The 8th airing of the msysGit herald

Good morning git land!

This dark and quiet Sunday night is as good an occasion as any to offer to you the 8th airing of the msysGit Herald, the not-quite-biweekly news letter to keep you informed about msysGit, the effort to bring one of the most powerful Source Code Management systems to the poor souls stuck with Windows.

These are the covered topics:

	We have a working 'git svn'

	Interview with Christian Stimming

	GPL only needs to be acknowledged

	The installer was on diet

	core.autocrlf = true

	Update to a new ssh, problems with pulling msysGit

	Entering Babel: "rebase"

	git-cheetah is now a submodule

Maybe I should strike the "not quite biweekly"? But then, it _is_ not quite biweekly ;-)

Anyway, here it goes, the newest issue of the Herald, subtitle "Yeah, we have a working git svn!"

We have a working 'git svn' ===========================

Some time ago, we started work on compiling Perl ourselves, in order to be able to install the Subversion modules which are necessary to run 'git svn'. In the last issue of the Herald, I wrote something about our long and rocky road until we had a working 'git svn' (See the story "We managed to get git-svn to run!").

We played it safe, though, and had some alpha versions of a Git installer including 'git svn', and with good reason: Some testers came back and said that it would not work with https repositories, and that authentication was not supported. We included the necessary libraries and modules, cut a new Git installer, and the tests show that while it is slow, it works!

While working on git-svn, we realized that there are a few scripts in msysGit that cannot possibly work (yet), so we excluded them from the Git installer. These scripts are: archimport, cvsexportcommit, cvsimport, cvsserver, filter-branch, instaweb, send-email, and shell.

Interview with Christian Stimming =================================

Christian put in a big effort to get 'git svn' to work when I got stuck. He did such a great job, that I decided to interview him for the Herald, since I could not afford to buy him a house by way of saying thanks. So here it goes:

Some fellow developer in another Open Source project mentioned it and was enthusiastic about this new way of working. As my main reason for being involved in Open Source is learning new things, this got me curious.

In one project I had to work with a dead lame CVS server, where each contact to the repository took way way too long. Discovering that git would import all of the history and give me access to each and everything I wanted to know from the history brought back the fun into the project. From that moment on, I was hooked.

Speed. It's so fast, it's marvellous. This doesn't quite hold for the Windows version, but it is still faster than any other alternatives that I know of. And the GUI tools gitk and git-gui did a fantastic job of a low entry barrier - it's now really easy to start branching and merging a lot.

The command line options. Sometimes it seems to me all the really cool actions in git can only be invoked by a mysterious collection of weird option switches which 1. I've never heard of, 2. are hidden deep down in long manual pages which make it impossible to distinguish important options from the unimportant ones, and 3. change very frequently in new versions. For example, "git rebase -i" is a nice feature, but it is simply a different action than a one-shot "git rebase". Hence, if it is supposed to be really used, it should rather be a command such as "git interactive-rebase". The GUI tools go a long way to hide those mysterious option collections, but some of the daily workflow steps are still unavailable in the GUI. Rebase being the most prominent, I guess.

Seeing it up and running on Windows just as well as on Linux. 

Finding that "git pull" will create many more merge commits than I wanted, and that there doesn't seem to be an easy way of running "git fetch; git rebase" in one command. All in all there hasn't been much frustration potential in git... I guess this is a good thing.

Desktop machines with Linux or Windows, always using code from a central SVN repository. Really: I'm switching back and forth between Linux or Windows, and seeing git available on Windows just as well makes this fun again.

Programming. Oh, and programming. Well, there's also family: A wonderful wife and two cute daughters.

Schulze gets the Blues. 

go?) 

Git will empower the user! (Well, here the user is more the developer, but this doesn't rhyme as good.) The GUI tools will make it possible for everyone to use git instead of any of the previously popular central VCSs, but now with the decentralized features on top. This is step one in introducing a true decentralized model into the normal company workflows. Step two is to start using push and pull between the decentralized repositories, so that developers use the tools to represent their actual workflow. Step three: World domination!

GPL only needs to be acknowledged =================================

It is funny how licenses can make people speak up who normally would keep quiet, and so we got a complaint that we forced the users of our Git installer to accept the GNU Public License.

Personally, I think it is not nice to demand changes, especially license changes, or changes in the visibility of the license, when you get the software for free.

Also, I try to stay on the very safe side of legal issues, in order to avoid contact with lawyers, especially ones where they demand that I cross their hands with silver. So I was quite unwilling to change that.

