3.9. Contribute your changes

If you have finished changing the Wireshark sources to suit your needs, you might want to contribute your changes back to the Wireshark community. You gain the following benefits by contributing your improvements:

There's no direct way to commit changes to the SVN repository. Only a few people are authorised to actually make changes to the source code (check-in changed files). If you want to submit your changes, you should make a diff file (a patch) and upload it to the bug tracker.

3.9.1. What is a diff file (a patch)?

A diff file is a plain text file containing the differences between a pair of files (or a multiple of such file pairs).

[Tip]Tip!

A diff file is often also called a patch, as it can be used to patch an existing source file or tree with changes from somewhere else.

The Wireshark community is using patches to transfer source code changes between the authors.

A patch is both readable by humans and (as it is specially formatted) by some dedicated tools.

Here is a small example of a patch for file.h that makes the second argument in cf_continue_tail() volatile. It was created using svn diff, described below:


Index: file.h
===================================================================
--- file.h      (revision 21134)
+++ file.h      (revision 22401)
@@ -142,7 +142,7 @@
  * @param err the error code, if an error had occured
  * @return one of cf_read_status_t
  */
-cf_read_status_t cf_continue_tail(capture_file *cf, int to_read, int *err);
+cf_read_status_t cf_continue_tail(capture_file *cf, volatile int to_read, int *err);

 /**
  * Finish reading from "end" of a capture file.

	

The plus sign at the start of a line indicates an added line, a minus sign indicates a deleted line compared to the original sources.

We prefer to use so called "unified" diff files in Wireshark development, three unchanged lines before and after the actual changed parts are included. This makes it much easier for a merge/patch tool to find the right place(s) to change in the existing sources.

3.9.2. Generate a patch

There are several ways to generate patches. The preferred way is to generate them from an updated Subversion tree, since it avoids unnecessary integration work.

3.9.2.1. Using the svn command-line client

svn diff [changed_files] > svn.diff

Use the command line svn client to generate a patch in the required format from the changes you've made to your working copy. If you leave out the name of the changed file the svn client searches for all changes in the working copy and usually produces a patch containing more than just the change you want to send. Therefore you should always check the produced patch file.

If you've added a new file, e.g. packet-myprotocol.c, you can use svn add to add it to your local tree before generating the patch. Similarly, you can use svn rm for files that should be removed.

3.9.2.2. Using the diff feature of the GUI Subversion clients

Most (if not all) of the GUI Subversion clients (RapidSVN, TortoiseSVN, ...) have a built-in "diff" feature.

If you use TortoiseSVN:

TortoiseSVN (to be precise Subversion) keeps track of the files you have changed in the directories it controls, and will generate for you a unified diff file compiling the differences. To do so - after updating your sources from the SVN repository if needed - just right-click on the highest level directory and choose "TortoiseSVN" -> "Create patch...". You will be asked for a name and then the diff file will be created. The names of the files in the patch will be relative to the directory you have right-clicked on, so it will need to be applied on that level too.

When you create the diff file, it will include any difference TortoiseSVN finds in files in and under the directory you have right-clicked on, and nothing else. This means that changes you might have made for your specific configuration - like modifying "config.nmake" so that it uses your lib directory - will also be included, and you will need to remove these lines from the diff file. It also means that only changes will be recorded, i.e. if you have created new files -say, a new packet-xxx for a new protocol dissector- it will not be included in the diff, you need to add it separately. And, of course, if you have been working separately in two different patches, the .diff file will include both topics, which is probably not a good idea.

3.9.2.3. Using the diff tool

A diff file is generated, by comparing two files or directories between your own working copy and the "official" source tree. So to be able to do a diff, you should have two source trees on your computer, one with your working copy (containing your changes), and one with the "official" source tree (hopefully the latest SVN files) from www.wireshark.org.

If you have only changed a single file, you could type something like this:

diff -r -u --strip-trailing-cr svn-file.c work-file.c > foo.diff

To get a diff file for your complete directory (including subdirectories), you could type something like this:

diff -N -r -u --strip-trailing-cr ./svn-dir ./working-dir > foo.diff

It's a good idea to do a make distclean before the actual diff call, as this will remove a lot of temporary files which might be otherwise included in the diff. After doing the diff, you should edit the foo.diff file and remove unnecessary things, like your private changes to the config.nmake file.

Table 3.1. Some useful diff options

OptionPurpose
-NAdd new files when used in conjunction with -r.
-rRecursively compare any subdirectories found.
-uOutput unified context.
--strip-trailing-crStrip trailing carriage return on input. This is useful for Win32
-x PATExclude files that match PAT. This could be something like -x *.obj to exclude all win32 object files.

