I have downloaded Vowpal Wabbit ( Version 7.4 ) on to my Mac ( OS X Mavericks ) with the following version of g++ :
Configured with: --prefix=/Applications/Xcode.app/Contents/Developer/usr --with-gxx-include- dir=/Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.9.sdk/usr/include/c++/4.2.1
Apple LLVM version 5.0 (clang-500.2.79) (based on LLVM 3.3svn)
Target: x86_64-apple-darwin13.0.0
Thread model: posix
I modified the Makefile that comes with the installation so that it will pick up Boost ( Version 1.55 ) headers and libraries from the correct folders on my Mac ( I checked by writing and compiling a simple program and everything works ). However, when I run make it cannot find a header file, for example:
In file included from library_example.cc:2:
./../vowpalwabbit/parser.h:15:10: fatal error: 'boost/program_options.hpp' file
not found
Grateful for any help in tracking down the problem.
vowpal wabbit has a dependency on the C++ library+headers boost/program_options.
This dependency must be installed before vowpal wabbit can be built from source.
Since other people may have a similar issue, I will try to answer the more general question:
How to build vowpal wabbit from scratch/source, on Mac OS-X?
The following set of instructions assumes you have no development environment at all. If you do, you may skip any steps you deem unnecessary.
Important: credit where credit is due. These instructions are almost 100% written by Aaron Shumacher: I sourced them from his page following a web-search, with relatively minor modifications, additions, and clarifications.
Stage 0, basic prerequisites needed for developing on Mac OS-X:
Install the Xcode command line tools. This will give you an Apple’s development environment (compiler, make, etc.).
Install the Homebrew package manager.
Once you have Homebrew installed, everything else that's missing, is easy to install.
Install needed tools to build from source:
brew install libtool
brew install automake
brew install boost
brew install git
Now, you're ready to download the vowpal wabbit source tree (latest):
Download:
git clone https://github.com/JohnLangford/vowpal_wabbit.git
cd vowpal_wabbit
You’re now ready to go through the standard build (aka make) process.
Build:
[Edit: 2014-07-20]: ./autogen.sh may not work well in some environments. I recommend trying to build using a simple make first. The included Makefile is already pretty robust and should work just fine in most environments. The problem with running autogen/automake is that it may overwrite the (pretty good) original Makefile with a bad one.
# ./autogen.sh # commented out, use only if 'make' alone doesn't work
make
make test
Note: it is important to verify (using make test) that your binary works as expected. If you see any failure, please report it on the mailing list, and proceed with the installation only if the failure is immaterial to you (e.g. only some rarely used options/tests you don't care about fail). See the test/RunTests script for full details on the tests.
Install:
make install
At this point, you should have vowpal wabbit at /usr/local/bin/vw
Related
I am trying to build from source for RedHawk installation and trying to replace uhd3.5.3 with a higher version of uhd for USRP_UHD module. I saw USRP_UHD source code, but found nowhere for uhd3.5.3 source code. I am guessing it might simply use "yum install ..." to download and install uhd host code and suporting libraries. Does anybody know where this code is located in redhawk_src_2.0 package? Or is there an efficient way in linux to search for this among all the files in redhawk_src_2.0 package?
Thanks in advance!
UHD is the USRP Harware driver softwware from Ettus Research. The UHD is distributed from:
http://files.ettus.com/binaries/uhd_stable/
Version 3.5.3 is here:
http://files.ettus.com/binaries/uhd_stable/uhd_003.005.003-release/
The current release is 3.9.2:
http://files.ettus.com/binaries/uhd_stable/uhd_003.009.002-release/
Hopefully this helps. I wasn't 100% sure what you are asking for.