However, none less than our benevolent dictator, Linus Torvalds spoke up, pointing out that it is not really a _choice_. You need not accept the GPL, it is sufficient that you acknowledge that this is the only license regulating your rights with regard to Git.

Now, that was very reasonable, so we changed it. 

Personal sidenote: I would do quite a few things for people whose work I benefit from every day; it is that fundamental idea of fairness (or tit for tat) encoded in the GPL which makes me extremely comfortable with that license, and I think that I am not the only one.

The installer was on diet =========================

The installer had a diet on Doctor's orders, but no worry, it is not bulimic. As stated previously, now even 'git svn' was added. How was that possible?

Pretty easy. We dragged around a lot of garbage in our installer. Some time ago, it was janitized, and lost a few megabytes already, but this time, even more effort was put into removing unneeded parts.

Mainly inside the Perl library, files were removed. Since the Git installer is not meant to be any development environment -- we do not even ship gcc -- shipping header files and linkable libraries does not make sense.

Neither is documentation for Perl needed, since nobody is supposed to develop the Perl scripts of Git without the development environment: that is what 'msysGit' is for.

So away went the documentation! 

A few files have been removed from Tcl, vim, some unneeded Perl modules, and cvs, too, so we actually ended up with an 8.0 megabyte installer _with_ 'git svn', whereas the previous installer weighed in with 8.8 megabyte.

core.autocrlf = true ====================

We finally addressed issue 21. For Windows projects, it seems to be the safest bet to just set core.autocrlf = true. We did that in the /etc/gitconfig we ship with msysGit.

There is a subtle problem here, and we will see how to solve it: Git's source code itself is not supposed to be checked out with core.autocrlf = true.

For new cloners, this should be easy to address (even if it has not been done yet): Git is checked out by /etc/profile, and we can set the config variable there, too.

However, existing users who pull (and do not read the msysGit mailing list regularly enough to have heard about the problem) will have to update their setup manually.

Update to a new ssh, problems with pulling msysGit ==================================================

There have been reports about hanging ssh processes, and for a long time, we had no idea how to solve it.

Until it was reported that an update to a newer MSys helps. So we updated some of MSys' libraries, and ssh.exe.

Little did we know what this would involve: one peculiarity of the Windows platform that has never been addressed properly by Git is that you cannot overwrite files that are in use. The symptoms in Git are that you check out, or merge, new revisions, but the files-in-use are not updated, and marked as dirty afterwards.

... such as the MSys libraries, when you are using the bash. While pulling msysGit, for example.

Now, a workaround is to actually install Git from our installer, then cd to the msysGit root, and perform the pull. That is a bit involved, though.

A few ideas arose, such as writing a script, or using git-gui. But a script does not seem feasible, as it would not be able to close the bash, perform the pull, and then start the bash again.

Also, "git merge" is still a shell script, so even git-gui (which runs via Tcl/Tk, and thusly does not use the MSys libraries) cannot perform the merge -- except when you call the git-gui from the Git installer.

We'll see if we can find an elegant way to tackle that problem. 

Entering Babel: "rebase" ========================

What is a "rebase"? For Git users, it means replaying commits on top of a new branch point.

Not so in Windows Speak: Windows' .dll files have fixed address ranges assigned to them, and when they overlap, problems occur. The tool "rebase" is meant to help there, by assigining non-overlapping ranges to existing .dll files.

Peter Harris -- about whose work you will likely read more in the next Heralds -- was so kind to diagnose and fix that problem, by running "rebase" himself.

Unfortunately, this tool is a closed source tool. But happily, Peter could point me to an Open Source implementation of it, and I succeeded in compiling it.

As a consequence, we have now a script to fetch, compile and install that Open Source "rebase" tool in our "full" branch, for the joy and enlightenement of interested developers.

git-cheetah is now a submodule ==============================

We readded the necessary COM infrastructure, and added git-cheetah (the hopefully-soon lookalike of TortoiseCVS for Git) as a submodule in /src/git-cheetah.

It is not checked out by default, so you will have to update the submodule as usual:

$ cd / 

$ git submodule init src/git-cheetah 

$ git submodule update src/git-cheetah 

$ cd src/git-cheetah 

$ make 

Speaking of git-cheetah: mailing list member Kirill is working pretty hard to get a context sensitive menu in place, and on separating the platform dependent parts out into their own source files.

