Compile GCC Branch

Dae · 08-31-2009, 02:34 AM

How would I compile the GCC lambda branch? All I've got is the svn checkout. There's makefile's and such but I can't get any of them to run. I don't know anything about linux, or much about command-line compiling.

**CornedBee** · 08-31-2009, 08:51 AM

The lambda branch is compiled exactly like any other GCC source checkout. If there's a configure script, run

Code:

./configure --help

to see the options, then run configure again with the proper options, preferably from a different directory, e.g:

Code:

svn co <gcc-url> gcc-lambda-src
mkdir gcc-lambda-obj
cd gcc-lambda-obj
../gcc-lambda-src/configure <options>
make
sudo make install

If you compile an experimental GCC branch, I recommend two things for configure:
1) Disable GCJ and its libjava. It's huge and takes forever to compile.
2) Give the GCC binaries a prefix or suffix. This allows you to put them into the PATH without conflicts with your system GCC.

If there's no configure script, you need to run autoconf in the source directory, but I think GCC policy commits the generated files.

To get all build dependencies for GCC, if you have Debian/Ubuntu, you can run something like

Code:

sudo apt-get builddep gcc

Dae · 08-31-2009, 08:28 PM

Thanks for that CornedBee. sudo apt-get build-dep gcc was helpful. I was going through manually finding the dependencies.

I keep running into this error though, after about 20 minutes of re-compiling.

Quote:

cp: cannot stat mm_malloc.h: No such file or directory
chmod: cannot access include/mm_malloc.h: No such file or directory

It's really annoying.

That's using default configure: ../srcdir/configure. I also tried it with --disable-libgcj as you suggested.

I can't find much help about mm_malloc on Google.

**CornedBee** · 09-01-2009, 01:50 AM

mm_malloc is a malloc with an additional alignment parameter, used for allocating buffers that will be used with vectorized code. It's part of the compiler suport library, which is why the build script wants to copy it around.
The file is apparently called gmm_intrin.h in the source tree (in $SRCDIR/gcc/config/i386) and the build system should rename it and copy it to the target. I have no idea why this fails. The build system is just as big a disaster as the rest of GCC.

Dae · 09-01-2009, 05:26 AM

I actually got it working

Thanks for the help CornedBee, I appreciate it

My final command list was as such.

Quote:

cd /home/daemn/
sudo apt-get build-dep gcc
svn co svn://gcc.gnu.org/svn/gcc/branches/cxx0x-lambdas-branch/ /home/daemn/srcdir/
mk objdir
cd objdir
../srcdir/configure --program-prefix=lambda --disable-libgcj --enable-languages=c++ --disable-werror
make
sudo make install

I had to do a make distclean and it took a few hours, but phew.. I love lambda's. I also love properties - hope those make it into c++. Anything that helps with ease and usability (such as auto). The implementation of C++0x sure is slow going though.. but lots of good stuff in there. I'm a lot more stoked than before, having read a lot more of the docs and proposals waiting for this to build.

I'll probably run into some roadblock. Just wait.

**CornedBee** · 09-01-2009, 06:51 AM

How far along is the lambda branch, by the way? Is it actually usable?

Dae · 09-01-2009, 07:04 AM

Quote:

Originally Posted by CornedBee

How far along is the lambda branch, by the way? Is it actually usable?

I'll let you know once I finish converting my project over to Linux. See if it survives Asio server/client and some other stuff.

Dae · 09-04-2009, 08:52 AM

Phew, that was intense. I ran into a whole series of issues. First of all, there seems to be a bug in GCC 4.4 and 4.5 (which the lambda-branch uses) in rvalue references (&&). So I disabled rvalue ref macro in Boost, and undefined the c++0x macro in the STD library (which brought forth an #error definition, which I casually deleted

This obviously means I cannot use the new stuff like std::thread and std::chrono. That's fine, because I tried them and they didn't seem complete (ie. this_thread::sleep which is now sleep_for apparently and is macro blocked, and not using std::seconds using std::chrono, etc). Then I ran into issues with libmysqld 5.1, switching to 5.4 didn't fix it, 4.1 worked but sucked, and after some tweaking 5.4 magically works.

