .exe
> files are written in C and not C++. If C++ is a powerful OO language
> and it's much greater than C for large application development, then
> why haven't Microsoft used it in its operating system? Is there a
> compatibility issue here that I am not aware of? Or something else?[/color]

A great deal of Windows, especially in the user-mode components, is written
in C++. It simply exposes a pure-C API since that's the only ABI
(Application Binary Interface) defined for Windows. A C++ API would not be
callable by any language other than C++.

The Kernel is written almost entirely in C, with a very few bits of
assembler here and there.

-cd

re: Why no C++ in Windows itself?

Herr Lucifer wrote:[color=blue]
> I don't know what you think of C++, but in its earliest definition, C++ was
> called "C with classes". So if there is no class in there, there is no C++
> code; it is just pure C....
> PS: dependency walker (version 2.1.36668) is able to show functions inside
> their classes if their are C++. Take a look if you have VS.NET 2003[/color]

Only if they are exported though! The *interface* is C (to allow
interoperability with as many other languages as possible, such as C,
C++, Delphi, VB, etc.), but that doesn't tell you anything about the
language with which the dll is implemented. The DLL code could include
thousands of C++ classes, but the interface still be exposed as pure C.

For example, I've got a C++ DLL that I've written containing lots of
classes but it only exposes a C API (since Java JNI only interacts with
specific exported C functions). Dependency Walker doesn't show the
classes at all, and it doesn't tell you what language was used.

IOW, Dependency Walker is a useless tool for determining what language a
module was implemented with.

Tom

re: Why no C++ in Windows itself?

GDI+, not GDI.

"benben" <benhongh@yahoo.com.au> wrote in message
news:uKMrvxpDFHA.4072@TK2MSFTNGP10.phx.gbl...[color=blue]
> The GDI system is written in C++, at least.
>[color=green]
>> Herr Lucifer wrote:[color=darkred]
>> > It is a long time now that I am using "Dependency Walker for Win32" and
>> > navigating through different Dlls in "System32" folder to find my
>> > target[/color][/color]
> API[color=green][color=darkred]
>> > functions. The thing I have noticed is that, 'almost' all system .dll[/color][/color]
> and[color=green][color=darkred]
>> > .exe
>> > files are written in C and not C++. If C++ is a powerful OO language
>> > and
>> > it's much greater than C for large application development, then why[/color][/color]
> haven't[color=green][color=darkred]
>> > Microsoft used it in its operating system? Is there a compatibility[/color][/color]
> issue[color=green][color=darkred]
>> > here that I am not aware of? Or something else?[/color]
>>
>> How can you tell that they are written in C? Have you examined the
>> assembler to look for the presence of C++ startup code (such as the
>> running of global object constructors)? Remember that C++ code need not
>> dynamically link with the C++ runtime, and it need not present a C++
>> interface to the outside world.
>>
>> Tom[/color]
>
>[/color]

re: Why no C++ in Windows itself?

On Wed, 9 Feb 2005 14:07:44 +0330, "Herr Lucifer"
<"\n"HerrLucifer\n@microsoft.com> wrote:
[color=blue]
>It is a long time now that I am using "Dependency Walker for Win32" and
>navigating through different Dlls in "System32" folder to find my target API
>functions. The thing I have noticed is that, 'almost' all system .dll and
>.exe
>files are written in C and not C++. If C++ is a powerful OO language and
>it's much greater than C for large application development, then why haven't
>Microsoft used it in its operating system? Is there a compatibility issue
>here that I am not aware of? Or something else?
>---[/color]

IMHO:

For time-critical issues, smaller code size, better memory
consumption, improved control over efficiency, improved code
generation, simpler interface to assembler, and easier debugging
(especially the kernel and low-level code) the guts of Windows are
mostly written in C.

Additionally, much of the Windows (NT) kernel is based on VMS, which
was written in Bliss32 and VAX assembly, which were much easier to
implement as C.

Microsoft's C++ code (for the shell, directx, COM, etc.) are bizarre
and obfuscated anyway. Were Windows internals written in C++, I would
cry.

--
Sev

re: Why no C++ in Windows itself?