So my dream about a git-cheetah for Dolphin, Konqueror and Finder has actually a chance to become reality soon!

Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:

It is not that much different. The basis operation is the same; I don't think that "rebase -i" differs from "rebase" more than "add -i" differs from "add".

"git pull --rebase", but it is quite new feature.

Hi,

On Sun, 2 Mar 2008, Jakub Narebski wrote:

Well, first of all, this was a personal view.

And then, I think he _has_ a point. Our interfaces are not bad, but they are not really obvious, either.

If I only were a bit more gifted in interface design...

Yes, it is. And it is not that easy for our users to find out about what new features got into Git, since there are _so many_ new features.

Ciao, Dscho

Johannes Schindelin wrote:

There are always RelNotes ;-))))

For me the sign how incredibly fast the git development is is the fact that git version from a year ago is considered "ancient".

Jakub Narebski schrieb:

Yes, and that is in itself a problem for people like me who just want to use git to get some work done. The time I spend installing new git versions, reading RelNotes and sorting through a rather high-volume mailing list goes off the time I can spare for working on the Linux driver I maintain. :-(

Tilman

Hi,

On Mon, 3 Mar 2008, Tilman Schmidt wrote:

Well, you do not _have_ to upgrade, if you are comfortable with what you have...

Ciao, Dscho

Am 03.03.2008 13:05 schrieb Johannes Schindelin:

[...]

True as far as it goes, and for appropriate values of "what I have". Neither do I _have_ to subscribe to the mailing list, or, for that matter, to read the release notes of a new version I install.

But git is not particularly easy to learn on my own, so I end up asking for help. (Arguably this qualifies as "not comfortable with what I have.") And then "ancient version" translates all too easily into "you should upgrade to a newer one".

Tilman Schmidt wrote:

I think the assumption made was that you were already getting some work done with the version of git that you had.

If you do not yet have a comfortable grasp of git usage I would suggest the following steps:

    1) Grab "a" version of git (preferably latest stable version)
	-If you need to compile it, read the "INSTALL" file.
    2) Read intro/tutorials
	-tutorial.txt
	-everyday.txt
	-user-manual.txt (more in-depth)
	These three are mentioned in the git man page.
    3) Start using it
	-refer to man pages to answer questions as you
	 encounter issues
	-ask the mailing list when man pages or other docs
	 are unable to answer your questions
    4) End of story

Upgrading, reading RelNotes, using new features, and reading the high-volume mailing list are not required, so it is wrong to suggest that they are barriers to entry for using git.

-brandon

Am 03.03.2008 19:48 schrieb Brandon Casey:

True.

True, too.

Got as far as that. In fact, I'm talking about one specific aspect of that last substep. (let's call it 3.2)

I can't see that yet.

I never meant to suggest that. They only are barriers for proceeding to step 3.2 and onwards to step 4.

Hope that clears it up a bit.

Tilman Schmidt <tilman@imap.cc> writes:

Yeah, we can stop fixing issues and enhancing features.  Would that help?

Am 03.03.2008 18:18 schrieb Junio C Hamano:

Oh dear, I didn't want to offend anyone. Of course fixing issues and enhancing features is a Good Thing(TM).

Tilman Schmidt <tilman@imap.cc> writes:

Heh, no offence taken, and sorry I forgot the obligatory smiley ;-)

But if you s/stop/slow down/ what I said, it may start to resemble a more serious question.

On Tue, Mar 4, 2008 at 11:37 AM, Junio C Hamano <gitster@pobox.com> wrote:

Given that git dev has such a frantic pace... would it make sense to give way to some "version inflation"?

This would give end users a more clear sense of how much things have changed -- a 1.4.x to 1.5.x doesn't seem like much. But a 1.5 to 2.0 with a "new features summary" will grab a bit of attention, get its slashdot article, and be a more frank communication of the work that's happened, and what the user can expect.

In other words, the 'linux versioning' scheme sucks when dealing with people who aren't sub'd to the mailing list. Yes, from git v0.99 to today we haven't broken anything too significant, but from an end user POV, several of the smaller changes carry enough incompatibility that v1.4.x and v1.5.x are not actually compatible (all the remote heads handling changes, for example).