First, clone the uhd library:
git clone git://github.com/EttusResearch/uhd.git
Then checkout the 3.9.2 tag:
git checkout release_003_009_002
Using the instructions from Ettus (http://files.ettus.com/manual/page_build_guide.html), install the dependencies listed under the Fedora section:
sudo yum -y install boost-devel libusb1-devel python-mako doxygen python-docutils cmake make gcc gcc-c++
Next, generate the Makefiles with CMake:
cd <uhd-repo-path>/host
mkdir build
cd build
cmake -DCMAKE_INSTALL_PREFIX=/opt/uhd ../
I recommend using an alternate install prefix at first to preserve the 3.5.3 version, just in case, but if you don't want to, just run:
cmake ../
instead of the longer command with the -DCMAKE_INSTALL_PREFIX flag. (Note: I couldn't get cmake to find the installed boost version, so you may have to do some other research on that since it is probably a whole different stackoverflow post in and of itself)
Now build and install the uhd library:
make
make test
sudo make install
Now that the library is installed, you should edit the USRP_UHD/cpp/configure.ac so that the PKG_CONFIG_PATH points to your install prefix (either the one specified in the flag above, or if you did the default, just leave it, as it should already be pointing to /usr/local/lib/pkgconfig). Also in the configure.ac file, make sure to change the uhd version number to 3.9.2.
Now you should be able to rebuild the USRP_UHD Device:
./reconf && ./configure && make install
I don't guarantee that the Device will build against the new version of the uhd library, that will depend on what (if anything) has been deprecated an removed from the library between the supported version and the newest one.
Finally, if you get through all of those steps successfully and if you specified a non-standard install prefix, you will have to add this prefix to either the LD_LIBRARY_PATH or an /etc/ld.so.conf.d/ script in order for the Device to correctly execute.
Clarification - you can follow the Redhawk manual appendix B for building Redhawk from source AFTER installing UHD v3.9.3 from source as per pwolframs instructions. The UHD v3.5.3 is not packaged with the Redhawk source code, only the Redhawk RPM, disregard all those RPMs when building UHD and Redhawk from source.
I'm on Mac OS X Yosemite 10.10.2. As far as I know, wx-widgets were preinstalled on older versions on OS X (probably that's why google search doesn't help), but that's not the case anymore. I need to get started on a C++ project using wx-widgets, using any IDE (Eclipse, Codelite, or whatever is best). However, I simply cannot get wx-widgets to work. After a workaround I was able to compile on my machine wxWidgets-3.0.2. I called from terminal make and make install, but the libraries seem to not be correctly installed. Indeed, if I create a project on Codelite I get many "wx-config: Command not found" errors. Can someone help me to get me started on installing the libraries and creating a simple project that includes wx-widgets?
I use homebrew which is installed by copying and pasting one line from the homebrew website.
Then, the installation of wxWidgets is simplicity itself with:
brew install wxmac
Any problems, you just call the doctor with
brew doctor
Now you have a proper package manager you can install ImageMagick, GNU Parallel, GNU awk, Oxygen, webkit2html, zeromq, tree, SDL, Redis, sqlitebrowser, pandoc, exiftool, ffmpeg, 7zip, awscli, etc...
You can then compile your wxWidgets programs with
g++ `wx-config --cxxflags` -o sample sample.cpp `wx-config --libs`
I can successfully compile and install wxWidgets on Yosemite with these commands:
- Get the latest sources of wxWidgets 3 from wxwidgets.org and unpack them.
- Move the unpacked directory someplace where you want to keep it.
- Open a terminal and change into the wxwidgets directory.
- Create two directories: build-release and build-debug (don't rename those!)
- Change into wxwidgets/build-release
- Run
../configure --with-osx_cocoa --disable-shared --with-opengl --enable-universal-binary=i386,x86_64 --with-macosx-sdk=/Developer/SDKs/MacOSX10.6.sdk --with-macosx-version-min=10.6
- Run make, then sudo make install
- Change into wxwidgets/build-debug
- Run
../configure --enable-debug --with-osx_cocoa --with-opengl --enable-universal-binary=i386,x86_64 --with-macosx-sdk=/Developer/SDKs/MacOSX10.6.sdk --with-macosx-version-min=10.6
- Run make, then sudo make install
Be aware that these instructions are for compiling against the Mac OS X 10.6 SDK, which I have at /Developer/SDKs/MacOSX10.6.sdk. You can try to omit the --with-macosx-sdk parameter altogether, which will compile using the latest SDK. Also ensure that you have Xcode and the command line developer tools installed.
These commands will install static libraries for the release build and shared libraries for the debug build. You can change this by supplying or omitting the --disable-shared parameter.
The instructions for building and installing wxWidgets are included in the distribution and can also be viewed online.
If your IDE doesn't find wx-config, you must not have the location where you installed it (/usr/local/bin) by default in your PATH. Add it there to fix this.