Severian wrote:[color=blue]
> On Wed, 9 Feb 2005 14:07:44 +0330, "Herr Lucifer"
> <"\n"HerrLucifer\n@microsoft.com> wrote:
>
>[color=green]
>>It is a long time now that I am using "Dependency Walker for Win32" and
>>navigating through different Dlls in "System32" folder to find my target API
>>functions. The thing I have noticed is that, 'almost' all system .dll and
>>.exe
>>files are written in C and not C++. If C++ is a powerful OO language and
>>it's much greater than C for large application development, then why haven't
>>Microsoft used it in its operating system? Is there a compatibility issue
>>here that I am not aware of? Or something else?
>>---[/color]
>
>
> IMHO:
>
> For time-critical issues,[/color]

Do you mean that C is faster executing? Or that its performance is more
consistent (for a realtime system)? I don't think either statement is
particularly true, though some C and C++ features should be avoided in
code that should run in a bounded amount of time (e.g. no dynamic
allocation, no disk IO, etc.).

smaller code size, better memory[color=blue]
> consumption,[/color]

C does generally produce slightly smaller code, due to the overhead EH
and RTTI in C++. However, both of those may be avoided if desired.

improved control over efficiency,

How so?

improved code[color=blue]
> generation,[/color]

The code generation is generally worse in C, since I believe that type
based alias analysis isn't as good in C. Is there any other respect in
which you think C can generate better code?

simpler interface to assembler,

What do you mean?

and easier debugging[color=blue]
> (especially the kernel and low-level code)[/color]

But macros are far harder to debug than inline functions!

the guts of Windows are[color=blue]
> mostly written in C.[/color]

That I know nothing about, but it certainly sounds true. But I believe
much of the Linux kernel is now written in C++ (as was BeOS). C++ is
primarily a systems programming language - it was specifically designed
for writing OSes (among other things), hence the priniciple of "You
don't pay for what you don't use."

I'd definitely prefer to write an OS in C++ rather than C, and the
resulting OS would probably be faster and better for it (though possibly
not smaller).

Tom

re: Why no C++ in Windows itself?

On Thu, 10 Feb 2005 18:07:35 +0000, Tom Widmer
<tom_usenet@hotmail.com> wrote:
[color=blue]
>Severian wrote:[color=green]
>> On Wed, 9 Feb 2005 14:07:44 +0330, "Herr Lucifer"
>> <"\n"HerrLucifer\n@microsoft.com> wrote:
>>
>>[color=darkred]
>>>It is a long time now that I am using "Dependency Walker for Win32" and
>>>navigating through different Dlls in "System32" folder to find my target API
>>>functions. The thing I have noticed is that, 'almost' all system .dll and
>>>.exe
>>>files are written in C and not C++. If C++ is a powerful OO language and
>>>it's much greater than C for large application development, then why haven't
>>>Microsoft used it in its operating system? Is there a compatibility issue
>>>here that I am not aware of? Or something else?
>>>---[/color]
>>
>>
>> IMHO:
>>
>> For time-critical issues,[/color]
>
>Do you mean that C is faster executing? Or that its performance is more
>consistent (for a realtime system)? I don't think either statement is
>particularly true, though some C and C++ features should be avoided in
>code that should run in a bounded amount of time (e.g. no dynamic
>allocation, no disk IO, etc.).[/color]

Sorry, I wasn't trying to start a language war. I program in both C
and C++, and have seen plenty of good and bad examples of both. I was
comparing well-written C with well-written C++ (not simply C-like code
compiled by a C++ compiler).

But generally, when using the constructs that make C++ a great
language, it is very much simpler to estimate the overhead of C
statements vs. C++ statements. (I'm not talking about C code compiled
by a C++ compiler!)
[color=blue]
> smaller code size, better memory[color=green]
>> consumption,[/color]
>
>C does generally produce slightly smaller code, due to the overhead EH
>and RTTI in C++. However, both of those may be avoided if desired.[/color]

In my experience, you have to know a lot more about C++ (and how it is
implemented on a particular platform) to make it as efficient (memory
and speed-wise) as well-written C.
[color=blue]
> improved control over efficiency,
>
>How so?[/color]