So say rock on, but label the next feature release 2.0 or at least 1.6 and declare it is "mostly compatible, but you'll do well in re-cloning your projects to keep things simple" -- in practice, I've had to do that anyway on the 1.3->1.4->1.5 transitions.

cheers,

martin

Am 03.03.2008 23:58 schrieb Martin Langhoff:

Well, I seriously wouldn't want you to slow down fixing bugs. :-) Enhancing features is a slightly different matter, because that tends to impress upon insiders a feeling of "how could we ever have done without that?" and onwards from there to "how can anybody seriously without that?" which then easily creates the impression that the answer to every other question from simple users like me is "install the newest version, it has a nifty feature that solves your problem better than anything already present in the version you have!"

You've got a point there.

But I'd like to suggest something else still: seeing that my git mailing list folder has already grown to 363 mails again, of which probably only a small fraction will concern me as a user - would it be possible to have separate mailing lists for usage topics and for discussions of ongoing development? I imagine that might help those who just want to use git (like me) to find their way around.

Just an idea ...

Thanks, Tilman

Hi,

On Thu, 6 Mar 2008, Tilman Schmidt wrote:

AFAIAC you can have your "users-only" mailing list. Personally, I will never look at it, though, since all I am interested in is the development of Git. If that holds true for the majority of Git _developers_, it might even be a bad idea to have a separate users' list, since then

- no ideas from strictly-users would flow to the developers, and

- new developments would not reach you, and

- you would not get help by the people knowing the internals _deeply_.

FWIW I think what you wish for is already covered -- and in an instantaneous manner at that -- by the IRC channel.

Ciao, Dscho

Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:

Personally, I suspect I would end up subscribing to both, but two mailing lists would make it much more cumbersome than necessary to correlate the original user "itch" request that triggered an enhancement, the discussion that clarified the design constraints and requirements, and the patch and the review comments that lead to the final implementation, especially if you do not encourage cross posting to both lists. And of course cross posting will make user-only list more technical which would defeat the original point of having two lists.

"users-only" list could probably created by readers' MUA, by picking emails that do not have "diff --git" in its body; that would probably be a good enough approximation for people who are not interested in the technical discussions.

Perhaps adopting a convention for the subject line, like "Usage question: xxxxxxxxxxxxxxxxxxxxx"? We're still a small list, so it wouldn't be horribly cumbersome. If the list grew beyond 150 or so<http://en.wikipedia.org/wiki/Dunbar%27s_number>active users, or if the signal:noise ratio grew too low, perhaps then that would be a good time to readdress the issue.

I confess that I haven't had to deal with such things before, so I'm not familiar with the practices that may work in other groups faced with a similar issue.

On Wed, Mar 5, 2008 at 7:41 PM, Junio C Hamano <gitster@pobox.com> wrote:

Hi,

On Wed, 5 Mar 2008, Jim Raden wrote:

I fail to see how that would accomplish something different from two mailing list, except the very likely chance that people are unaware of that convention.

Recently Shawn mentioned that there are 853 members.

<tongue-in-cheek>It is pretty easy to tell that somebody top-posting is not a regular on this list</tongue-in-cheek>

Don't get me wrong, I'd _love_ to see a proper solution to the problem, alas, I do not see one yet.

Ciao, Dscho

Msysgit has apparently 60 members, at least according to http://groups.google.com/group/msysgit. It was that number to which I was referring.

I saw that this message was cross-posted to git@vger.kernel.org, which no doubt has many, many more than 60. Perhaps this is where the 853 figure comes from?

On Wed, Mar 5, 2008 at 9:29 PM, Johannes Schindelin <Johannes.Schindelin@gmx.de> wrote:

You're right, many people probably *would* be unaware of the convention on anything other than a small list, i.e. 60 members. ;)

If the list were bigger, say for instance 853, the solution I proposed would be too cumbersome.

Those 853 members are on msysgit or git@vger.kernel.org?

On Wed, Mar 5, 2008 at 9:42 PM, Jim Raden <james.raden@gmail.com> wrote:

http://vger.kernel.org/vger-lists.html#git

Oh no, only 851 members now. They're dropping like flies! :-)

j.

Obviously I must have scared them off with my suggestion!  ;-)

On Wed, Mar 5, 2008 at 11:32 PM, Jay Soffian <jaysoffian@gmail.com> wrote:

Jim Raden wrote:

Nah they switched to Mercurial. :-)

Paul Franz

On Wed, Mar 5, 2008 at 6:40 PM, Tilman Schmidt <tilman@imap.cc> wrote:

The mercurial lists are split-up this way. I hate it.

As for wanting a git-users list, just try filtering out everything from this list with PATCH in the subject and you'll have a close-enough proxy, IMO.

j.