Greetings everyone...
As you might know: I'm the tech freak guy who compiled the GlibC and some Linux applications for HTC Desire.
As I have published my findings, I've recieved messages about how I'm compiling the applications given the source code. I try to help about this, but sadly I think half of what I say is not understandable without a proper guide. Ergo, this thread will suply this!
When?
Well, soon; because I'm still preparing it.
What will it include?
It'll be a basic guide about how to compile applications with your toolchain (Here is the guide about how to make a toolchain of yourself: How to make a toolchain). I'm planning to show it with GlibC compilation.
What's the extend of this help?
Frankly, compiling from the source sometimes can be quite messy and most of the time, a compilation/configuration line is never same with different applications. However, since most sources come with either Autotools or with a Makefile, most of the times, all you need is to change a few command line variable with a proper value.
I'm not planning to show you the code changes or "patchworks" they might be necessary, because this is quite application specific thing and I'm not an expert about this either. I'm just going to show you a guide about the compilation process.
What's the catch / scum of this help?
There is no catch in it. I just want to help curious nerdy guys like me )), so that we can learn from each other; or maybe improve each others work! I learned a lot from guys here, and I want to help to this community as much as I can as well.
Compilation Guide from Source code
Well, let's begin
We're going to compile the GlibC, with all internals to make it ready. After you prepare this, you might also check DoomLord's post here (http://forum.xda-developers.com/showthread.php?t=1041064) to make a recovery zip to flash this to your device. Make yourself comfortable now; because I'm going to explain everything we use here, thus this might be loong guide
1- Obtain the Sources
First, obtain the sources that you're going to compile. Check all the dependencies of the package and download them (and compile them) before. In our case, since we'll compile GlibC and all it needs is a working toolchain; we don't need to worry about this.
You can obtain GlibC source code from http://ftp.gnu.org/gnu/glibc/ address. Note that, since we're building GlibC for ARM devices and since GlibC seperated ARM support to "Ports" package, we need to download that as well. Make sure you download the same versions of Glibc and Glibc-Ports packages.
{
"lightbox_close": "Close",
"lightbox_next": "Next",
"lightbox_previous": "Previous",
"lightbox_error": "The requested content cannot be loaded. Please try again later.",
"lightbox_start_slideshow": "Start slideshow",
"lightbox_stop_slideshow": "Stop slideshow",
"lightbox_full_screen": "Full screen",
"lightbox_thumbnails": "Thumbnails",
"lightbox_download": "Download",
"lightbox_share": "Share",
"lightbox_zoom": "Zoom",
"lightbox_new_window": "New window",
"lightbox_toggle_sidebar": "Toggle sidebar"
}
2- Extract the Sources, and make a build directory
Extract the sources to some folder; say /home/<user>/Desktop/glibc . Also extract the glibc-ports package to some direcorty, say /home/<user>/Desktop/glibc-ports.
After this, rename the folder "glibc-ports" to just "ports" and then move it inside to the "glibc" folder. If you don't do this step, you'll have "ARM is unsupported" error in configuration step.
Since glibc cannot be compiled from the directory which sources are in it (a restriction made by the developers, to make sources unchanged and ready to be compiled again), we need to make another folder from which we're going to call the compilation tools. Let's say this directory is named as "glibc-build", also at /home/<user>/Desktop .
3- Start "configure"
Now, open a terminal emulator window ( you can use Ctrl+Alt+T keys under Ubuntu to easily open one ). Change into the glibc-build directory.
Code:
cd /home/<user>/Desktop/glibc-build
After this operation, we've to call the glibc's auto-configure script from this folder. Note that ".." is a "special folder name" which denotes the upper level directory (so basicly, we're changing into the upper level in directory tree - Desktop in our case - and then call something from a folder inside that)
Code:
../glibc-2.14/configure --help
When you run the above command, configure will give you the options you can use to configure this package for your needs. Nearly 90% of the time, checking the output of this command gives you the options you need to set to compile properly.
Code:
`configure' configures GNU C Library (see version.h) to adapt to many kinds of systems.
Usage: ../glibc-2.14/configure [OPTION]... [VAR=VALUE]...
To assign environment variables (e.g., CC, CFLAGS...), specify them as
VAR=VALUE. See below for descriptions of some of the useful variables.
Defaults for the options are specified in brackets.
Configuration:
-h, --help display this help and exit
--help=short display options specific to this package
--help=recursive display the short help of all the included packages
-V, --version display version information and exit
-q, --quiet, --silent do not print `checking ...' messages
--cache-file=FILE cache test results in FILE [disabled]
-C, --config-cache alias for `--cache-file=config.cache'
-n, --no-create do not create output files
--srcdir=DIR find the sources in DIR [configure dir or `..']
Installation directories:
--prefix=PREFIX install architecture-independent files in PREFIX
[/usr/local]
--exec-prefix=EPREFIX install architecture-dependent files in EPREFIX
[PREFIX]
By default, `make install' will install all the files in
`/usr/local/bin', `/usr/local/lib' etc. You can specify
an installation prefix other than `/usr/local' using `--prefix',
for instance `--prefix=$HOME'.
For better control, use the options below.
Fine tuning of the installation directories:
--bindir=DIR user executables [EPREFIX/bin]
--sbindir=DIR system admin executables [EPREFIX/sbin]
--libexecdir=DIR program executables [EPREFIX/libexec]
--sysconfdir=DIR read-only single-machine data [PREFIX/etc]
--sharedstatedir=DIR modifiable architecture-independent data [PREFIX/com]
--localstatedir=DIR modifiable single-machine data [PREFIX/var]
--libdir=DIR object code libraries [EPREFIX/lib]
--includedir=DIR C header files [PREFIX/include]
--oldincludedir=DIR C header files for non-gcc [/usr/include]
--datarootdir=DIR read-only arch.-independent data root [PREFIX/share]
--datadir=DIR read-only architecture-independent data [DATAROOTDIR]
--infodir=DIR info documentation [DATAROOTDIR/info]
--localedir=DIR locale-dependent data [DATAROOTDIR/locale]
--mandir=DIR man documentation [DATAROOTDIR/man]
--docdir=DIR documentation root [DATAROOTDIR/doc/c-library]
--htmldir=DIR html documentation [DOCDIR]
--dvidir=DIR dvi documentation [DOCDIR]
--pdfdir=DIR pdf documentation [DOCDIR]
--psdir=DIR ps documentation [DOCDIR]
System types:
--build=BUILD configure for building on BUILD [guessed]
--host=HOST cross-compile to build programs to run on HOST [BUILD]
Optional Features:
--disable-option-checking ignore unrecognized --enable/--with options
--disable-FEATURE do not include FEATURE (same as --enable-FEATURE=no)
--enable-FEATURE[=ARG] include FEATURE [ARG=yes]
--disable-sanity-checks really do not use threads (should not be used except
in special situations) [default=yes]
--enable-check-abi do "make check-abi" in "make check" (no/warn/yes)
[default=no]
--enable-shared build shared library [default=yes if GNU ld & ELF]
--enable-profile build profiled library [default=no]
--enable-omitfp build undebuggable optimized library [default=no]
--enable-bounded build with runtime bounds checking [default=no]
--disable-versioning do not include versioning information in the library
objects [default=yes if supported]
--enable-oldest-abi=ABI configure the oldest ABI supported [e.g. 2.2]
[default=glibc default]
--enable-stackguard-randomization
initialize __stack_chk_guard canary with a random
number at program start
--enable-add-ons[=DIRS...]
configure and build add-ons in DIR1,DIR2,... search
for add-ons if no parameter given
--disable-hidden-plt do not hide internal function calls to avoid PLT
--enable-bind-now disable lazy relocations in DSOs
--enable-static-nss build static NSS modules [default=no]
--disable-force-install don't force installation of files from this package,
even if they are older than the installed files
--enable-kernel=VERSION compile for compatibility with kernel not older than
VERSION
--enable-all-warnings enable all useful warnings gcc can issue
--enable-multi-arch enable single DSO with optimizations for multiple
architectures
--enable-experimental-malloc
enable experimental malloc features
--enable-nss-crypt enable libcrypt to use nss
Optional Packages:
--with-PACKAGE[=ARG] use PACKAGE [ARG=yes]
--without-PACKAGE do not use PACKAGE (same as --with-PACKAGE=no)
--with-gd=DIR find libgd include dir and library with prefix DIR
--with-gd-include=DIR find libgd include files in DIR
--with-gd-lib=DIR find libgd library files in DIR
--with-fp if using floating-point hardware [default=yes]
--with-binutils=PATH specify location of binutils (as and ld)
--with-elf if using the ELF object format
--with-selinux if building with SELinux support
--with-xcoff if using the XCOFF object format
--without-cvs if CVS should not be used
--with-headers=PATH location of system headers to use (for example
/usr/src/linux/include) [default=compiler default]
--with-tls enable support for TLS
--without-__thread do not use TLS features even when supporting them
--with-cpu=CPU select code for CPU variant
Some influential environment variables:
CC C compiler command
CFLAGS C compiler flags
LDFLAGS linker flags, e.g. -L<lib dir> if you have libraries in a
nonstandard directory <lib dir>
LIBS libraries to pass to the linker, e.g. -l<library>
CPPFLAGS (Objective) C/C++ preprocessor flags, e.g. -I<include dir> if
you have headers in a nonstandard directory <include dir>
CPP C preprocessor
CXX C++ compiler command
CXXFLAGS C++ compiler flags
Use these variables to override the choices made by `configure' or to help
it to find libraries and programs with nonstandard names/locations.