Because, as examples, overloaded operators and virtual functions
introduce non-obvious overhead that is more explicit in C.
[color=blue]
> improved code[color=green]
>> generation,[/color]
>
>The code generation is generally worse in C, since I believe that type
>based alias analysis isn't as good in C. Is there any other respect in
>which you think C can generate better code?[/color]

You're probably right, in the general case of compiling simple C++
code; and in some cases (i.e., templates) even when generating complex
code. But once you start using the things that make C++ such a
wonderful language, the generated code tends, in my experience, to
become bloated and wasteful -- and much more difficult to debug.
[color=blue]
>simpler interface to assembler,
>
>What do you mean?[/color]

I am speaking of calling assembly from C++ or vice versa, not inline
assembly, which is quite platform-specific.

One example: since assembly language has no concept of C++ objects,
you have to "deobjectify" any data you want to access from assembler
(or C for that matter!), or know a lot about how a particular C++
compiler represents objects in memory and write extra assembly
language to work with them.
[color=blue]
> and easier debugging[color=green]
>> (especially the kernel and low-level code)[/color]
>
>But macros are far harder to debug than inline functions![/color]

Most modern C compilers include inline functions, though by looking at
code generated by MS and GNU C++, macros are often much more
efficient, and thus often more appropriate for OS code.

The difficulty I was referring to was the semi-invisible things that
C++ does, that require you to know the implementation of classes you
use. Overloaded operators (-> and [] come immediately to mind) make
debugging a loathesome task unless you have access to the source code
for the original class. Even in those cases, it's often difficult to
figure out what exactly is going on behind the scenes.

I've seen well-written C++ code that is a joy to work with, but a lot
more of it is obfuscated (unintentionally?) to the point of
hair-pulling. Most of Microsoft's C++ falls into the latter category!

When writing C++, I love STL and wish that MS had not tried to make
their own horrid messes first.
[color=blue]
> the guts of Windows are[color=green]
>> mostly written in C.[/color]
>
>That I know nothing about, but it certainly sounds true. But I believe
>much of the Linux kernel is now written in C++ (as was BeOS). C++ is
>primarily a systems programming language - it was specifically designed
>for writing OSes (among other things), hence the priniciple of "You
>don't pay for what you don't use."[/color]

I agree that well-written C++ code (in a _specific_ rather than
generic way) is very appropriate for operating systems, as long as
provisions are made for using the OS API from other languages.
[color=blue]
>I'd definitely prefer to write an OS in C++ rather than C, and the
>resulting OS would probably be faster and better for it (though possibly
>not smaller).[/color]

I think it boils down to whether the code is well-written or not; the
simpler basis of C makes it easier, in my opinion, to write efficient
and understandable code. That being said, I have worked with and
(hope) I have written efficient and easily maintainable C++ code as
well.

--
Sev

re: Why no C++ in Windows itself?

In article <#yZy58tDFHA.624@TK2MSFTNGP15.phx.gbl>, phil@private-
citizen.com says...

[ top posting fixed ]
[color=blue]
> "benben" <benhongh@yahoo.com.au> wrote in message
> news:uKMrvxpDFHA.4072@TK2MSFTNGP10.phx.gbl...[color=green]
> > The GDI system is written in C++, at least.[/color][/color]
[color=blue]
> GDI+, not GDI.[/color]

GDI doesn't attempt to present an OO interface like GDI+ does, but
GDI is written (primarily) in C++ as well.

Going back to the original question, I think it's worth considering
the simple age of a great deal of the code -- much of the code in
Windows XP goes all the way back to Windows NT 3.1.

Microsoft had only been selling a C++ compiler for about a year when
Windows NT 3.1 started beta testing. Unless we assume that Windows NT
was written from the ground up in less than a year we have to admit
that they really only had a couple of choices: they could use
somebody else's compiler, or they could write in C. Given how much MS
was betting on Windows NT, I can't imagine them depending on anybody
else's technology to build it.

That pretty much ruled out using C++ except for a few of the last
parts to be written.

It's also been pointed out that the exported functions don't
necessarily reflect the language of origin. For a solid example of
this, I'd advise anybody to use dumpbin to look at the functions
exported from ATL.DLL and note that they all look distinctly C-like.
Looking at the source code (which is distributed as part of VC++ and
VS++) will leave little doubt that it really is written in C++.

