Hello possible new developer, and welcome to the Xastir project!
In order to get contributed patches accepted more easily by the Xastir developers:
- ) Read HowTo:Developer_Guidelines_and_Notes. Make sure to follow the formatting and indentation rules, and in particular the tab format (spaces, not tabs). If you don't like some of the formatting rules, abide by them anyway for consistency (Hey, I don't like some of them either, but consistency is more important than ideas I might have of coding style! --we7u).
- ) Check the Xastir pages at Sourceforge. There are sections for bug reports and feature requests. Anyone can log new bugs or feature requests, whether or not they are logged into Sourceforge. This is one of the best places to see what needs to be worked on, and to see if anyone else has had a similar idea.
- ) Check with the Xastir-Dev list first to see if anyone else is working on that particular idea or section of code.
- ) Verify on the Xastir-Dev list that your idea has some merit and chance of being accepted before you put your valuable time into the patch.
- ) Make sure you're willing to abide by the GPL license with respect to your patch.
- ) Use as generic C as possible, and comment what you've done, in English please!
- ) Keep in mind that Xastir runs on multiple operating systems, so code for that. Some #ifdef's may be required in order to make it work for the various operating systems.
- ) Xastir can be run in multiple languages, so code for that. If any user text is added, make sure to add language strings for them to the language files. If you don't know a particular language, add it to all of the language files in English. It will be translated by others later.
- ) Put your patches into either "context diff" or "unified diff" format. Don't submit complete files. Make sure your diff is against the latest Git sources, as that is the version that the devlopers work from. Create the diff file against the directories recursively if more than one file has been changed:
diff -ur xastir-1.7.0 xastir-modified-directory > mypatch
Here's a shorthand method if you're doing all this against the Git version. Do this from the top-level Xastir directory:
git diff -u > mypatch
- ) By compressing the patch and attaching it to an e-mail you preserve the lines exactly so that the "patch" program will have no trouble with it:
This creates mypatch.gz, which you can attach to an e-mail or put up on an ftp or http site for download.
- ) Contribute patches either directly to one of the active developers or to the Xastir-Dev mailing list.
- ) To apply a patch I typically do this:
gzip -d mypatch.gz
Check the path listed inside the patch:
Based on the paths listed inside the patch and the directory at which you're currently sitting, adjust the "p" number below so that the proper number of directories are stripped from the patch. This example strips off the first two directories:
patch -p2 <mypatch
To keep track of commits, rely on GitHub "Notifications". Sign up for a github account and "Watch" the project. You'll get notifications when things happen and can use GitHub's settings to control how you're notified (email or otherwise).
Xastir is a multi-threaded and multi-process application. It uses both threads and forks to do it's work. You must have a debugger that is capable of following these kinds of twists and turns in a program. Many older versions cannot.
Notes from Tom Russo
According to everything I can find about GDB, debugging of multithreaded apps has been supported in gdb for some time, and are certainly supported in gdb 6.x. For the last couple days I've been running xastir inside gdb instead of at the command prompt --- perhaps I won't keep forgetting to start it with -t now, and when it crashes I'll be where I need to be.
Jim --- this is possibly an option for you, too, if you're seeing segfaults frequently enough and can't solve the core file question. Try this:
gdb /usr/local/bin/xastir > run -t
When it crashes, it'll pop you right out into the debugger, whether a core file was created or not. You could then view the active threads:
> info threads
and get a stack trace of where the crash happened:
You could also probably list the code near where the crash happened:
If you send the output of those three commands to the xastir-dev list then it might help us narrow down the causes.
Notes from Curt
Note that the meaning of the "-t" command-line flag has been reversed. "xastir -?" should show the change once you compile it. This means that we'll do core dumps by default upon segfault instead of using the internal Xastir segfault handler. We've also re-enabled the "-g" compile option for GCC so that debugging information will remain embedded in the executable (unless you strip it). This should aid the development team to debug problems.
> Also, I cannot find anything in man bash that talks about core > dumps nor segmentation faults.
Seems like some stuff we're just supposed to know. _How_ we're supposed to know I don't know either... ;-)
Perhaps that last bit about SUID/SGID might be a clue: There are often exceptions to the rules for SUID/SGID programs. Try _not_ setting Xastir SUID (if you're not using AX.25 kernel networking ports that is) to see if you get a core file. Can you send a SIGSEGV to the running process and make it dump? I just tried this and I'm not getting a core file either.
