MSysGit mergetool broken after 1.7.7

I've been using Git on Windows (MSysGit) happily with BeyondCompare 3 for quite a while. Mostly I use BC3 to view changes. Occasionally I use it to merge conflicts. I don't merge very often, thanks to Git's powerful merging capability. It has worked fine. Recently I upgraded to version 1.7.10-preview20120409, and suddenly the git's merge command always fail with below error messages:

Merging:
my/path/filename.cs

Normal merge conflict for 'my/path/filename.cs':
  {local}: modified file
  {remote}: modified file
Hit return to start merge resolution tool (BC3):
git_path/libexec/git-core/mergetools/BC3: line 9: bcompare: command not found
my/path/filename.cs seems unchanged.
Was the merge successful? [y/n] n

Where is this line 9? It turns out that it seems MSysGit (or Git?) has put all merge tools to a folder named "git_path/libexec/git-core/mergetools/". In my installation, I can already find quite a bunch of tools:

araxis
bc3
defaults
deltawalker
diffuse
ecmerge
emerge
kdiff3
kompare
meld
opendiff
p4merge
tkdiff
tortoisemerge
vim
xxdiff

and they are actually bash scripts. For example, the bc3 file is like below:

diff_cmd () {
 "$merge_tool_path" "$LOCAL" "$REMOTE"
}

merge_cmd () {
 touch "$BACKUP"
 if $base_present
 then
  "$merge_tool_path" "$LOCAL" "$REMOTE" "$BASE" \
   -mergeoutput="$MERGED"
 else
  "$merge_tool_path" "$LOCAL" "$REMOTE" \
   -mergeoutput="$MERGED"
 fi
 check_unchanged
}

translate_merge_tool_path() {
 if type bcomp >/dev/null 2>/dev/null
 then
  echo bcomp
 else
  echo bcompare
 fi
}

and there is really is a line 9! Of course directly modify the file will do the trick. But I guess there are better ways. After testing around, trying to override the merge_tool_path environment variable, I found it's easier if I just put BC3's folder to path, and nothing else need to be changed!

Git: generate patches, after or before commit happens.

Occasionally, if not often, I need to generate patches to be applied to another git copy. This can happen either before I commit or after. It turns out, the ways to do it for the two scenarios are different.

If the change set is already committed, then below command line can generate the patch:

git format-patch -1

This will create a patch in the current folder with file name like 0001-comments-in-the-commit.patch.

-1 in the command line means generating patch for last one commit. If you want to patch in more commits, specify the number here. If you don't like the default file name and want to specify your own output file name, then you can direct output to stdout and the redirect stdout to a file, like below:

  git format-patch -1 --stdout > your_file_name.patch

So how do you do it when you don't want to commit the changes? A simple way is to generate the diff output and redirect it to a file. The problems is, like most people, I'm not satisfied with git's diff output and I use a graphical tool, BeyandCompare, to do the diff viewing. If I execute

git diff > pathfile.txt


the BC is invoked and there is nothing redirected to the patch file, unsurprisingly. So how to force the diff command to skip the GUI tool? As in most of the cases, a command line switch comes to help: --no-ext-diff. Below command line will happily output the result to console and skip BC:

 git diff --no-ext-diff Applying patches generated with git diff is simple:
patch -p1 < patch-file.txt

git diff: ignore the deleted files and newly added files

I have a project that has gone through some serious refactoring, during which a lot of garbage files were removed. Then during a code review, the HEAD was compared with the last release tag and bang!, BeyandCompare UI was popped up so many times that we had to abort it after a while.

git diff sha-for-pre-release

Then I though there must be a way to skip this and it turned out there is. When using git diff, there is an option, --diff-filter, among tons of others. And below are excerpted from the manual:

--diff-filter=[(A|C|D|M|R|T|U|X|B)…[*]] 
Select only files that are Added (A), Copied (C), Deleted (D), Modified (M), Renamed (R), have their type(i.e. regular file, symlink, submodule, …) changed (T), are Unmerged (U), are Unknown (X), or have had their pairing Broken (B). Any combination of the filter characters (including none) can be used. When * (All-or-none) is added to the combination, all paths are selected if there is any file that matches other criteria in the comparison; if there is no file that matches other criteria, nothing is selected. 

My favorite option is --diff-filter=MRT, which will skip added and delete files.