Report bugs to <glibc>.
GNU C Library home page: <http://www.gnu.org/software/c-library/>.
General help using GNU software: <http://www.gnu.org/gethelp/>.
The help is quite straightforward, and all explanations are already done. The most important options for our case here is "prefix" and "host". Also, for compilation of GlibC, we need to give "with-headers", "with-tls" and "with-ports" proper values.
"Prefix" is a very important option. This options specifies where the installed binaries and libraries "think" they are. Some binaries and libs. are quite flexible: they don't care about where they are, thus the value of prefix isn't much of importance, but most of the time, it is. Omitting a prefix value here will make PREFIX value of "/usr/local" - which is quite dangerous because it may screw your own system (/usr/local is usually where your own PC applications are - most probably you cannot screw this folder if you're not a super-user but, well, anyway...); so always change this value to something proper when you're cross compiling.
Also, make sure that this path exists in your target device. If you set a prefix, say /cache, and then copy the stuff at, say, /data; it's highly probable that your applications will search for it's config files at /cache, and will complain that they are not found - since they're at /data. Since AFAIK, there is no way to change this in most application after compilation of software, you have to recompile the program to fix such errors (or move the application to the prefix it's set).
I usually use /data as prefix, because I install the stuff to there.
Host is another very important option. If you don't give this parameter, auto-configure will think you're compiling this application for the PC you're using. We've to give "host" the machine specification we're compiling the app for. For ARM based devices, this is mostly something line "arm-xxx-linux" or "arm-linux" or "arm-xxxx-linux-gnueabi" or such.
How do you found this out? Well, simple: If you followed my crostool-ng tutorial and used the same options as I did; your host name will be in "arm-xxxxxx-linux-gnueabi" format. "xxxxx" here is what your vendor name is. If you set a vendor value when making the toolchain, you must write it here. If you left it empty, xxxxx will be "unknown".
The other way to check this is your toolchain's name of gcc command: If your ARM GCC toolchain name is "arm-myvendor-linux-gnueabi-gcc" then, your host name is "arm-myvendor-linux-gnueabi".
Most of the times, there are other options that you must give. As for an example, here is what I generally use to compile glibC:
Code:
../glibc-2.14/configure --host=arm-msm-linux-gnueabi --prefix=/data --with-tls --with-ports="nptl, ports"
I use host as "arm-msm-linux-gnueabi" because this is my cross compile target name.
We set "with-tls" option here: it's for GlibC to support Thread Local Storage. This is a programming scheme that allows threads to have their own storage area; and we give this configuration here to give GlibC a support for this. More info about TLS can be found online in programming wiki's.
We also set "with-ports" because we're also installing some extra plugins for our GlibC install. NPTL is an advanced threading library that supports many advanced features like TLS and such. There is also a "linuxthreads" but it's older. New versions of GlibC is not shipped with "linuxthreads" either. The other port, "ports" is for ARM support.
You must also give the kernel source directory here with a parameter of "with-headers" if you're using some other toolchain and your kernel sources are in some other place. Thanks to crostool-ng that it moves all the headers to the toolchain folder and makes them automatically reachable, so you can omit this parameter.
Well, when you give the command and supply the options and press ENTER, it shall start configuring the package:
When there is a missing dependency or option, it shall generally give you a warning at this step and terminate the configuration. You must then find the solution of this error (install the necessary dependency etc.) and then try to configure the package again - as I said, this is usually not the case for GlibC .
3- Start "make"
After configuration process is done without errors, you can pass to the compilation phase. If you did configuration properly, you don't have to use special parameters and such for most packages. Just issue the command "make" and wait
Code:
make
Once this is completed without errors, you're nearly ready for the shipment Note that, if you get any errors in this process, you should definitely google it; since errors in this level is quite application and build environment (your machines configs. etc.) specific. The errors in this phase is usually needs some tricks to fix, and they are usually not so easy to fix
3- Start "make install"
Once the compilation is successful, you can install the application to the "PREFIX" folder you specified whilst configuring the package. In order to do that, just issue:
Code:
make install
However, issuing just this command isn't what we want generally: Since we've cross compiled the application, we're not interested in installing that to our PC! We want it to be somewhere we can easily access, so that we can distribute easily right?
For most packages, three important variables are necessary to make this. Which variable is heeded is highly dependent of package, however it's DESTDIR usually used.
DESTDIR: The variable usually used in makefiles. When you set DESTDIR while giving "make install", the application becomes installed to DESTDIR/PREFIX folder, instead of just PREFIX folder without changing the prefix info inside the application so it makes it easier for us to distribute the package.
install_root : GlibC uses this variable, instead of DESTDIR. I didn't see any other packages using this.
INSTALL_TOP: OpenSSL and Lua packages use this variable instead of DESTDIR.
Which of these variable is used cannot be identified either without examining the Makefile, or without testing it
How are you going to give it a value? Simple: you just assign the variable a value, after giving "make install" command:
Code:
make install DESTDIR=/home/<user>/myapp
For GlibC, this becomes:
Code:
make install install_root=/home/<user>/glibc
Which makes glibc installed to /home/<user>/glibc folder.
You can take the files from this folder and make a flashable zip; and send it to your device!
5- NOTES
- Be careful about the PREFIX value, you should install the files to this place in your device as well.
- Most of the times, you don't need the "include" folder - this folder keeps the header files for your packages, so that other applications may be compiled with them. Since we're not compiling applications in our device, we don't need headers. Compiled applications don't use headers anymore.
- Most of the times, .a files under the "lib" folder is unnecessary - these are static libraries that are usually used when applications are being compiled. Since we don't compile stuff at our device, we don't need them - they are usually quite big too! Be careful though: some packages don't offer .so files (which are dynamic libraries for applications to use dynamically) - it might be necessary to keep .a files then.
- .la files are needed for application compilation with libraries, you can erase them as well.
- pkgconfig can be erased in distribution packages. PKGConfig is a tool for compilation that automatically parses the files in pkgconfig dirs to give necessary compilation parameters with dynamic libraries to the applications at compile time. Like we don't need headers, we don't need those files.
Well, that's it. I hope I could be of some help. See you next time and happy Android'ing!
< Reserved for other stuff for guide, like hints.. >
Maybe also include a precompiled toolchain for download?
Can do that, but first need to compile a "static" toolchain for this. Actually, this seems like a good idea
Let me work on this next
Nice...big thx...i love such guide´s
with kind regards...Alex
theGanymedes said:
Can do that, but first need to compile a "static" toolchain for this. Actually, this seems like a good idea
Let me work on this next
Click to expand...
Click to collapse
Tried that: The output is huge, and even though the toolchain works without any need to any library, it still needs the libraries to be compiled in order to compile the other binaries - makes sense, since you need the libraries to compile against them anyway
So isn't it possible to package your toolchain which contains gcc, initial glibc and libraries, so that we can download and use them right away? It would save a lot of time and would be a far better option than the ndk.
Sent from my HTC Desire using Tapatalk
Droidzone said:
So isn't it possible to package your toolchain which contains gcc, initial glibc and libraries, so that we can download and use them right away? It would save a lot of time and would be a far better option than the ndk.
Sent from my HTC Desire using Tapatalk
Click to expand...
Click to collapse
That's possible, but only if you're also using Ubuntu 10.10 or 10.04 and a 64-bit machine
For some reason - guess library version differences - toolchains done in 11.10 was not working at 10.04; so I had to redo it, for instance. This technique you say might or might not work - there is no guarantee in it.
Ah well.. I'm using a 32 bit Ubuntu 11.10 on a 64 bit machine. Anyway I have the toolchain I compiled on this one with your help..But if you remember I took a couple of days to get it done, even when I had your excellent help..Newcomers might find it difficult I guess. But if it werent for your toolchain compilation guide in the glibc thread, it'd have been next to impossible!