The diff tool has a lot options; they can be listed with:

diff --help

3.9.3. Some tips for a good patch

Some tips that will make the merging of your changes into the SVN tree much more likely (and you want exactly that, don't you :-):

  • Use the latest SVN sources, or alike. It's a good idea to work with the same sources that are used by the other developer's, this makes it usually much easier to apply your patch. For information about the different ways to get the sources, see Section 3.3, “Obtain the Wireshark sources”.

  • Update your SVN sources just before making a patch. For the same reasons as the previous point.

  • Do a "make clean" before generating the patch. This removes a lot of unneeded intermediate files (like object files) which can confuse the diff tool generating a lot of unneeded stuff which you have to remove by hand from the patch again.

  • Find a good descriptive filename for your patch. Think a moment to find a proper name for your patch file. Often a filename like wireshark.diff is used, which isn't really helpful if keeping several of these files and find the right one later. For example: If you want to commit changes to the datatypes of dissector foo, a good filename might be: packet-foo-datatypes.diff.

  • Don't put unrelated things into one large patch. A few smaller patches are usually easier to apply (but also don't put every changed line into a separate patch :-).

  • Remove any parts of the patch not related to the changes you want to submit. You can use a text editor for this. A common example for win32 developers are the differences in your private config.nmake file.

In general: making it easier to understand and apply your patch by one of the maintainers will make it much more likely (and faster) that it will actually be applied.

Please remember: you don't pay the person "on the other side of the mail" for his/her effort applying your patch!

3.9.4. Code Requirements

The core maintainers have done a lot of work fixing bugs and making code compile on the various platforms Wireshark supports.

To ensure Wireshark's source code quality, and to reduce the workload of the core maintainers, there are some things you should think about before submitting a patch.

[Warning]Warn!

Ignoring the code requirements will make it very likely that your patch will be rejected!

  • Follow the Wireshark source code style guide. Just because something compiles on your platform, that doesn't mean it'll compile on all of the other platforms for which Wireshark is built. Wireshark runs on many platforms, and can be compiled with a number of different compilers. See Section 7.2, “Coding styleguides” for details.

  • Submit dissectors as build-in whenever possible. Developing a new dissector as a plugin is a good idea because compiling is quicker, but it's best to convert dissectors to the build-in style before submitting for checkin. This reduces the number of files that must be installed with Wireshark and ensures your dissector will be available on all platforms.

    This is no hard-n-fast rule though. Many dissectors are straightforward so they can easily be put into 'the big pile', while some are ASN.1 based which takes a different approach, and some multiple sourcefile dissectors are more suitable to be placed separate as plugin.

  • Fuzz test your changes! Fuzz testing is a very effective way to automatically find a lot of dissector related bugs. You'll take a capture file containing packets affecting your dissector and the fuzz test will randomly change bytes in this file, so that unusual code paths in your dissector are checked. There are tools available to automatically do this on any number of input files, see: http://wiki.wireshark.org/FuzzTesting for details.

3.9.5. Sending your patch for inclusion

After generating a patch of your changes, you might want to have your changes included into the SVN repository.

To submit a patch, open a new ticket in the Wireshark bug database at http://bugs.wireshark.org/bugzilla/enter_bug.cgi?product=Wireshark. You must first create a bug, then attach your patch or patches.

  • Set the Product, Priority, and Severity as needed.

  • Add a Summary and Description, and create a bug using the Commit button. If your code has passed fuzz testing, please say so in the description.

  • Once the bug has been created, select Create a New Attachment and upload your patch or patches. Set the review_for_checkin flag to ?.

  • If possible and applicable, attach a capture file that demonstrates your new feature or protocol.

[Tip]Tip!

Setting the review_for_checkin is important. Without it, your patch won't show up in the pending patch request queue.

You might get one of the following responses to your patch request:

  • Your patch is checked into the SVN repository. Congratulations!

  • You are asked to provide additional information, capture files, or other material. If you haven't fuzzed your code, you may be asked to do so.

  • Your patch is rejected. You should get a response with the reason for rejection. Common reasons include not following the style guide, buggy or insecure code, and code that won't compile on other platforms. In each case you'll have to fix each problem and upload another patch.

  • You don't get any response to your patch (even after a few days or so). Possible reason: your patch might simply get lost, as all core maintainers were busy at that time and forgot to look at your patch. Simply send a mail asking if the patch was forgotten or if someone is still looking at it.