--
Later,
Jerry.

The universe is a figment of its own imagination.

re: Why no C++ in Windows itself?

I think a point regarding operating systems is being missed. Any operating
system has 3 key compents process management, memory management and I/O.
Thats it. Everything else is application. The OS allocates memory, loads a
process into said memory and begins execution of said process. After that it
switches context between services interrupts, passing instructions to the CPU
and moving bits about memory.

All of this has to be small and preferrable fast. So its all assembly or C,
because C++ requires a fat runtime and excessive memory management. The guts
of windows primarily reside in nt.dll and kernel32.dll, both under 1MB. These
binary contain the Native API in 100% C wrapped around assembly. The
subsystems that sit on top of this, the most well known being the Win32
subsytem, basically wrap around these native api calls, translating the
low-level system objects like disks, partition, and file descriptors into
things we can visualize like drives, directories etc. Could the Win32
subsystem be implemented in C++, of course. But unless something genuinely
need to be implemented in OOP, why bother.

In my experience, and I use both C and C++ regularly, many developers lock
themselves into one contruct...either they hate OOP or they can't do without
it. The result, you get a lot of over-coding from both camps. There are
things you can do in 10 line of C++ that requires hunders of lines of C, and
the inverse is true as well.

Programming languages are tools and like any good mechanic, a programmer has
toolbox too heavy for amatuers to even lift.


"Severian" wrote:
[color=blue]
> On Thu, 10 Feb 2005 18:07:35 +0000, Tom Widmer
> <tom_usenet@hotmail.com> wrote:
>[color=green]
> >Severian wrote:[color=darkred]
> >> On Wed, 9 Feb 2005 14:07:44 +0330, "Herr Lucifer"
> >> <"\n"HerrLucifer\n@microsoft.com> wrote:
> >>
> >>
> >>>It is a long time now that I am using "Dependency Walker for Win32" and
> >>>navigating through different Dlls in "System32" folder to find my target API
> >>>functions. The thing I have noticed is that, 'almost' all system .dll and
> >>>.exe
> >>>files are written in C and not C++. If C++ is a powerful OO language and
> >>>it's much greater than C for large application development, then why haven't
> >>>Microsoft used it in its operating system? Is there a compatibility issue
> >>>here that I am not aware of? Or something else?
> >>>---
> >>
> >>
> >> IMHO:
> >>
> >> For time-critical issues,[/color]
> >
> >Do you mean that C is faster executing? Or that its performance is more
> >consistent (for a realtime system)? I don't think either statement is
> >particularly true, though some C and C++ features should be avoided in
> >code that should run in a bounded amount of time (e.g. no dynamic
> >allocation, no disk IO, etc.).[/color]
>
> Sorry, I wasn't trying to start a language war. I program in both C
> and C++, and have seen plenty of good and bad examples of both. I was
> comparing well-written C with well-written C++ (not simply C-like code
> compiled by a C++ compiler).
>
> But generally, when using the constructs that make C++ a great
> language, it is very much simpler to estimate the overhead of C
> statements vs. C++ statements. (I'm not talking about C code compiled
> by a C++ compiler!)
>[color=green]
> > smaller code size, better memory[color=darkred]
> >> consumption,[/color]
> >
> >C does generally produce slightly smaller code, due to the overhead EH
> >and RTTI in C++. However, both of those may be avoided if desired.[/color]
>
> In my experience, you have to know a lot more about C++ (and how it is
> implemented on a particular platform) to make it as efficient (memory
> and speed-wise) as well-written C.
>[color=green]
> > improved control over efficiency,
> >
> >How so?[/color]
>
> Because, as examples, overloaded operators and virtual functions
> introduce non-obvious overhead that is more explicit in C.
>[color=green]
> > improved code[color=darkred]
> >> generation,[/color]
> >
> >The code generation is generally worse in C, since I believe that type
> >based alias analysis isn't as good in C. Is there any other respect in
> >which you think C can generate better code?[/color]
>
> You're probably right, in the general case of compiling simple C++
> code; and in some cases (i.e., templates) even when generating complex
> code. But once you start using the things that make C++ such a
> wonderful language, the generated code tends, in my experience, to
> become bloated and wasteful -- and much more difficult to debug.
>[color=green]
> >simpler interface to assembler,
> >
> >What do you mean?[/color]
>
> I am speaking of calling assembly from C++ or vice versa, not inline
> assembly, which is quite platform-specific.
>
> One example: since assembly language has no concept of C++ objects,
> you have to "deobjectify" any data you want to access from assembler
> (or C for that matter!), or know a lot about how a particular C++
> compiler represents objects in memory and write extra assembly
> language to work with them.
>[color=green]
> > and easier debugging[color=darkred]
> >> (especially the kernel and low-level code)[/color]
> >
> >But macros are far harder to debug than inline functions![/color]
>
> Most modern C compilers include inline functions, though by looking at
> code generated by MS and GNU C++, macros are often much more
> efficient, and thus often more appropriate for OS code.
>
> The difficulty I was referring to was the semi-invisible things that
> C++ does, that require you to know the implementation of classes you
> use. Overloaded operators (-> and [] come immediately to mind) make
> debugging a loathesome task unless you have access to the source code
> for the original class. Even in those cases, it's often difficult to
> figure out what exactly is going on behind the scenes.
>
> I've seen well-written C++ code that is a joy to work with, but a lot
> more of it is obfuscated (unintentionally?) to the point of
> hair-pulling. Most of Microsoft's C++ falls into the latter category!
>
> When writing C++, I love STL and wish that MS had not tried to make
> their own horrid messes first.
>[color=green]
> > the guts of Windows are[color=darkred]
> >> mostly written in C.[/color]
> >
> >That I know nothing about, but it certainly sounds true. But I believe
> >much of the Linux kernel is now written in C++ (as was BeOS). C++ is
> >primarily a systems programming language - it was specifically designed
> >for writing OSes (among other things), hence the priniciple of "You
> >don't pay for what you don't use."[/color]
>
> I agree that well-written C++ code (in a _specific_ rather than
> generic way) is very appropriate for operating systems, as long as
> provisions are made for using the OS API from other languages.
>[color=green]
> >I'd definitely prefer to write an OS in C++ rather than C, and the
> >resulting OS would probably be faster and better for it (though possibly
> >not smaller).[/color]
>
> I think it boils down to whether the code is well-written or not; the
> simpler basis of C makes it easier, in my opinion, to write efficient
> and understandable code. That being said, I have worked with and
> (hope) I have written efficient and easily maintainable C++ code as
> well.
>
> --
> Sev
>[/color]