Droidzone said:
Ah well.. I'm using a 32 bit Ubuntu 11.10 on a 64 bit machine. Anyway I have the toolchain I compiled on this one with your help..But if you remember I took a couple of days to get it done, even when I had your excellent help..Newcomers might find it difficult I guess. But if it werent for your toolchain compilation guide in the glibc thread, it'd have been next to impossible!
Click to expand...
Click to collapse
Thanks Actually, it took me 3 days to make a bootstrap GCC and 8 days to make a Stage 1 GCC when I first began this I could never compile GlibC with those, because of the messy patches needed
I'm never claiming the process to be easy; but believe me: making a toolchain is the most irritating part. Most of the other programs do compile and run just fine, once their dependencies met. Only "very massive projects" like Xorg gives headaches time to time: because normal users do never try to compile those stuff from the source; and because of that, the developers don't try to make the process error-free.
Still, "grep" and knowledge of C makes wonders time to time
Droidzone said:
Maybe also include a precompiled toolchain for download?
Click to expand...
Click to collapse
I have a precompiled toolchain for android, it is in alpha stage with work still ongoing, and needs 2Go of storage. See: http://forum.xda-developers.com/showthread.php?p=43256170#post43256170
Hi Folks
I wasn't sure where this should belong but as it is a bit of an Hack this forum is probably the most appropriate
Introduction
This short tutorial will show you how to patch the Android Build System to allow you to cross-compile Android AOSP host tools ( adb, fastboot etc ) for OSX using a linux based machine. This is something Google said was impossible or at the very least unsupported.
Assumptions
You have a linux based machine and working copy of the AOSP source tree.
You can/have successfully compile(d) a full Android Release from this tree.
A basic idea of how the Android Build System works is beneficial.
Getting Started
I've set-up a git repository which contains a binary copy of the OSX SDK 10.6 and the apple-darwin10-gcc cross compiler. So first things first. open a terminal and set the root of the AOSP sources tree to the current directory.
STAGE 1: Copy the OSX SDK
Step 1.
Clone the repo with the SDK and toolchain
Code:
git clone https://github.com/trevd/android_platform_build2.git build2
Step 2.
Create /Developer directory at your filesystem root, this is a known location for the SDKs
Code:
sudo mkdir /Developer
sudo chown $USER.$USER /Developer
Step 3.
Copy and unpack the SDK package
Code:
cp build2/osxsdks10.6.tar.gz /Developer
cd /Developer
tar -zxvf osxsdks10.6.tar.gz
rm osxsdks10.6.tar.gz
cd - # back to aosp root
STAGE 2 : Swapping the Toolchain
This is where the fun begins :laugh:
The Android Build system has the majority of the infrastructure in place already to build for OSX, the only problem is that you need OSX to build for OSX. However we can remedy that with a couple of dirty hacks :laugh:.
The prebuilts/gcc/darwin-x86 directory contains a toolchain compatible with osx ( mach-o binaries ). We are going to swap this for a linux compatible ( elf ) executables.
Step 4:
Copy and unpack the elf compatible darwin cross toolchain
Code:
cp build2/i686-apple-darwin-4.2.1.tar.gz prebuilts/gcc/linux-x86/host
cd prebuilts/gcc/linux-x86/host
tar -zxvf i686-apple-darwin-4.2.1.tar.gz
cd - # back to aosp root
Step 5:
Remove the mach-o binaries and symlink the elf binaries in it's place
Code:
cd prebuilts/gcc
rm -rf darwin-x86
ln -s linux-x86 darwin-x86
cd - # back to aosp root
Step 6:
We also need to replace the mach-o version of the ccache executable which live in the prebuilt/misc directory
Code:
cd prebuilts/misc
rm -rf darwin-x86
ln -s linux-x86 darwin-x86
cd - # back to aosp root
STAGE 3: Patching the build system .mk files
We need to patch a couple of files in the build directory namely the build/core/combo/HOST_darwin-x86.mk the main crux of this is swapping the ar tool for libtool so static libraries can be created without error.
Code:
patch -p1 < build2/build.patch
If the patch has been applied successfully you should see the following
Code:
patching file system/core/adb/Android.mk
patching file build/core/combo/HOST_darwin-x86.mk
patching file build/core/combo/select.mk
patching file build/core/envsetup.mk
patching file build/envsetup.sh
You are now ready to cross compile!! :good: ..... well not quite, but nearly.... here's why!
The Android Build System will attempt to build both the Target and Host files for most modules so I'd advise using a lunch option which already has a full target built for it or alternatively you can build the generic sdk using the following commands at the AOSP source tree root.
Code:
. build/envsetup.sh
lunch sdk-eng
make sdk
This will stop target dependency errors occurring when you build individual modules.
NOW we're ready to cross compile.
STAGE 4: Building Modules
At present module build is very much a piecemeal process. To build adb for example we need to build the dependencies first. This is not too onerous as most host modules have very few dependencies.
Building adb
adb has dependencies on the following libraries
Code:
external/zlib
external/openssl
system/core/liblog
system/core/libcutils
system/core/libzipfile
I've found the easiest way to compile the dependencies is to navigate to each directory in turn an use to "mm" build system command to compile the individual module. the commands I run to compile adb are as follows.
From AOSP Source Root
Code:
cd external/zlib
USE_DARWIN=true mm -j8
cd ../openssl
USE_DARWIN=true mm -j8
croot # go back to the AOSP root
cd system/core/liblog
USE_DARWIN=true mm -j8
cd ../libcutils
USE_DARWIN=true mm -j8
cd ../libzipfile/
USE_DARWIN=true mm -j8
cd ../adb
USE_DARWIN=true mm -j8
All being well you should now have and adb binary located at out/host/darwin-x86/bin/adb. running the file command on this binary should produce the following output
Code:
adb: Mach-O executable i386
Conclusion
Although this method is a little rough and ready, it should produce the desired results if you need to cross compile for OSX. The eventual goal would be to compile a full OSX Android SDK on linux in a similar manner to the way the windows-sdk is currently compiled. This requires more investigation as compiling the windows sdk on linux employs a little bit of trickery on the part of the build system.
Final Notes and FAQs:
Why can't I just type make <module> from the root?
Doing this triggers building of additional modules such as LLVM and clang which are to deployed out/host/darwin-x86/bin the build system then attempts to use binary later on. These are obviously built for the Mach-o architecture and as such are incompatible with the linux. This results in a build error which can and would be resolved by the above mentioned trickery ( see conclusion )
I use OSX binaries (along with Windows and my native Linux) in one of my projects. Thanks, I have always relied on finding compiled binaries elsewhere. Lack of an OSX aapt held up an update at one point.
One of those things that you don't really use until you need it, but I will try to remember to give it a shot. I don't have any doubt that it works.
mateorod said:
I use OSX binaries (along with Windows and my native Linux) in one of my projects. Thanks, I have always relied on finding compiled binaries elsewhere. Lack of an OSX aapt held up an update at one point.
One of those things that you don't really use until you need it, but I will try to remember to give it a shot. I don't have any doubt that it works.
Click to expand...
Click to collapse
Thanks. Yes this really is an edge case. Hopefully It will help some folks out.
Regarding aapt in particular.... It's perfectly possible to build aapt, however, we need to do some slight of hand with the clang and clang++ executables as libpng on which aapt depends uses these 2 binaries as part of it's build process.
Here's the build list and the clang trick if you want to try it some time.
Code:
build/libs/host
external/expat
external/zlib
system/core/liblog
system/core/libcutils
mkdir out/host/darwin-x86/bin
cp out/host/linux-x86/bin/clang out/host/darwin-x86/bin
cp out/host/linux-x86/bin/clang++ out/host/darwin-x86/bin
external/libpng
frameworks/base/libs/androidfw
frameworks/native/libs/utils
frameworks/base/tools/aapt
I started off with a clean out/host/darwin-x86 directory so I didn't miss any dependencies.
like I mentioned the clang "swap out" is something the make win_sdk option does automatically so with it a little more research I should be able to get the mac build to do the same but you'll have to "fill yer boots" with the ghetto method for now
For reference here's a link to the sdk building instructions http://tools.android.com/build which describes how to cross compile the windows sdk on linux ( in case anyone was wondering what the hell i'm on about)
My use case has come up
I will be cross-compiling for OSX today...specifically with aapt in mind. I will report back, but I fully expect it to work as described.
mateorod said:
My use case has come up
I will be cross-compiling for OSX today...specifically with aapt in mind. I will report back, but I fully expect it to work as described.
Click to expand...
Click to collapse
Cheers Man!