Like suggested by VZ, you must change your PATH.
But not the variable PATH in your environment Windows but the variable PATH in your C++ project. It's strange but maybe it's different.
Go into (for eclipse) :
C/C++Build->Environment->PATH (add the path of wx-config in the beginning).
Press : Apply, OK and build the project.
I think is strange that Eclipse don't update the PATH after rebooting the OS. Maybe we must add an option to Eclipse for building WxWidget and force Eclipse to update the PATH, but I don't know it.
This change work for me.
Eclipse / MinGW-MSYS / WxWidget 2.8
I installed Boost with homebrew(brew install boost) on my Mac running Lion with the purpose of extending python with an existing C++ program I have. Now I can't follow the starting guide because 1 - I don't have 'bjam' installed in my system and 2-I don't see the examples folder anywhere. My question is: how I am supposed to do the the c++-python linking with this homebrew installation without 'bjam'? is there some alternative method I should use or something I am missing here?
The right formula to install boost for linking c++ with python programs is:
brew install boost-build
as pointed out by senderle in the comments to my question. This installs bjam automatically.
For Windows, you go into $(BoostDir)\tools\build\v2\engine and run build.bat which automatically builds bjam (into the bin.ntx86 directory on windows). There is a build.sh file there too, but I've never used a Mac so I don't know if that will work for you. Otherwise, just do a Google search for a precompiled bjam executable for Mac OS X.
It should just be a case of downloading Boost then running "Bootstrap" followed by "Bjam".
Bootstrap should build you a Bjam executable in the Boost folder.
I'm used to manually install GCC from source before on Ubuntu and it was a painful process. So I really don't want to do repeat this process. Currently, I have MinGW and GCC (4.6.2) installed on my machine. So is there an easy way to update GCC without building it entirely from source? Has anyone done it before can share me some tips?
Update using terminal:
Run the command prompt/terminal (cmd or sh).
Update the package list:
mingw-get update
After updating the package list, run:
mingw-get upgrade
When the command finishes running, all of your packages will be upgraded.
Update using the GUI version:
If you aren't used to the terminal, there is also a GUI version of MinGW called "MinGW Installation Manager", which is normally located at:
C:\MinGW\libexec\mingw-get\guimain.exe
When the GUI is open, tap Installation -> Update Catalogue. This will update the package list.
After that, tap Installation -> Mark All Upgrades. This will select all of the packages which can be upgraded.
Finally, tap Installation -> Apply Changes to apply the upgrades.
Snapshots and release builds of the MinGW http://code.google.com/p/mingw-builds/downloads/list
UPDATE: As of Nov-5-2012, MinGW includes 4.7.2 of the GCC compiler, which is the most current version. I personally will use the MinGW package, since it includes MSYS and other tools, and since it is the defacto standard, I hope that it is better supported. The only drawback is the included GDB does not include Python scripting, but I think that can be updated separately.
I don't know if you can update the GCC in MinGW, but there are alternative projects to MinGW with newer versions of GCC.
The MinGW-builds project provides a package similar to MinGW but with an updated GCC (4.7.2). I extracted the files to c:\MinGW (so I could use the same path) after moving my existing MinGW to another folder. Only MinGW has MSYS, so if you need that (I did) copy the msys folder from the original MinGW.
After installing MinGW-builds, the result of g++ --version:
g++ (Built by MinGW-builds project) 4.7.2
The latest GCC that comes with standard MinGW is currently 4.7.0. I compiled a few simple projects in Code::Blocks, and the latest wxWidgets source, and everything seems to work.
It's worth noting that MinGW-builds includes a Python script enabled version of GDB, in case you want STL pretty printer support, or other GDB Python features.
Another good choice is TDM-GCC, which is a project that hosts an impressive installer that includes an updated GCC compiler (4.7.1) and support tools. It is a standalone package. As far as I can tell, Python scripting is not supported in the GDB that comes with this package.
Finally there is the MinGW-w64 project, which is a fork of MinGW. Rubenvb, who posted an answer to this question, has good standalone packages (in the personal builds folder) based on this project which include an updated GCC (4.7.2). The included GDB seems to be Python script enabled.
if you use MSYS2
just open mingw64-console and type in:
# Update the package database and core system packages with:
pacman -Syu
# If needed, close MSYS2, run it again from Start menu. Update the rest with:
pacman -Su
I have built a more up to date GCC 4.7 (it's built within code freeze, so it should be equivalent with the release). Also, this build contains <thread> support.