re: Why no C++ in Windows itself?

"Rob" <Rob@discussions.microsoft.com> wrote in message
news:449702AA-537F-4218-90AC-98451355017A@microsoft.com...
[color=blue]
> All of this has to be small and preferrable fast. So its all assembly or
> C,
> because C++ requires a fat runtime and excessive memory management.[/color]

A common misperception, but largely untrue. The big exception is C++
IOStreams which IS a huge runtime, but an OS, even one written in C++ would
be unlikely to use IOStreams IMO.
[color=blue]
> The guts
> of windows primarily reside in nt.dll and kernel32.dll, both under 1MB.
> These binary contain the Native API in 100% C wrapped around assembly. The
> subsystems that sit on top of this, the most well known being the Win32
> subsytem, basically wrap around these native api calls, translating the
> low-level system objects like disks, partition, and file descriptors into
> things we can visualize like drives, directories etc. Could the Win32
> subsystem be implemented in C++, of course. But unless something genuinely
> need to be implemented in OOP, why bother.[/color]

Not quite true. ntdll.dll is nothing but a gateway - there's virtually no
actual code in there. The kernel code is in ntoskrnl.exe (2Mb) plus a
number of kernel-mode DLLs and tons of device drivers.

kernel32.dll is the basic Win32 API (Process/Thread API, File/IO API
mostly). These are generally thin wrappers over the "native API" exposed by
ntdll.dll and are written in C (mostly if not entirely).

The windowing portion of the OS is primarily implemented by user32.dll,
gdi32.dll and win32k.sys, which are written in C (mostly if not entirely).

Many of the other Win32 API components are in fact written in C++.

In all of the OS code there is virtually no assembly language other than
some start up code and the lowest level CPU support code in the HAL.

-cd