Hopefully no bitrot has crept in since april and now. I know I've changed my OS version since to Lubuntu 13.04, not like the OS version really matters any.
mateorod said:
but I fully expect it to work as described.
Click to expand...
Click to collapse
Then you Sir, are either Drunk or a Fool! LOL Keep expectations Quantum and only decided when the result is observed a'la Schrodinger Cat
okay...So I was trying to compile SlimRom (so as to get an OSX aapt binary with the SlimRom changes) and things did not necessarily go as planned. There were enough changes to the SlimRom android_build that your build/build.patch does not apply cleanly. I spent some time and tried to modify the patch so that it would work for both SlimRom, AOSP and probably others, but each android_build repo has some differences in surrounding the HOST_AR line, so commenting that just was not portable between flavors. Not cool.
Anyway, turns out that this method does not quite work out of the box for non-AOSP versions (not that you claimed that it did). I got some unfamiliar errors related to (I believe) some OSX toolchains. But in both times I tried this, I actually had to pretty immediately swap out of that flavor and so I was unable to do much debugging. (I keep all the flavors I build {CM, AOKP, SlimRom, PAC, PA, OpenPDroid, etc, etc, etc} all layered in one android/system/jellybean directory. It saves a ton of space, but only allows me to do one thing at a time.)
So the only feedback I have is nothing...I even formatted my hard drive in-between and forgot to put up a paste, so the errors are currently lost to history.
Things that I noticed, for better or worse
You recommend putting the SDKs in the root dir. I believe the documentation is recommending the Developer be placed in home (as per the SDK/ADT bundle docs).
You might want a
Code:
mv android_platform_build2 build2
line. I normally wouldn't bother, but it looks like you are trying to post a line-by-line guide.
I would put the recommendation that a full build be available to the out folder (or a built generic sdk) right at the top, since it is a preliminary step. I had to revert my handwritten changes, then build, then reapply the changes and rebuild since I thought it needed a clean out dir.
Did you have any trouble with git reverting the toolchain swap? On two separate machines, I had to go so far as to delete .repo/projects/prebuilts/gcc/* and prebuilts/gcc/darwin-x86/arm/arm-eabi-4.6. It kept complaining of that the project in the .repo folder was a bad match. No amount of git trickery (which I am not terrible at) let me back out more easily.
I am willing to try again...but I have some other small things to attend to first. It is an admirable hack you have here sir. I will return to it soon and report back once more.
mateorod said:
okay...So I was trying to compile SlimRom (so as to get an OSX aapt binary with the SlimRom changes) and things did not necessarily go as planned. There were enough changes to the SlimRom android_build that your build/build.patch does not apply cleanly. I spent some time and tried to modify the patch so that it would work for both SlimRom, AOSP and probably others, but each android_build repo has some differences in surrounding the HOST_AR line, so commenting that just was not portable between flavors. Not cool.
Anyway, turns out that this method does not quite work out of the box for non-AOSP versions (not that you claimed that it did). I got some unfamiliar errors related to (I believe) some OSX toolchains. But in both times I tried this, I actually had to pretty immediately swap out of that flavor and so I was unable to do much debugging. (I keep all the flavors I build {CM, AOKP, SlimRom, PAC, PA, OpenPDroid, etc, etc, etc} all layered in one android/system/jellybean directory. It saves a ton of space, but only allows me to do one thing at a time.)
So the only feedback I have is nothing...I even formatted my hard drive in-between and forgot to put up a paste, so the errors are currently lost to history.
Things that I noticed, for better or worse
You recommend putting the SDKs in the root dir. I believe the documentation is recommending the Developer be placed in home (as per the SDK/ADT bundle docs).
You might want a
Code:
mv android_platform_build2 build2
line. I normally wouldn't bother, but it looks like you are trying to post a line-by-line guide.
I would put the recommendation that a full build be available to the out folder (or a built generic sdk) right at the top, since it is a preliminary step. I had to revert my handwritten changes, then build, then reapply the changes and rebuild since I thought it needed a clean out dir.
Did you have any trouble with git reverting the toolchain swap? On two separate machines, I had to go so far as to delete .repo/projects/prebuilts/gcc/* and prebuilts/gcc/darwin-x86/arm/arm-eabi-4.6. It kept complaining of that the project in the .repo folder was a bad match. No amount of git trickery (which I am not terrible at) let me back out more easily.
I am willing to try again...but I have some other small things to attend to first. It is an admirable hack you have here sir. I will return to it soon and report back once more.
Click to expand...
Click to collapse
Hi
Thanks for this, It sounds like you've suffered an exercise in frustration there. I wasn't aware that "SlimRom" had a different aapt ( just out of general ignorance and not having paid any attention )
SDK - My Tree last time I used this was /Developer directory in the root - I think It comes from what the toolchain is expecting, I just gave it what it wants
mv android_platform_build2 build2 - Yep I did mean that it's the git clone line which wants changing
Code:
git clone https://github.com/trevd/android_platform_build2.git build2
SDK Recommendation - I shall move that to the top, even though it is already in there, It should probably be highlighted better and possible it's own "Stage"
Reverting the toolchain - Ahh , It appears I work slightly different from most in this respect. I have a general mistrust of SCM's ( I lost too much code on too many different SCM's, Probably through my own inability to use them correctly but ) what I do to revert to change is
Code:
cd prebuilts/gcc/darwin-x86/host/
rm -rf i686-apple-darwin-4.2.1
repo sync i686-apple-darwin-4.2.1
You can do this "trick" on any project in the source tree it's only on rare occasions where I screwed up badly that I have to delete anything in .repo/projects but I also have my distro in their own individual directories with there own full git trees, which is a massive waste of space and has a ton of redundancy due to the AOSP repositories being mirror by every single one but switching between them is a lot easier
If SlimRom's changes are localized to aapt, I'd be more inclined to drop it into the AOSP build and try that... If you have a link to slimrom's frameworks/base repository I'll grab it and try it myself.
On a final note there's a "full version" of the HOST_darwin make file in the build2/core/combo directory the changes to envsetup.mk and select.mk are minimal and can easily be applied manually. You don't need to patch the adb makefile if your not building it.
Again Thanks for the feedback
[SIZE="+1"]What is a Cross-Compiler?[/SIZE]
A cross compiler is a compiler capable of creating executable code for a platform other than the one on which the compiler is running. Cross compiler tools are used to generate executables for embedded system or multiple platforms. It is used to compile for a platform upon which it is not feasible to do the compiling, like microcontrollers that don't support an operating system. From wikipedia
Click to expand...
Click to collapse
[SIZE="+1"]How is that connected with an Android?[/SIZE]
In order to create a native C/C++ binary for an Android, you must firstly compile the source code. Usually you can't do so on an Android itself due to lack of proper tools and environment, or hardware barriers, especially amount of RAM. This is why you should learn how to cross-compile, to create a binary on your PC, that your ARM-based Android will understand.
[SIZE="+1"]Why do I need it?[/SIZE]
You need to learn cross-compiling technique if you want to run native C/C++ programs on an Android. Actually, if you've already built your own custom ROM from AOSP sources (i.e. CyanogenMod), then you used cross-compiling tools and methods even without noticing .
Building an AOSP ROM is fairly easy, there's one command like brunch, which does the job. However, what if you want to compile a custom, not natively included binary? This is the purpose of this tutorial.
[SIZE="+1"]What I will learn from this guide?[/SIZE]
How to properly set C/C++ building environment
How to build a native C/C++ application for Android device
How to optimize native binaries for my device
[SIZE="+1"]Step 1 - The Beginning[/SIZE]
You should start from installing any Linux-based OS, I highly suggest trying a Debian-based distro (such as Ubuntu), or even Debian itself, as this tutorial is based on it .
In general, I highly suggest to compile an AOSP ROM (such as CyanogenMod) for your device firstly. This will help you to get familiar with cross-compiling on Android. I also suggest to compile one or two programs from source for your Linux, but if you're brave enough to learn cross-compiling without doing any of these, you can skip those suggestions .
[SIZE="+1"]Step 2 - Setting up[/SIZE]
Firstly you should make sure that you have all required compile tools already.
[email protected]:~# apt-get update && apt-get install checkinstall
Click to expand...
Click to collapse
This should do the trick and install all required components.
I suggest creating a new folder and navigating to it, just to avoid a mess, but you can organize everything as you wish.
Start from downloading NDK from here.
The NDK is a toolset that allows you to implement parts of your app using native-code languages such as C and C++.
Click to expand...
Click to collapse
[email protected]:~# wget http://dl.google.com/android/ndk/android-ndk-r9d-linux-x86_64.tar.bz2
[email protected]:~# tar xvf android-ndk-r9d-linux-x86_64.tar.bz2
[email protected]:~# mv android-ndk-r9d ndk
Click to expand...
