GCC Moves Forward With Conversion To C++

[curaga]

Why would it take longer?

You've not compiled both C and C++ lately, have you?

[mirv]

You've not compiled both C and C++ lately, have you?

you haven't made statements that make sense lately, have you?
(sorry, couldn't resist that one)

Seriously though, got anything to back up that statement, or do you believe it just because? If it's just because, you should probably go reading through the idea a bit more.

[TheBlackCat]

You've not compiled both C and C++ lately, have you?

So as far as I can tell, you are saying one of the two things:

1. Compiling the exact same code for takes longer if you are compiling for C++.
2. One or more of the specific changes they plan to make will result in a 20x slowdown in the build

If it is 1, please explain why. If 2, please tell us which of those changes will have the effect and why. Otherwise I am curious how you justify this position. Just saying "C code compiles faster than C++ code" doesn't really mean much if you are comparing different code.

[RealNC]

If you use the STL, the compiles will be quite longer. Also, compiling the same classes in header files again and again just takes much longer then function prototypes.

[gamerk2]

*rolls eyes*

[sarcasm]
Ya because a language determines the quality of the code.
[/sarcasm]

Tell me about it; working on a major ADA program right now. You can write horrid code in any programming language.

[curaga]

you haven't made statements that make sense lately, have you?
(sorry, couldn't resist that one)

Seriously though, got anything to back up that statement, or do you believe it just because? If it's just because, you should probably go reading through the idea a bit more.

It's common knowledge. Even compiling plain C with g++ is slower than with gcc, by 35% here (n=6, hello world). If you actually use any C++ features, the gap grows.

time gcc -o hello hello.c
56ms

time g++ -o hello hello.c
76ms

[curaga]

So as far as I can tell, you are saying one of the two things:

1. Compiling the exact same code for takes longer if you are compiling for C++.
2. One or more of the specific changes they plan to make will result in a 20x slowdown in the build

If it is 1, please explain why. If 2, please tell us which of those changes will have the effect and why. Otherwise I am curious how you justify this position. Just saying "C code compiles faster than C++ code" doesn't really mean much if you are comparing different code.

Both. See my post above about compiling C as C++. Then make a hello world in C++, using C++ idioms, features, and STL and time that.

20x is simply a ballpark figure. But it's a fact that C++ is slow to compile, especially if templates are involved, as mentioned by RealNC.

[mirv]

It's common knowledge. Even compiling plain C with g++ is slower than with gcc, by 35% here (n=6, hello world). If you actually use any C++ features, the gap grows.

time gcc -o hello hello.c
56ms

time g++ -o hello hello.c
76ms

Which is like using glxgears for opengl benchmarking. But just for shits and giggles, I ran that here and got....wildly different results for every run through, sometimes with gcc faster, sometimes with g++ faster. Also, -ftime-report is probably a bit better...

Now if you want to talk about how it's easier to abuse C++, which results in increased compile time, that's entirely different. Let's look at the flip side too: for comparable structures, if not using STL, then something else has to be written, which may be written poorly. Actually, you could theoretically save compile time in some areas due to OOP principles (who knows? that's why gcc-in-c++ is being looked at). You could also make it worse - none of which is a reflection on the language, but rather on the how it's being used.
Now let's also look at C++ vs C in terms of pure semantics which must be observed...in which case, sure, C++ has a bit more in the lexical analysis which must be handled (for example), but such overhead is going to be nowhere near what you're claiming - it's mostly likely going to be completely unnoticed.

[RealNC]

Which is like using glxgears for opengl benchmarking.

Here's building ncurses 5.9 with C and C++. For C++:

./configure CC="g++" CFLAGS="-fpermissive -Wno-write-strings" CXXFLAGS="-fpermissive -Wno-write-strings" --without-cxx-binding
make -j4 # to warm disk caches
make clean
time make -j1

real    0m47.392s
user    0m41.033s
sys     0m4.815s

For C:

make distclean
./configure CFLAGS="-Wno-write-strings" CXXFLAGS="-fpermissive -Wno-write-strings" --without-cxx-binding
make -j4 # to warm disk caches
make clean
time make -j1

real    0m42.323s
user    0m36.326s
sys     0m4.294s

All was done in a tmpfs so it's in RAM. And caches were warm by doing the build twice.

[gamerk2]

It's common knowledge. Even compiling plain C with g++ is slower than with gcc, by 35% here (n=6, hello world). If you actually use any C++ features, the gap grows.

time gcc -o hello hello.c
56ms

time g++ -o hello hello.c
76ms

Question: How are you measuring time? Because I hope you're doing it in such a way where you choice of scheduler isn't affecting the results in some way...