Rvalue bug: http://www.mail-archive.com/gcc-bugs...msg264019.html

The end result? I've deleted my PoS "callback" macro (which only worked with vc9), and have 5000 lines of lambda, roughly 40 lambda definitions, working with thousands of sockets per second (multiple asynchronous servers) and not a problem in sight yet. Works great!

This is pretty much exactly what I wanted (for now), your standard c/c++ (pre-0x), boost, and auto / lambda. I'll expand my use of lambdas / variadic templates / initializers etc. later, and the rest when we get them. I'll be back if I run into any problems ^^ you can bet on it.

**CornedBee** · 09-04-2009, 09:17 AM

Fun! Maybe I should implement lambdas for Clang now.

Dae · 09-05-2009, 10:28 AM

Haha, it shouldn't be too hard if you allow local structs as template arguments, then just change the syntax (which is surprising since I don't believe GCC allows local structs as template arguments yet).

Edit: actually recursion is working, it just has to be passed by reference not value.. obviously. my bad.

Code:

std::function<void()> hello = [&hello]() { hello(); };

I'm expanding my network code so we'll see how it'll handle some added complexity.

Dae · 09-06-2009, 02:33 AM

I didn't realize how restrictive that known bug from the official page would be:

Quote:

Although they can appear within templates, lambda functions may not be used with dependent types.

So at the most I'll only be using them for usage code. I guess that's good, but I can wish library code was easier to read/understand/navigate for everyone (especially colleagues). Oh well.

Rvalue references seem to be working for me, just not for parts of the STD/boost library (maybe an incompatibility issue - boost 1.4 only shows testing up to gcc 3.3 and some 3.4).

08-31-2009, 02:34 AM	#1
Dae Deprecated Join Date: Oct 2004 Location: Canada Posts: 1,032	Compile GCC Branch How would I compile the GCC lambda branch? All I've got is the svn checkout. There's makefile's and such but I can't get any of them to run. I don't know anything about linux, or much about command-line compiling. __________________ Warning: Have doubt in anything I post. GCC 4.5, Boost 1.40, Code::Blocks 8.02, Ubuntu 9.10 010001000110000101100101
Dae is offline

08-31-2009, 08:51 AM	#2
CornedBee Cat without Hat Join Date: Apr 2003 Posts: 8,492	The lambda branch is compiled exactly like any other GCC source checkout. If there's a configure script, run Code: ./configure --help to see the options, then run configure again with the proper options, preferably from a different directory, e.g: Code: svn co <gcc-url> gcc-lambda-src mkdir gcc-lambda-obj cd gcc-lambda-obj ../gcc-lambda-src/configure <options> make sudo make install If you compile an experimental GCC branch, I recommend two things for configure: 1) Disable GCJ and its libjava. It's huge and takes forever to compile. 2) Give the GCC binaries a prefix or suffix. This allows you to put them into the PATH without conflicts with your system GCC. If there's no configure script, you need to run autoconf in the source directory, but I think GCC policy commits the generated files. To get all build dependencies for GCC, if you have Debian/Ubuntu, you can run something like Code: sudo apt-get builddep gcc __________________ All the buzzt! CornedBee "There is not now, nor has there ever been, nor will there ever be, any programming language in which it is the least bit difficult to write bad code." - Flon's Law
CornedBee is offline

09-01-2009, 01:50 AM	#4
CornedBee Cat without Hat Join Date: Apr 2003 Posts: 8,492	mm_malloc is a malloc with an additional alignment parameter, used for allocating buffers that will be used with vectorized code. It's part of the compiler suport library, which is why the build script wants to copy it around. The file is apparently called gmm_intrin.h in the source tree (in $SRCDIR/gcc/config/i386) and the build system should rename it and copy it to the target. I have no idea why this fails. The build system is just as big a disaster as the rest of GCC. __________________ All the buzzt! CornedBee "There is not now, nor has there ever been, nor will there ever be, any programming language in which it is the least bit difficult to write bad code." - Flon's Law
CornedBee is offline