Click to collapse
Now you should make a standalone toolchain, navigate to root of your ndk (this is important) and then build your toolchain:
[email protected]:~# cd ndk/
[email protected]:~/ndk# build/tools/make-standalone-toolchain.sh --toolchain=arm-linux-androideabi-4.8 --platform=android-18 --install-dir=/root/ndkTC
Copying prebuilt binaries...
Copying sysroot headers and libraries...
Copying libstdc++ headers and libraries...
Copying files to: /root/ndkTC
Cleaning up...
Done.
Click to expand...
Click to collapse
You should edit bolded variables to your preferences. Toolchain is the version of GCC you want to use, 4.8 is currently the newest one, in the future it may be 4.9 and so on. Platform is a target API for your programs, this is important only for android-specific commands, such as logging to logcat. When compiling a native Linux program, this won't matter (but it's a good idea to set it properly, just in case). Install dir specifies destination of your toolchain, make sure that it's other than ndk (as you can see I have ndk in /root/ndk and toolchain in /root/ndkTC).
Now you need to download my exclusive cc.sh script from here and make it executable.
[email protected]:~# wget https://dl.dropboxusercontent.com/u/23869279/Files/cc.sh
[email protected]:~# chmod 755 cc.sh
Click to expand...
Click to collapse
This script is a very handy tool written by me to make your life easier while cross-compiling. Before running it make sure to edit "BASIC" options, especially NDK paths. Apart from that it's a good idea to take a look at DEVICEFLAGS and setting them properly for your device, or clearing for generic build. You don't need to touch other ones unless you want/need them.
Just for a reference, I'll include currently editable options:
#############
### BASIC ###
#############
# Root of NDK, the one which contains $NDK/ndk-build binary
NDK="/root/ndk"
# Root of NDK toolchain, the one used in --install-dir from $NDK/build/tools/make-standalone-toolchain.sh. Make sure it contains $NDKTC/bin directory with $CROSS_COMPILE binaries
NDKTC="/root/ndkTC"
# Optional, may help NDK in some cases, should be equal to GCC version of the toolchain specified above
export NDK_TOOLCHAIN_VERSION=4.8
# This flag turns on ADVANCED section below, you should use "0" if you want easy compiling for generic targets, or "1" if you want to get best optimized results for specific targets
# In general it's strongly suggested to leave it turned on, but if you're using makefiles, which already specify optimization level and everything else, then of course you may want to turn it off
ADVANCED="1"
################
### ADVANCED ###
################
# Device CFLAGS, these should be taken from TARGET_GLOBAL_CFLAGS property of BoardCommonConfig.mk of your device, eventually leave them empty for generic non-device-optimized build
# Please notice that -march flag comes from TARGET_ARCH_VARIANT
DEVICECFLAGS="-march=armv7-a -mtune=cortex-a9 -mfpu=neon -mfloat-abi=softfp"
# This specifies optimization level used during compilation. Usually it's a good idea to keep it on "-O2" for best results, but you may want to experiment with "-Os", "-O3" or "-Ofast"
OLEVEL="-O2"
# This specifies extra optimization flags, which are not selected by any of optimization levels chosen above
# Please notice that they're pretty EXPERIMENTAL, and if you get any compilation errors, the first step is experimenting with them or disabling them completely, you may also want to try different O level
OPTICFLAGS="-s -flto=8 -ffunction-sections -fdata-sections -fvisibility=hidden -funswitch-loops -frename-registers -frerun-cse-after-loop -fomit-frame-pointer -fgcse-after-reload -fgcse-sm -fgcse-las -fweb -ftracer -fstrict-aliasing"
# This specifies extra linker optimizations. Same as above, in case of problems this is second step for finding out the culprit
LDFLAGS="-Wl,-O1 -Wl,--as-needed -Wl,--relax -Wl,--sort-common -Wl,--gc-sections"
# This specifies additional sections to strip, for extra savings on size
STRIPFLAGS="-s -R .note -R .comment -R .gnu.version -R .gnu.version_r"
# Additional definitions, which may help some binaries to work with android
DEFFLAGS="-DNDEBUG -D__ANDROID__"
##############
### EXPERT ###
##############
# This specifies host (target) for makefiles. In some rare scenarios you may also try "--host=arm-linux-androideabi"
# In general you shouldn't change that, as you're compiling binaries for low-level ARM-EABI and not Android itself
CONFIGANDROID="--host=arm-linux-eabi"
# This specifies the CROSS_COMPILE variable, again, in some rare scenarios you may also try "arm-eabi-"
# But beware, NDK doesn't even offer anything apart from arm-linux-androideabi one, however custom toolchains such as Linaro offer arm-eabi as well
CROSS_COMPILE="arm-linux-androideabi-"
# This specifies if we should also override our native toolchain in the PATH in addition to overriding makefile commands such as CC
# You should NOT enable it, unless your makefile calls "gcc" instead of "$CC" and you want to point "gcc" (and similar) to NDKTC
# However, in such case, you should either fix makefile yourself or not use it at all
# You've been warned, this is not a good idea
TCOVERRIDE="0"
# Workaround for some broken compilers with malloc problems (undefined reference to rpl_malloc and similar errors during compiling), don't uncomment unless you need it
#export ac_cv_func_malloc_0_nonnull=yes
Click to expand...
Click to collapse
As you can notice, my magic script already contains bunch of optimizations, especially device-based optimizations, which are the most important. Now it's the time to run our script, but in current shell and not a new one.
[email protected]:~# source cc.sh
Done setting your environment
CFLAGS: -O2 -march=armv7-a -mtune=cortex-a9 -mfpu=neon -mfloat-abi=softfp -s -flto=8 -ffunction-sections -fdata-sections -fvisibility=hidden -funswitch-loops -frename-registers -frerun-cse-after-loop -fomit-frame-pointer -fgcse-after-reload -fgcse-sm -fgcse-las -fweb -ftracer -fstrict-aliasing -DNDEBUG -D__ANDROID__
LDFLAGS: -Wl,-O1 -Wl,--as-needed -Wl,--relax -Wl,--sort-common -Wl,--gc-sections
CC points to arm-linux-androideabi-gcc and this points to /root/ndkTC/bin/arm-linux-androideabi-gcc
Use "$CC" command for calling gcc and "$CCC" command for calling our optimized CC
Use "$CXX" command for calling g++ and "$CCXX" for calling our optimized CXX
Use "$STRIP" command for calling strip and "$SSTRIP" command for calling our optimized STRIP
Example: "$CCC myprogram.c -o mybinary && $SSTRIP mybinary "
When using makefiles with configure options, always use "./configure $CONFIGANDROID" instead of using "./configure" itself
Please notice that makefiles may, or may not, borrow our CFLAGS and LFLAGS, so I suggest to double-check them and eventually append them to makefile itself
Pro tip: Makefiles with configure options always borrow CC, CFLAGS and LDFLAGS, so if you're using ./configure, probably you don't need to do anything else
Click to expand...
Click to collapse
Command "source cc.sh" executes cc.sh and "shares" the environment, which means that any exports will be exported to our current shell, and this is what we want. It acts the same as AOSP's ". build/envsetup.sh", so you can also use . instead of source.
As you can see above, my script should let you know if it properly set everything, especially if $CC points to our ndkTC. It also set a generic "$CCC" and "$CCXX" commands, which are optimized versions of standard $CC. $CC points to our cross-compiler, $CCC points to our cross-compiler and also includes our optimization flags.
[email protected]:~# echo $CC
arm-linux-androideabi-gcc
[email protected]:~# echo $CCC
arm-linux-androideabi-gcc -O2 -march=armv7-a -mtune=cortex-a9 -mfpu=neon -mfloat-abi=softfp -s -flto=8 -ffunction-sections -fdata-sections -fvisibility=hidden -funswitch-loops -frename-registers -frerun-cse-after-loop -fomit-frame-pointer -fgcse-after-reload -fgcse-sm -fgcse-las -fweb -ftracer -fstrict-aliasing -DNDEBUG -D__ANDROID__ -Wl,-O1 -Wl,--as-needed -Wl,--relax -Wl,--sort-common -Wl,--gc-sections
Click to expand...
Click to collapse
[SIZE="+1"]Step 3 - Cross-Compiling[/SIZE]
Now we'll compile our first program for Android!
Create a new file hello.c, and put inside:
Code:
#include <stdio.h>
int main (void)
{
puts ("Hello World!");
return 0;
}
Now you compile and strip it:
[email protected]:~# $CCC hello.c -o hello && $SSTRIP hello
Click to expand...