32-bit
64-bit
The -gcc_linux package is what you want if you want to use it on Linux. Just extract somewhere and add the mingw??/bin directory to PATH.
PS: There's a native Windows compiler and a Windows Clang build. I suggest using Clang only with one of my GCC 4.6 builds, as it has trouble with GCC 4.7's libstd++. Extract the Clang package in the same directory as the GCC package and it will work out of the box.
The MinGW-w64 build from http://winlibs.com/ has the latest GCC version and requires no installation, just unzip the download. To upgrade you can just replace the mingw32 or mingw64 folder with the new version.
To get latest MinGW64 in Windows: https://sourceforge.net/projects/mingw-w64/?source=typ_redirect
For all platforms: https://mingw-w64.org/doku.php/download
You just need to update your g++ compiler if its not.
First problem can be solved through this:
Update using the GUI version:
If you aren't used to the terminal, there is also a GUI version of MinGW called "MinGW Installation Manager", which is normally located at:
C:\MinGW\libexec\mingw-get\guimain.exe
When the GUI is open, tap
Installation -> Update Catalogue
. This will update the package list.
After that, tap
Installation -> Mark All Upgrades
. This will select all of the packages which can be upgraded.
Finally, tap
Installation -> Apply Changes
to apply the upgrades.
2nd Problem could be that if you didn't updated the path of mingw in environment variables.
I want to compile C++ code on MacOS X, using the g++ compiler. How do I install it?
That's the compiler that comes with Apple's XCode tools package. They've hacked on it a little, but basically it's just g++.
You can download XCode for free (well, mostly, you do have to sign up to become an ADC member, but that's free too) here: http://developer.apple.com/technology/xcode.html
Edit 2013-01-25: This answer was correct in 2010. It needs an update.
While XCode tools still has a command-line C++ compiler, In recent versions of OS X (I think 10.7 and later) have switched to clang/llvm (mostly because Apple wants all the benefits of Open Source without having to contribute back and clang is BSD licensed). Secondly, I think all you have to do to install XCode is to download it from the App store. I'm pretty sure it's free there.
So, in order to get g++ you'll have to use something like homebrew (seemingly the current way to install Open Source software on the Mac (though homebrew has a lot of caveats surrounding installing gcc using it)), fink (basically Debian's apt system for OS X/Darwin), or MacPorts (Basically, OpenBSDs ports system for OS X/Darwin) to get it.
Fink definitely has the right packages. On 2016-12-26, it had gcc 5 and gcc 6 packages.
I'm less familiar with how MacPorts works, though some initial cursory investigation indicates they have the relevant packages as well.
Installing XCode requires:
Enrolling on the Apple website (not fun)
Downloading a 4.7G installer
To install g++ *WITHOUT* having to download the MASSIVE 4.7G xCode install, try this package:
https://github.com/kennethreitz/osx-gcc-installer
The DMG files linked on that page are ~270M and much quicker to install. This was perfect for me, getting homebrew up and running with a minimum of hassle.
The github project itself is basically a script that repackages just the critical chunks of xCode for distribution. In order to run that script and build the DMG files, you'd need to already have an XCode install, which would kind of defeat the point, so the pre-built DMG files are hosted on the project page.
Type g++(or make) on terminal.
This will prompt for you to install the developer tools, if they are missing.
Also the size will be very less when compared to xcode
Download Xcode, which is free with an ADC online membership (also free):
http://developer.apple.com/technology/xcode.html
xcode is now available for free from the app store. Just "buy it" (for free) and it will download. To get the command line tools go into preferences/downloads and "install command line compiler tools".
Instead of gcc you are using clang, but it works the same.
Here is how to do it on the newer mac chips and how to switch from clang(default) to g++
Install g++ through home-brew.
Check out the version you just installed, probably 12th or higher
You can make a symbolic link from g++-12 to g++
In order to do it, just type in your terminal:
sudo ln -s $(which g++-12) /usr/local/bin/g++.
Now open a new terminal and check your version again and you should see g++ instead of clang
g++ --version