09-01-2009, 06:51 AM	#6
CornedBee Cat without Hat Join Date: Apr 2003 Posts: 8,492	How far along is the lambda branch, by the way? Is it actually usable? __________________ All the buzzt! CornedBee "There is not now, nor has there ever been, nor will there ever be, any programming language in which it is the least bit difficult to write bad code." - Flon's Law
CornedBee is offline

09-04-2009, 08:52 AM	#8
Dae Deprecated Join Date: Oct 2004 Location: Canada Posts: 1,032	Phew, that was intense. I ran into a whole series of issues. First of all, there seems to be a bug in GCC 4.4 and 4.5 (which the lambda-branch uses) in rvalue references (&&). So I disabled rvalue ref macro in Boost, and undefined the c++0x macro in the STD library (which brought forth an #error definition, which I casually deleted This obviously means I cannot use the new stuff like std::thread and std::chrono. That's fine, because I tried them and they didn't seem complete (ie. this_thread::sleep which is now sleep_for apparently and is macro blocked, and not using std::seconds using std::chrono, etc). Then I ran into issues with libmysqld 5.1, switching to 5.4 didn't fix it, 4.1 worked but sucked, and after some tweaking 5.4 magically works. Rvalue bug: http://www.mail-archive.com/gcc-bugs...msg264019.html The end result? I've deleted my PoS "callback" macro (which only worked with vc9), and have 5000 lines of lambda, roughly 40 lambda definitions, working with thousands of sockets per second (multiple asynchronous servers) and not a problem in sight yet. Works great! This is pretty much exactly what I wanted (for now), your standard c/c++ (pre-0x), boost, and auto / lambda. I'll expand my use of lambdas / variadic templates / initializers etc. later, and the rest when we get them. I'll be back if I run into any problems ^^ you can bet on it. __________________ Warning: Have doubt in anything I post. GCC 4.5, Boost 1.40, Code::Blocks 8.02, Ubuntu 9.10 010001000110000101100101 Last edited by Dae; 09-04-2009 at 09:00 AM.
Dae is offline

09-04-2009, 09:17 AM	#9
CornedBee Cat without Hat Join Date: Apr 2003 Posts: 8,492	Fun! Maybe I should implement lambdas for Clang now. __________________ All the buzzt! CornedBee "There is not now, nor has there ever been, nor will there ever be, any programming language in which it is the least bit difficult to write bad code." - Flon's Law
CornedBee is offline

09-05-2009, 10:28 AM	#10
Dae Deprecated Join Date: Oct 2004 Location: Canada Posts: 1,032	Haha, it shouldn't be too hard if you allow local structs as template arguments, then just change the syntax (which is surprising since I don't believe GCC allows local structs as template arguments yet). Edit: actually recursion is working, it just has to be passed by reference not value.. obviously. my bad. Code: std::function<void()> hello = [&hello]() { hello(); }; I'm expanding my network code so we'll see how it'll handle some added complexity. __________________ Warning: Have doubt in anything I post. GCC 4.5, Boost 1.40, Code::Blocks 8.02, Ubuntu 9.10 010001000110000101100101 Last edited by Dae; 09-05-2009 at 07:53 PM.
Dae is offline

Thread Tools
Show Printable Version Email this Page
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
gcc compiler for windows?	td4nos	C Programming	8	07-19-2009 02:28 PM
gcc asm code syntax	Uberapa	C Programming	4	06-15-2007 01:16 AM
gcc not for C++?	tin	C++ Programming	4	09-15-2004 08:26 AM
GCC bool undefined	Stack Overflow	Linux Programming	3	07-06-2004 12:54 PM
gcc compile on cmd line, then run, get: bash: hello: command not found	cjtotheg	Linux Programming	4	11-14-2001 10:28 AM