Click to collapse
Remember that $CCC and $SSTRIP command will only work if you source'd cc.sh script explained above. $CCC command compiles source code to a binary with already optimized flags (device flags, optimization level, optimization flags, linker flags), while $SSTRIP command strips "bloat" from output binary, such as comments and notices. The purpose is to make a binary smaller and faster.
You can check if your binary has been compiled properly through readelf command.
[email protected]:~# readelf -A hello
Attribute Section: aeabi
File Attributes
Tag_CPU_name: "ARM v7"
Tag_CPU_arch: v7
Tag_CPU_arch_profile: Application
Tag_ARM_ISA_use: Yes
Tag_THUMB_ISA_use: Thumb-2
Tag_FP_arch: VFPv3
Tag_Advanced_SIMD_arch: NEONv1
Tag_ABI_PCS_wchar_t: 4
Tag_ABI_FP_denormal: Needed
Tag_ABI_FP_exceptions: Needed
Tag_ABI_FP_number_model: IEEE 754
Tag_ABI_align_needed: 8-byte
Tag_ABI_enum_size: int
Tag_ABI_HardFP_use: SP and DP
Tag_ABI_optimization_goals: Aggressive Speed
Tag_CPU_unaligned_access: v6
Tag_DIV_use: Not allowed
Click to expand...
Click to collapse
As you can see, I've compiled a binary optimized for ARM v7, with THUMB-2 instructions and NEON support. How nice! Is it because of device-specific flags? Let's check what happens if we use $CC instead of $CCC:
[email protected]:~# readelf -A hello2
Attribute Section: aeabi
File Attributes
Tag_CPU_name: "5TE"
Tag_CPU_arch: v5TE
Tag_ARM_ISA_use: Yes
Tag_THUMB_ISA_use: Thumb-1
Tag_FP_arch: VFPv2
Tag_ABI_PCS_wchar_t: 4
Tag_ABI_FP_denormal: Needed
Tag_ABI_FP_exceptions: Needed
Tag_ABI_FP_number_model: IEEE 754
Tag_ABI_align_needed: 8-byte
Tag_ABI_enum_size: int
Tag_ABI_optimization_goals: Aggressive Speed
Tag_DIV_use: Not allowed
Click to expand...
Click to collapse
As you can see, if you do not specify flags, you'll lose major portion of optimizations. Of course binary will work properly, hence it has been cross-compiled for ARM, but we can always make it smaller and faster!
[SIZE="+1"]Step 4 - Testing[/SIZE]
A favourite part of everything, tests!
[email protected]:~/shared# adb shell
[email protected]:/ # sysrw
[email protected]:/ # exit
[email protected]:~/shared# adb push hello /system/bin/hello
95 KB/s (5124 bytes in 0.052s)
[email protected]:~/shared# adb shell
[email protected]:/ # chmod 755 /system/bin/hello
[email protected]:/ # chown root:system /system/bin/hello
[email protected]:/ # exit
Click to expand...
Click to collapse
In above example I pushed my binary straight to /system/bin directory, which is in the Android's PATH. If you don't have rooted device that's not a problem, you can use /data/local directory or /storage/sdcard0. You can also upload your binary anywhere you want and download it as any other file, then run from /storage/sdcard0/Download, this way doesn't require even working ADB . Just don't forget about setting proper permissions afterwards!
Now let's try to run it!
{
"lightbox_close": "Close",
"lightbox_next": "Next",
"lightbox_previous": "Previous",
"lightbox_error": "The requested content cannot be loaded. Please try again later.",
"lightbox_start_slideshow": "Start slideshow",
"lightbox_stop_slideshow": "Stop slideshow",
"lightbox_full_screen": "Full screen",
"lightbox_thumbnails": "Thumbnails",
"lightbox_download": "Download",
"lightbox_share": "Share",
"lightbox_zoom": "Zoom",
"lightbox_new_window": "New window",
"lightbox_toggle_sidebar": "Toggle sidebar"
}
If your binary is not in the PATH, you should write full path to your binary of course. As I pushed my binary to /system/bin, I don't need to do so.
If everything finished successfully and you got your very first Hello World response as above, congratulations. You've just compiled and ran your first native C/C++ program on Android device.
[SIZE="+1"]What to do next?[/SIZE]
In theory, you can now compile anything you want. Here are some apps that I'm using in my ArchiDroid ROM:
Pixelserv
Haveged
Dnsmasq
DNRD
Rinetd
These are only a few examples. You can compile anything you want, or even write your own native applications. Good luck!
JustArchi said:
[SIZE=+1]What is a Cross-Compiler?[/SIZE]
[SIZE=+1]How is that connected with an Android?[/SIZE]
In order to create a native C/C++ binary for an Android, you must firstly compile the source code. Usually you can't do so on an Android itself due to lack of proper tools and environment, or hardware barriers, especially amount of RAM. This is why you should learn how to cross-compile, to create a binary on your PC, that your ARM-based Android will understand.
[SIZE=+1]Why do I need it?[/SIZE]
You need to learn cross-compiling technique if you want to run native C/C++ programs on an Android. Actually, if you've already built your own custom ROM from AOSP sources (i.e. CyanogenMod), then you used cross-compiling tools and methods even without noticing .
Building an AOSP ROM is fairly easy, there's one command like brunch, which does the job. However, what if you want to compile a custom, not natively included binary? This is the purpose of this tutorial.
[SIZE=+1]What I will learn from this guide?[/SIZE]
How to properly set C/C++ building environment
How to build a native C/C++ application for Android device
How to optimize native binaries for my device
[SIZE=+1]Step 1 - The Beginning[/SIZE]
You should start from installing any Linux-based OS, I highly suggest trying a Debian-based distro (such as Ubuntu), or even Debian itself, as this tutorial is based on it .
In general, I highly suggest to compile an AOSP ROM (such as CyanogenMod) for your device firstly. This will help you to get familiar with cross-compiling on Android. I also suggest to compile one or two programs from source for your Linux, but if you're brave enough to learn cross-compiling without doing any of these, you can skip those suggestions .
[SIZE=+1]Step 2 - Setting up[/SIZE]
Firstly you should make sure that you have all required compile tools already.
This should do the trick and install all required components.
I suggest creating a new folder and navigating to it, just to avoid a mess, but you can organize everything as you wish.
Start from downloading NDK from here.
Now you should make a standalone toolchain, navigate to root of your ndk (this is important) and then build your toolchain:
You should edit bolded variables to your preferences. Toolchain is the version of GCC you want to use, 4.8 is currently the newest one, in the future it may be 4.9 and so on. Platform is a target API for your programs, this is important only for android-specific commands, such as logging to logcat. When compiling a native Linux program, this won't matter (but it's a good idea to set it properly, just in case). Install dir specifies destination of your toolchain, make sure that it's other than ndk (as you can see I have ndk in /root/ndk and toolchain in /root/ndkTC).
Now you need to download my exclusive cc.sh script from here and make it executable.
This script is a very handy tool written by me to make your life easier while cross-compiling. Before running it make sure to edit "BASIC" options, especially NDK paths. Apart from that it's a good idea to take a look at DEVICEFLAGS and setting them properly for your device, or clearing for generic build. You don't need to touch other ones unless you want/need them.
Just for a reference, I'll include currently editable options:
As you can notice, my magic script already contains bunch of optimizations, especially device-based optimizations, which are the most important. Now it's the time to run our script, but in current shell and not a new one.
Command "source cc.sh" executes cc.sh and "shares" the environment, which means that any exports will be exported to our current shell, and this is what we want. It acts the same as AOSP's ". build/envsetup.sh", so you can also use . instead of source.
As you can see above, my script should let you know if it properly set everything, especially if $CC points to our ndkTC. It also set a generic "$CCC" and "$CCXX" commands, which are optimized versions of standard $CC. $CC points to our cross-compiler, $CCC points to our cross-compiler and also includes our optimization flags.
[SIZE=+1]Step 3 - Cross-Compiling[/SIZE]
Now we'll compile our first program for Android!
Create a new file hello.c, and put inside:
Code:
#include <stdio.h>
int main (void)
{
puts ("Hello World!");
return 0;
}
Now you compile and strip it:
Remember that $CCC and $SSTRIP command will only work if you source'd cc.sh script explained above. $CCC command compiles source code to a binary with already optimized flags (device flags, optimization level, optimization flags, linker flags), while $SSTRIP command strips "bloat" from output binary, such as comments and notices. The purpose is to make a binary smaller and faster.
You can check if your binary has been compiled properly through readelf command.
As you can see, I've compiled a binary optimized for ARM v7, with THUMB-2 instructions and NEON support. How nice! Is it because of device-specific flags? Let's check what happens if we use $CC instead of $CCC:
As you can see, if you do not specify flags, you'll lose major portion of optimizations. Of course binary will work properly, hence it has been cross-compiled for ARM, but we can always make it smaller and faster!
