The interface between Subversion and external two- and three-way
differencing tools harkens back to a time when Subversion's only
contextual differencing capabilities were built around
invocations of the GNU diffutils toolchain, specifically the
diff and diff3 utilities.
To get the kind of behavior Subversion needed, it called these
utilities with more than a handful of options and parameters,
most of which were quite specific to the utilities. Some time
later, Subversion grew its own internal differencing library,
and as a failover mechanism, the
--diff3-cmd options were added to the
Subversion command-line client so that users could more easily
indicate that they preferred to use the GNU diff and diff3
utilities instead of the newfangled internal diff library. If
those options were used, Subversion would simply ignore the
internal diff library, and fall back to running those external
programs, lengthy argument lists and all. And that's where
things remain today.
It didn't take long for folks to realize that having such easy configuration mechanisms for specifying that Subversion should use the external GNU diff and diff3 utilities located at a particular place on the system could be applied toward the use of other differencing tools, too. After all, Subversion didn't actually verify that the things it was being told to run were members of the GNU diffutils toolchain. But the only configurable aspect of using those external tools is their location on the system—not the option set, parameter order, and so on. Subversion continues to throw all those GNU utility options at your external diff tool regardless of whether that program can understand those options. And that's where things get unintuitive for most users.
The decision on when to fire off a contextual two- or three-way
diff as part of a larger Subversion operation is made entirely
by Subversion and is affected by, among other things, whether
the files being operated on are human-readable as
determined by their
This means, for example, that even if you had the niftiest
Microsoft Word-aware differencing or merging tool in the
universe, it would never be invoked by Subversion as long as
your versioned Word documents had a configured MIME type that
denoted that they were not human-readable (such as
application/msword). For more about MIME
type settings, see the section called “File Content Type”
Much later, Subversion 1.5 introduced interactive resolution
of conflicts (described in
the section called “Resolve Any Conflicts”). One of the options
that this feature provides to users is the ability to
interactively launch a third-party merge tool. If this action
is taken, Subversion will check to see if the user has specified
such a tool for use in this way. Subversion will first check
SVN_MERGE environment variable for the
name of an external merge tool. If that variable is not set, it
will look for the same information in the value of
merge-tool-cmd runtime configuration
option. Upon finding a configured external merge tool, it will
invoke that tool.
While the general purposes of the three-way differencing and merge tools are roughly the same (finding a way to make separate-but-overlapping file changes live in harmony), Subversion exercises each of these options at different times and for different reasons. The internal three-way differencing engine and its optional external replacement are used when interaction with the user is not expected. In fact, significant delay introduced by such a tool can actually result in the failure of some time-sensitive Subversion operations. It's the external merge tool that is intended to be invoked interactively.
Now, while the interface between Subversion and an external merge tool is significantly less convoluted than that between Subversion and the diff and diff3 tools, the likelihood of finding such a tool whose calling conventions exactly match what Subversion expects is still quite low. The key to using external differencing and merge tools with Subversion is to use wrapper scripts, which convert the input from Subversion into something that your specific differencing tool can understand, and then convert the output of your tool back into a format that Subversion expects. The following sections cover the specifics of those expectations.
Subversion calls external diff programs with parameters suitable for the GNU diff utility, and expects only that the external program will return with a successful error code per the GNU diff definition thereof. For most alternative diff programs, only the sixth and seventh arguments—the paths of the files that represent the left and right sides of the diff, respectively—are of interest. Note that Subversion runs the diff program once per modified file covered by the Subversion operation, so if your program runs in an asynchronous fashion (or is “backgrounded”), you might have several instances of it all running simultaneously. Finally, Subversion expects that your program return an error code of 1 if your program detected differences, or 0 if it did not—any other error code is considered a fatal error.
Example 7.2. diffwrap.py
#!/usr/bin/env python import sys import os # Configure your favorite diff program here. DIFF = "/usr/local/bin/my-diff-tool" # Subversion provides the paths we need as the last two parameters. LEFT = sys.argv[-2] RIGHT = sys.argv[-1] # Call the diff command (change the following line to make sense for # your diff program). cmd = [DIFF, '--left', LEFT, '--right', RIGHT] os.execv(cmd, cmd) # Return an errorcode of 0 if no differences were detected, 1 if some were. # Any other errorcode will be treated as fatal.
Example 7.3. diffwrap.bat
@ECHO OFF REM Configure your favorite diff program here. SET DIFF="C:\Program Files\Funky Stuff\My Diff Tool.exe" REM Subversion provides the paths we need as the last two parameters. REM These are parameters 6 and 7 (unless you use svn diff -x, in REM which case, all bets are off). SET LEFT=%6 SET RIGHT=%7 REM Call the diff command (change the following line to make sense for REM your diff program). %DIFF% --left %LEFT% --right %RIGHT% REM Return an errorcode of 0 if no differences were detected, 1 if some were. REM Any other errorcode will be treated as fatal.