Ah, I see in the "man bash" page:
ulimit -a All current limits are reported
This gives me:
core file size (blocks, -c) 0 data seg size (kbytes, -d) unlimited file size (blocks, -f) unlimited max locked memory (kbytes, -l) 32 max memory size (kbytes, -m) unlimited open files (-n) 1024 pipe size (512 bytes, -p) 8 stack size (kbytes, -s) unlimited cpu time (seconds, -t) unlimited max user processes (-u) 6139 virtual memory (kbytes, -v) unlimited
So... Looks like I need to set "ulimit -c" and try again:
ulimit -c unlimited
Now when I do "kill -11 <pid>" I get this:
+ Segmentation fault (core dumped) xastir
-rw------- 1 archer users 12320768 2006-02-02 09:31 core.7586
I just added "ulimit -c unlimited" to my .profile. For what it's worth, if Xastir has SUID permission bit set (4755) I don't get a core dump, but if it is reset I do (0755).
I can invoke ddd like this (it uses gdb under the hood):
ddd /usr/local/bin/xastir core.7586 &
or gdb like this:
gdb -c core.7586 /usr/local/bin/xastir
Once you have the program and core file loaded into the debugger you can display a backtrace to see where the program died. In the case of ddd, it's Status->Backtrace, then you click on one of the entries to make the source code window display the exact location.
Another good debugger to try is "UPS".
Another note about core files: Sometimes they get written where you don't expect. I just had one appear in ~/.xastir/tmp, so if you think you should have a core file and can't find it, try:
find . -type f | grep core
The core file may be written to a directory that Xastir is currently in, instead of where you started Xastir from.
Customized debugging builds
Sometimes it is helpful to build xastir with specialized compiler options to aid in debugging. For example, if you're trying to hunt down elusive segfaults, and you're using GCC, it might be wise to build with "-O -g -fno-inline" to prevent the compiler from optimizing quite so vigorously; aggressive optimization can sometimes lead to the debugger saying the code died on one line when in fact it's happening somewhere else.
To build with custom CFLAGS like this, just tell configure what you want CFLAGS to be:
cd build ../configure CFLAGS="-O -g -fno-inline"
Naturally, you need to build every file after changing configure like this, so do a make clean before building:
make clean make make install
Remember not to "make install-strip" when trying to do debugging builds, or your core files will not have debugging symbols in them and it will be harder to get useful information out of the debugger.
Segregating specialized builds
The automake/autoconf setup of xastir makes it easy to maintain several different builds of the code without having to clean out the directory and rebuild every time. One does this with "build directories" and executable suffixes.
To use this capability, make sure you're starting with a completely clean source directory. In the directory where you normally do your "git update", do a make distclean. This will remove anything that configure created as well as anything that make created. From this point forward, you don't build xastir in the source code directory anymore.
Make an empty directory somewhere --- this will be your build directory. I put my build directories in parallel with the source code directory. So my tree looks like this:
XASTIR-DEV | +----/xastir (source directory) | +---/build-normal | +---/build-debug
and so on.
So I would do:
cd XASTIR-DEV mkdir build-normal -p
In the build directory, you run configure using the path to the configure script:
cd build-normal ../xastir/configure make make-install
The configure script uses the path that you gave when you ran it to figure out where to find the source code.
The beauty of this is you can make a second build without doing a make clean:
cd XASTIR-DEV mkdir build-debug -p cd build-debug ../xastir/configure --program-suffix="-debug" CFLAGS="-O -g -fno-inline" make && sudo make install
This will build a second binary called "xastir-debug" and install it, but because you've done the buil in a separate directory, your normal compile is untouched.
cd ~/XASTIR-DEV/xastir git pull cd ../build-normal make && sudo make install cd ../build-debug make && sudo make install
will update both of your builds.
I also use this technique to share a single xastir source tree over NFS, building the code for multiple operating systems in separate build directories.
(log in to CYGWIN machine, mount NFS directory) cd /mounted/directory/XASTIR-DEV mkdir build-cygnus -p cd build-cygnus ../xastir/configure make && make install
It can also be used to maintain specialized configurations, for example a build with "rtree" disabled, a build with map caching disabled, etc. This is a good development trick to keep yourself honest --- make sure you can still build all your custom builds when you've done a large hacking run, and want to check if you have broken things inside ifdefs. Doing that without build directories requires a huge number of configure/make/make distclean cycles.
Anything else? Let's hear about what's still confusing or needs to be expanded in this document. Thanks!
APRS(tm) is a Trademark of Bob Bruninga
Copyright (C) 2006 The Xastir Group
This page is based on $Id: README.Contributing,v 1.13 2006/07/07 16:10:55