[SIZE=+1]Step 4 - Testing[/SIZE]
A favourite part of everything, tests!
In above example I pushed my binary straight to /system/bin directory, which is in the Android's PATH. If you don't have rooted device that's not a problem, you can use /data/local directory or /storage/sdcard0. You can also upload your binary anywhere you want and download it as any other file, then run from /storage/sdcard0/Download, this way doesn't require even working ADB . Just don't forget about setting proper permissions afterwards!
Now let's try to run it!
If your binary is not in the PATH, you should write full path to your binary of course. As I pushed my binary to /system/bin, I don't need to do so.
If everything finished successfully and you got your very first Hello World response as above, congratulations. You've just compiled and ran your first native C/C++ program on Android device.
[SIZE=+1]What to do next?[/SIZE]
In theory, you can now compile anything you want. Here are some apps that I'm using in my ArchiDroid ROM:
Pixelserv
Haveged
Dnsmasq
DNRD
Rinetd
These are only a few examples. You can compile anything you want, or even write your own native applications. Good luck!
Click to expand...
Click to collapse
[Mod Edit: Please don't quote the whole OP]
Fricking awesome. Worked perfect on my builduntu running in VirtualBox
dicksteele said:
Fricking awesome. Worked perfect on my builduntu running in VirtualBox
Click to expand...
Click to collapse
I'm very glad it worked for you .
Maybe you happen to know which packages checkinstall depends on? I want to run this on Arch - pun not intended - and pacman doesn't exactly talk with debs.
(Przy okazji, świetny tutorial c: )
Dragoon Aethis said:
Maybe you happen to know which packages checkinstall depends on? I want to run this on Arch - pun not intended - and pacman doesn't exactly talk with debs.
(Przy okazji, świetny tutorial c: )
Click to expand...
Click to collapse
Checkinstall makes sure that you have all required packages installed. You can achieve nearly the same by installing "build-essential, gcc, g++, make", and that should be enough I guess .
Also, big kudos to @willverduzco for featuring my guide on XDA portal!
I would like to see a guide for llvm/ clang.
Sent from my GT-I9000 using xda app-developers app
maybe a bit irrelevant... but i wanted to learn how to cross compile/port a binary (for example "applypatch") for cygwin... any link to guide will be helpful
Thank You
DerRomtester said:
I would like to see a guide for llvm/ clang.
Sent from my GT-I9000 using xda app-developers app
Click to expand...
Click to collapse
When making standalone toolchain you should use clang instead of gcc. You should also study my cc.sh script and adapt to your own. After that, steps are nearly the same.
EnerJon said:
maybe a bit irrelevant... but i wanted to learn how to cross compile/port a binary (for example "applypatch") for cygwin... any link to guide will be helpful
Thank You
Click to expand...
Click to collapse
Using Cygwin for such kind of things is... bad. Install VirtualBox and any Linux distro if you want to master cross-compile technique.
JustArchi said:
Using Cygwin for such kind of things is... bad. Install VirtualBox and any Linux distro if you want to master cross-compile technique.
Click to expand...
Click to collapse
Actually i was making a tool for windows to generate/apply OTA for Android ROMs... i wanted to compile/port "IMGDIFF2" and "applypatch" from android sources...
EnerJon said:
Actually i was making a tool for windows to generate/apply OTA for Android ROMs... i wanted to compile/port "IMGDIFF2" and "applypatch" from android sources...
Click to expand...
Click to collapse
Then you should find your sources for IMGDIFF2 and applypatch and compile from source for Android, just like example hello.c above.
@JustArchi I saw this guide mentioned on the portal and read through it. Very interesting stuff. Great work explaining. I've got several questions, however, perhaps you can elaborate on.
My primary PC OS is Gentoo Linux (I've been using it for 10 years), in patricular ~amd64 which is the equivalent of Debian unstable. In Gentoo, all packages are compiled from the sources. I have a very up to date complete toolchain already installed and functioning properly as part of the native package installation system which uses portage for maintaining and updating.
I've already compiled CM and AOSP for my device, but I can't for the life of me understand why when setting up my build environment using either Google or CM tools several much older versions of GCC and GLIBC are installed into my source repos and used to build the ROM when the prerequisites for building the environment already require a working toolchain on the host build box?
Isn't there a way to just use the native toolchain from the host? Ideally, I'd love to free up the space used by these extra compilers and libraries for sources instead. Additionally, since my toolchain is much newer (gcc-4.8.2, glibc-2.19, etc) and optimized for my hardware than these generic prebuilt binaries, my ROM builds would compile faster and more optimized if I could use it instead.
The big question I ask is would you know what I'd have to do to setup my native environment to build Android? I'd truly love to be able to get rid of these other toolchains and free up the space on my harddrive. Any help would be greatly appreciated. TIA
JustArchi said:
When making standalone toolchain you should use clang instead of gcc. You should also study my cc.sh script and adapt to your own. After that, steps are nearly the same.
Using Cygwin for such kind of things is... bad. Install VirtualBox and any Linux distro if you want to master cross-compile technique.
Click to expand...
Click to collapse
I try this. I would like to cross compile a kernel with clang. Hopefully i get it working.
Odysseus1962 said:
@JustArchi I saw this guide mentioned on the portal and read through it. Very interesting stuff. Great work explaining. I've got several questions, however, perhaps you can elaborate on.
My primary PC OS is Gentoo Linux (I've been using it for 10 years), in patricular ~amd64 which is the equivalent of Debian unstable. In Gentoo, all packages are compiled from the sources. I have a very up to date complete toolchain already installed and functioning properly as part of the native package installation system which uses portage for maintaining and updating.
I've already compiled CM and AOSP for my device, but I can't for the life of me understand why when setting up my build environment using either Google or CM tools several much older versions of GCC and GLIBC are installed into my source repos and used to build the ROM when the prerequisites for building the environment already require a working toolchain on the host build box?
Isn't there a way to just use the native toolchain from the host? Ideally, I'd love to free up the space used by these extra compilers and libraries for sources instead. Additionally, since my toolchain is much newer (gcc-4.8.2, glibc-2.19, etc) and optimized for my hardware than these generic prebuilt binaries, my ROM builds would compile faster and more optimized if I could use it instead.
The big question I ask is would you know what I'd have to do to setup my native environment to build Android? I'd truly love to be able to get rid of these other toolchains and free up the space on my harddrive. Any help would be greatly appreciated. TIA
Click to expand...
Click to collapse
You need special compiler capable of compiling for specific architecture, this is not the same as native GCC toolchain for amd64. When you're using native compiler, output is always designed for amd64 or i386, when using cross-compiler, output is always designed for ARM, or other specific architecture.
JustArchi said:
You need special compiler capable of compiling for specific architecture, this is not the same as native GCC toolchain for amd64. When you're using native compiler, output is always designed for amd64 or i386, when using cross-compiler, output is always designed for ARM, or other specific architecture.
Click to expand...
Click to collapse
Thanks for the quick response. I'm a bit disappointed, but I'm still wondering that there has to be some way for me to utilize the ARM toolchain I currently have installed to cross-compile from the sources a more updated optimized toolchain for me to build with. Unfortunately (for me), that Gentoo is more of a niche Linux distro so finding help in their forums for working with ARM is difficult. As it is, it took much effort and trial and error to setup my current configuration to build with since nearly everything on the net is geared towards Ubuntu / Debian (both of which I feel are loaded with useless cruft and dependencies for things I have never and will never use).
Anyhow thanks again for this great guide, and for your continued work here helping us all.