Configurations of application using IoC, Spring.NET

I have being using Spring when I worked in Java world. Now that I'm in C# part of the world, I'd naturally go with Spring.NET. To be honest, before using Spring.NET, I did not even heard about StructureMap, Windsor, Autofac, NInject, let alone the Funq, Munq, Light.Core, etc. In general, I could live with XML configuration, and some configuration feature derived from XML really makes passing configuration data much easier, by putting below object definition in the container, you can isolate your often changed settings to a file without writing any glue code:

  <object id="SYSTEM.ConfigPropertyHolder" type="Spring.Objects.Factory.Config.PropertyPlaceholderConfigurer, Spring.Core">
    <property name="LastLocationOverrides" value="true"/>
    <property name="ConfigSections" value="config.app"/>
    <property name="Locations">
      <list>
        <value>~/Config/config.app.xml</value>
      </list>
    </property>
  </object>

now I can use a file named "config.app.xml", in folder Config, to hold my config settings:

<?xml version="1.0" encoding="utf-8" ?>
<configuration>
  <configSections>
    <section name="config.app" type="System.Configuration.NameValueSectionHandler, System" />
  </configSections>

  <config.app>
    <add key="value1" value="some data" />
    <add key="value2" value="other data" />
    ...
  </config.app>
</configuration>

and in the container definition xml file, I can use the settings defined in the config file:

  <object id="obj_id" type="type.name">
    <constructor-arg name="para1name" value="${value1}"/>
    <property name="pro1name" value="${value2}"/>
  </object>

And since most of Spring.NET's objects are defined with an id, there is even a further way to do the customization, using another class "Spring.Objects.Factory.Config.PropertyOverrideConfigurer, Spring.Core". Put such an object definition in container,

  <object id="SYSTEM.ConfigPropertyOverride" type="Spring.Objects.Factory.Config.PropertyOverrideConfigurer, Spring.Core">
    <property name="IgnoreResourceNotFound" value="true"/>
    <property name="ConfigSections" value="config.system"/>
    <property name="Locations">
      <list>
        <value>~/Config/config.sys.xml</value>
      </list>
    </property>
  </object>

you gain the capability to directly change object properties in config.sys.app:

  <config.system>

    <!--the key must be a <object_in_context>.<property>-->
    <add key="obj_id.someProperty" value="12345"/>
    <add key="another_obj.anotherProp" value="good!"/>
    ...
  </config.system>

Ugly or not, it's really convenient and powerful. In general, Spring.NET is really a feature rich software package, not only a IoC container. And also the document is really good, comparing to most other OSS projects.

So what's the problem? The problem is I'm getting more and more tired with the angle brackets, and I would like to try something new, like NInject, Autofac.

SubWCRev.exe and its dependencies

Our team is using Subversion as CVS, and thus it's natural to use SubWCRev.exe to retrieve source code version and embed into VersionInfo.cs file, so we can already rebuild the software when we get a bug report.

The problem is when the build process is running on build server, there is no TortoiseSVN installed on the server and I don't want to install it on all the build agents. So I took some time to find out the dlls that SubWCRev.exe depends on, in case I forget it:

# intl3_tsvn32.dll
# libapr_tsvn32.dll
# libaprutil_tsvn32.dll
# libsasl32.dll
# libsvn_tsvn32.dll

tell Git to do ignore your changes to some files

Normally Git is so powerful with its branching capabilities, that you seldom need this feature. But sometimes it can be handy if you can simply tell Git to "pretend" that you have not changed some files, even they have really been changed.

If you are using Git as a SVN client, like I do, you will know that Git refuses sync with SVN with local changes. In a large project, it can be that there is a common development environment with a common deployed application server so the client developers can connect and debug without worrying about setup their own instances. So it's also common that when you checkout codes out of the server, in my case it's a SVN repository, the config file contains URL pointing to that common server. If you are the server side developer, then you have to start your own instance and use a client to connect to it. Then you have to change the config file to point the URL to your instance and remember never commit the changes ,otherwise your colleague will kill you :)

There are a bunch of ways to overcome this. You can first only check in your changes without the config files, then stash the config files and then commit to SVN. Or you can create your own local branch with your config files, checkout and rebase the branch each time you want to debug with your config, which is what I'm doing now.

A new way I realized today while reading this post was, I can tell Git simply ignore the changes you have made to some files, with below command:

git update-index --assume-unchanged somefile.txt

this way Git will assume the file is never changed and will perform happily whatever command you want that can normally only be performed to a clean copy. So what happens that the file is changed in upstream repository? By manual, "Git will fail (gracefully) in case it needs to modify this file in the index e.g. when merging in a commit; thus, in case the assumed-untracked file is changed upstream, you will need to handle the situation manually."

Trick to make LiveMeeting addon for Outlook 2007 automatically connected.

Yes, my company uses Microsoft Office and related tools to handle normal communication. But some times it happens that the add-on is loaded but not really activated. You'll still see the LiveMeeting toolbar in outlook toolbar, but just one button "Enable LiveMeeting" there. This is annoying since you have to remember to push it each time you start outlook.

There is a simple solution to it: go to "start->all programes->microsoft office->microsoft office tools" and select the "microsoft office 2007 language settings", change the language to some other value, save it. Then open it again and change it back to your preferred language. Then start outlook again, you'll find the add-on is activated automatically with 3 buttons.