Subversion invokes three-way differencing programs to perform non-interactive merges. When configured to use an external three-way differencing program, it executes that program with parameters suitable for the GNU diff3 utility, expecting that the external program will return with a successful error code and that the full file contents that result from the completed merge operation are printed on the standard output stream (so that Subversion can redirect them into the appropriate version-controlled file). For most alternative merge programs, only the ninth, tenth, and eleventh arguments, the paths of the files which represent the “mine”, “older”, and “yours” inputs, respectively, are of interest. Note that because Subversion depends on the output of your merge program, your wrapper script must not exit before that output has been delivered to Subversion. When it finally does exit, it should return an error code of 0 if the merge was successful, or 1 if unresolved conflicts remain in the output—any other error code is considered a fatal error.
Example 7.4. diff3wrap.py
#!/usr/bin/env python import sys import os # Configure your favorite three-way diff program here. DIFF3 = "/usr/local/bin/my-diff3-tool" # Subversion provides the paths we need as the last three parameters. MINE = sys.argv[-3] OLDER = sys.argv[-2] YOURS = sys.argv[-1] # Call the three-way diff command (change the following line to make # sense for your three-way diff program). cmd = [DIFF3, '--older', OLDER, '--mine', MINE, '--yours', YOURS] os.execv(cmd, cmd) # After performing the merge, this script needs to print the contents # of the merged file to stdout. Do that in whatever way you see fit. # Return an errorcode of 0 on successful merge, 1 if unresolved conflicts # remain in the result. Any other errorcode will be treated as fatal.
Example 7.5. diff3wrap.bat
@ECHO OFF REM Configure your favorite three-way diff program here. SET DIFF3="C:\Program Files\Funky Stuff\My Diff3 Tool.exe" REM Subversion provides the paths we need as the last three parameters. REM These are parameters 9, 10, and 11. But we have access to only REM nine parameters at a time, so we shift our nine-parameter window REM twice to let us get to what we need. SHIFT SHIFT SET MINE=%7 SET OLDER=%8 SET YOURS=%9 REM Call the three-way diff command (change the following line to make REM sense for your three-way diff program). %DIFF3% --older %OLDER% --mine %MINE% --yours %YOURS% REM After performing the merge, this script needs to print the contents REM of the merged file to stdout. Do that in whatever way you see fit. REM Return an errorcode of 0 on successful merge, 1 if unresolved conflicts REM remain in the result. Any other errorcode will be treated as fatal.
Subversion optionally invokes an external merge tool as part of its support for interactive conflict resolution. It provides as arguments to the merge tool the following: the path of the unmodified base file, the path of the “theirs” file (which contains upstream changes), the path of the “mine” file (which contains local modifications), the path of the file into which the final resolved contents should be stored by the merge tool, and the working copy path of the conflicted file (relative to the original target of the merge operation). The merge tool is expected to return an error code of 0 to indicate success, or 1 to indicate failure.
Example 7.6. mergewrap.py
#!/usr/bin/env python import sys import os # Configure your favorite merge program here. MERGE = "/usr/local/bin/my-merge-tool" # Get the paths provided by Subversion. BASE = sys.argv THEIRS = sys.argv MINE = sys.argv MERGED = sys.argv WCPATH = sys.argv # Call the merge command (change the following line to make sense for # your merge program). cmd = [DIFF3, '--base', BASE, '--mine', MINE, '--theirs', THEIRS, '--outfile', MERGED] os.execv(cmd, cmd) # Return an errorcode of 0 if the conflict was resolved; 1 otherwise. # Any other errorcode will be treated as fatal.
Example 7.7. mergewrap.bat
@ECHO OFF REM Configure your favorite merge program here. SET DIFF3="C:\Program Files\Funky Stuff\My Merge Tool.exe" REM Get the paths provided by Subversion. SET BASE=%1 SET THEIRS=%2 SET MINE=%3 SET MERGED=%4 SET WCPATH=%5 REM Call the merge command (change the following line to make sense for REM your merge program). %DIFF3% --base %BASE% --mine %MINE% --theirs %THEIRS% --outfile %MERGED% REM Return an errorcode of 0 if the conflict was resolved; 1 otherwise. REM Any other errorcode will be treated as fatal.
 The GNU diff manual page puts it this way: “An exit status of 0 means no differences were found, 1 means some differences were found, and 2 means trouble.”