Ciao
Dropbox link is down
Inviato dal mio GT-I9300 utilizzando Tapatalk
Code:
#!/bin/bash
# _ _ _ _ _
# | |_ _ ___| |_ / \ _ __ ___| |__ (_)
# _ | | | | / __| __| / _ \ | '__/ __| '_ \| |
# | |_| | |_| \__ \ |_ / ___ \| | | (__| | | | |
# \___/ \__,_|___/\__/_/ \_\_| \___|_| |_|_|
#
# Copyright 2014 Łukasz "JustArchi" Domeradzki
# Contact: [email protected]
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#############
### BASIC ###
#############
# Root of NDK, the one which contains $NDK/ndk-build binary
NDK="/root/ndk"
# Root of NDK toolchain, the one used in --install-dir from $NDK/build/tools/make-standalone-toolchain.sh. Make sure it contains $NDKTC/bin directory with $CROSS_COMPILE binaries
NDKTC="/root/ndkTC"
# Optional, may help NDK in some cases, should be equal to GCC version of the toolchain specified above
export NDK_TOOLCHAIN_VERSION=4.8
# This flag turns on ADVANCED section below, you should use "0" if you want easy compiling for generic targets, or "1" if you want to get best optimized results for specific targets
# In general it's strongly suggested to leave it turned on, but if you're using makefiles, which already specify optimization level and everything else, then of course you may want to turn it off
ADVANCED="1"
################
### ADVANCED ###
################
# Device CFLAGS, these should be taken from TARGET_GLOBAL_CFLAGS property of BoardCommonConfig.mk of your device, eventually leave them empty for generic non-device-optimized build
# Please notice that -march flag comes from TARGET_ARCH_VARIANT
DEVICECFLAGS="-march=armv7-a -mtune=cortex-a9 -mfpu=neon -mfloat-abi=softfp"
# This specifies optimization level used during compilation. Usually it's a good idea to keep it on "-O2" for best results, but you may want to experiment with "-Os", "-O3" or "-Ofast"
OLEVEL="-O2"
# This specifies extra optimization flags, which are not selected by any of optimization levels chosen above
# Please notice that they're pretty EXPERIMENTAL, and if you get any compilation errors, the first step is experimenting with them or disabling them completely, you may also want to try different O level
OPTICFLAGS="-s -flto=8 -ffunction-sections -fdata-sections -fvisibility=hidden -funswitch-loops -frename-registers -frerun-cse-after-loop -fomit-frame-pointer -fgcse-after-reload -fgcse-sm -fgcse-las -fweb -ftracer -fstrict-aliasing"
# This specifies extra linker optimizations. Same as above, in case of problems this is second step for finding out the culprit
LDFLAGS="-Wl,-O1 -Wl,--as-needed -Wl,--relax -Wl,--sort-common -Wl,--gc-sections"
# This specifies additional sections to strip, for extra savings on size
STRIPFLAGS="-s -R .note -R .comment -R .gnu.version -R .gnu.version_r"
# Additional definitions, which may help some binaries to work with android
DEFFLAGS="-DNDEBUG -D__ANDROID__"
##############
### EXPERT ###
##############
# This specifies host (target) for makefiles. In some rare scenarios you may also try "--host=arm-linux-androideabi"
# In general you shouldn't change that, as you're compiling binaries for low-level ARM-EABI and not Android itself
CONFIGANDROID="--host=arm-linux-eabi"
# This specifies the CROSS_COMPILE variable, again, in some rare scenarios you may also try "arm-eabi-"
# But beware, NDK doesn't even offer anything apart from arm-linux-androideabi one, however custom toolchains such as Linaro offer arm-eabi as well
CROSS_COMPILE="arm-linux-androideabi-"
# This specifies if we should also override our native toolchain in the PATH in addition to overriding makefile commands such as CC
# You should NOT enable it, unless your makefile calls "gcc" instead of "$CC" and you want to point "gcc" (and similar) to NDKTC
# However, in such case, you should either fix makefile yourself or not use it at all
# You've been warned, this is not a good idea
TCOVERRIDE="0"
# Workaround for some broken compilers with malloc problems (undefined reference to rpl_malloc and similar errors during compiling), don't uncomment unless you need it
#export ac_cv_func_malloc_0_nonnull=yes
############
### CORE ###
############
# You shouldn't edit anything from now on
if [ "$ADVANCED" -ne 0 ]; then # If advanced is specified, we override flags used by makefiles with our optimized ones, of course if makefile allows that
export CFLAGS="$OLEVEL $DEVICECFLAGS $OPTICFLAGS $DEFFLAGS"
export LOCAL_CFLAGS="$CFLAGS"
export CXXFLAGS="$CFLAGS" # We use same flags for CXX as well
export LOCAL_CXXFLAGS="$CXXFLAGS"
export CPPFLAGS="$CPPFLAGS" # Yes, CPP is the same as CXX, because they're both used in different makefiles/compilers, unfortunately
export LOCAL_CPPFLAGS="$CPPFLAGS"
export LDFLAGS="$LDFLAGS"
export LOCAL_LDFLAGS="$LDFLAGS"
fi
if [ ! -z "$NDK" ] && [ "$(echo $PATH | grep -qi $NDK; echo $?)" -ne 0 ]; then # If NDK doesn't exist in the path (yet), prepend it
export PATH="$NDK:$PATH"
fi
if [ ! -z "$NDKTC" ] && [ "$(echo $PATH | grep -qi $NDKTC; echo $?)" -ne 0 ]; then # If NDKTC doesn't exist in the path (yet), prepend it
export PATH="$NDKTC/bin:$PATH"
fi
export CROSS_COMPILE="$CROSS_COMPILE" # All makefiles depend on CROSS_COMPILE variable, this is important to set"
export AS=${CROSS_COMPILE}as
export AR=${CROSS_COMPILE}ar
export CC=${CROSS_COMPILE}gcc
export CXX=${CROSS_COMPILE}g++
export CPP=${CROSS_COMPILE}cpp
export LD=${CROSS_COMPILE}ld
export NM=${CROSS_COMPILE}nm
export OBJCOPY=${CROSS_COMPILE}objcopy
export OBJDUMP=${CROSS_COMPILE}objdump
export READELF=${CROSS_COMPILE}readelf
export RANLIB=${CROSS_COMPILE}ranlib
export SIZE=${CROSS_COMPILE}size
export STRINGS=${CROSS_COMPILE}strings
export STRIP=${CROSS_COMPILE}strip
if [ "$TCOVERRIDE" -eq 1 ]; then # This is not a a good idea...
alias as="$AS"
alias ar="$AR"
alias gcc="$CC"
alias g++="$CXX"
alias cpp="$CPP"
alias ld="$LD"
alias nm="$NM"
alias objcopy="$OBJCOPY"
alias objdump="$OBJDUMP"
alias readelf="$READELF"
alias ranlib="$RANLIB"
alias size="$SIZE"
alias strings="$STRINGS"
alias strip="$STRIP"
fi
export CONFIGANDROID="$CONFIGANDROID"
export CCC="$CC $CFLAGS $LDFLAGS"
export CXX="$CXX $CXXFLAGS $LDFLAGS"
export SSTRIP="$STRIP $STRIPFLAGS"
echo "Done setting your environment"
echo
echo "CFLAGS: $CFLAGS"
echo "LDFLAGS: $LDFLAGS"
echo "CC points to $CC and this points to $(which "$CC")"
echo
echo "Use \"\$CC\" command for calling gcc and \"\$CCC\" command for calling our optimized CC"
echo "Use \"\$CXX\" command for calling g++ and \"\$CCXX\" for calling our optimized CXX"
echo "Use \"\$STRIP\" command for calling strip and \"\$SSTRIP\" command for calling our optimized STRIP"
echo
echo "Example: \"\$CCC myprogram.c -o mybinary && \$SSTRIP mybinary \""
echo
echo "When using makefiles with configure options, always use \"./configure \$CONFIGANDROID\" instead of using \"./configure\" itself"
echo "Please notice that makefiles may, or may not, borrow our CFLAGS and LFLAGS, so I suggest to double-check them and eventually append them to makefile itself"
echo "Pro tip: Makefiles with configure options always borrow CC, CFLAGS and LDFLAGS, so if you're using ./configure, probably you don't need to do anything else"
Temporary replacement for cc.sh, as dropbox will be up soon.
Hi!
Great info.
To cross compile some packages with autotools (./configure; make; make install) it's needed to export the SYSROOT path ($ndkTC/sysroot) and include the option --sysroot=$SYSROOT on CFLAGS. Some need too --with-sysroot=$SYSROOT as configure option. This way the configure script and linker can find the libraries.
If i'm building a library that must be used as dependence to other program I use to include --static to build a static library and --prefix=$SYSROOT/usr on configure options to install the lib on toolchain sysroot folder...
Thanks.
sfortier said:
Hi!
Great info.
To cross compile some packages with autotools (./configure; make; make install) it's needed to export the SYSROOT path ($ndkTC/sysroot) and include the option --sysroot=$SYSROOT on CFLAGS. Some need too --with-sysroot=$SYSROOT as configure option. This way the configure script and linker can find the libraries.
If i'm building a library that must be used as dependence to other program I use to include --static to build a static library and --prefix=$SYSROOT/usr on configure options to install the lib on toolchain sysroot folder...
Thanks.
Click to expand...
Click to collapse
Hey.
Nice to know, I'll update my script with that. Thanks!
My last attempt to cross compile something was qemu (i'm was thinking on run windows on my tablet... )
I needed to build glib, pixmap, libpng, zlib, libjpeg-turbo, libiconv, libffi, libintl. Now I have my toolchain with all these usefull static (I prefer static libs to simplify binary installation) libs installed!