MSysGit:MSysGitHerald10

From Git SCM Wiki
Jump to: navigation, search

Contents

Good morning git land!

A warm and sunny Monday afternoon is as good an occasion as any to offer to you the 10th copy of the msysGit Herald, the quite irregular news letter to keep you informed about msysGit, the effort to bring one of the most powerful Source Code Management systems to the Operating System known as Windows.

These are the covered topics:

  • Upcoming branch layout of 4msysgit
  • Using pthreads
  • OpenSSL & cURL
  • Subversion 1.6.3
  • Moving Perl to MinGW32
  • Fixes to the "full installer"
  • Interview with Erik Faye-Lund


If things had continued as this developer expected, this issue would not be due for another year. As things turned out, though, there appeared some very interested and capable developers out of the blue, and msysGit is moving again!


Upcoming branch layout of 4msysgit

Since the last Herald, there have been quite a few infrastructure changes in msysGit, such as making it detectable if your Git runs in msysGit or in an installed Git for Windows, or improvements in the installer-making procedure (see also the story "Fixes to the 'full installer'" in this Herald).

But there has been also quite some work, primarily by Johannes Sixt and Steffen Prohaska, to minimize the differences between git.git and 4msysgit.git. Soon, we will be able to follow git.git much more closely than now.

So the plan is to follow git.git's 'maint', 'master' and 'next' branches, and rebase to git.git's branches whenever a new Git version comes out.

The temporary branches will still be called 'work/<branch>' for branches that are rebased often, and '<nick>/<branch>' for branches that a certain developer wants to work on alone.

Whenever something was merged into 'next', the corresponding work branch will be removed.


Using pthreads

As more and more Windows machines have multiple cores or even CPUs, we now have our own pthreads library, to make use of the machine's capabilities when compressing the database.


OpenSSL & cURL

Many people think our issue tracker is something like a list for Santa Clause or something. Many even just dump their (sometimtes unreasonable) ideas there and never come back. I made it a habit of being joyful when clicking on the "Invalid" link to move such issues out of the list of open issues.

One of these wishes was not too unreasonable, but I knew that it would not happen: in my experience, Windows users are very happy to ask for things, but they almost feel insulted when you ask them for help.

But I was wrong: the wish was to have an SSL-aware version of cURL, to be able to clone repositories from https:// URLs. And one guy, an honest-to-God Windows lover even, volunteered to work on this issue (see also the interview at the end).

It was not smooth sailing, but Erik continued to work on the issue, and with a little cooperation, we got something to compile.

The result is not only the addition of two new, shiny .dll files, but two shell scripts ("recipes") that can download, compile, install and commit the files. This helps not only with upgrading to newer versions, but also with upgrading our Subversion (see next story).


Subversion 1.6.3

Similarly to OpenSSL & cURL, there was a wish on the issue tracker that we should upgrade our Subversion, which is quite old by now.

However, I vividly remember how difficult it was to get the initial git-svn support working (albeit very slowly), and how many people and their enthusiastic effort it took (and I also remember how some people just complained about the slowness of the result, which really took almost all the fun out of msysGit for me, shame on them).

To my pleasant surprise, again someone came along and worked on the issue: Laurent Boulard.

As I feared, upgrading Subversion was not a simple task: its neon library now depends on OpenSSL. Unlike the OpenSSL we compiled successfully for MinGW, our Subversion is an MSys program, and therefore we needed an MSys version of OpenSSL.

Remember: MinGW is really a very thin compile-time layer over the Microsoft Runtime; MinGW programs are therefore real Windows programs, with no concept of Unix-style paths or POSIX niceties such as a fork() call.

MSys, in contrast, is a slimmed-down version of Cygwin (an old version at that), whose only purpose is to provide enough of a POSIX layer to run a bash.

It appeared easier to us to get something like git-svn, which really expects a POSIX environment, going in the MSys environment.

Here, the efforts not only to make OpenSSL compile in msysGit, but to make the result a script which can do it, really came in handy.

So now we are closing in on having a functional Subversion 1.6.3 in msysGit.


Moving Perl to MinGW32

Just when I thought that I was surprised enough by the arrival of new msysGit contributors, another guy names Bosko Ivanisevic came along and asked how to fix his Subversion build. It became quite apparent that he tried to compile it as a MinGW program.

I thought it might make sense to try to get git-svn to run in a MinGW setting (i.e. OpenSSL, Perl and Subversion all as MinGW programs), and Bosko agreed. I fully expect the performance issues of git-svn to be helped by that move (although I think that some of those who will benefit from it do not deserve it at all).

This project is in full action right as we speak.


Fixes to the "full installer"

According to our download site, the msysGit "full" installer is more than three times more popular than the "net" installer.

Just a quick reminder: "msysGit" is the build environment for working on "Git for Windows". The "msysGit" installer comes in two flavors: "net" and "full".

The "net" installer is really just a stripped-down "Git for Windows" that just clones our Git repositories, builds and installs Git.

The "full" installer is a stand-alone self-extracting 7z archive that brings everything needed to compile a particular Git for Windows version. It has no connection to our Git repositories, though.

To make a "full" installed msysGit environment updatable (by setting up Git remotes and fetching the branches), we provide the script /share/msysGit/initialize.sh.

Now, this would be all well and fine, if that script was not broken. As it happens after some rather obvious fixes, the script is no longer broken!


Interview with Erik Faye-Lund

I think Erik is the first msysGit contributor I "met" via an issue tracker. He was relatively active answering questions, also on the msysGit mailing list, before he dived into the adventure of compiling OpenSSL and cURL (see the corresponding story of this Herald issue).

He agreed to answer my Terrible Ten, as long as I was fine with his sometimes not-too-serious style of answering ;) -- which of course I am.

So here it goes:

1) How did you get involved with Git? >

I was looking for something better than SVN for a long time. The reason was mostly distribution, as I do a lot of work offline. I tried SVK, and I hated it. I tried Mercurial, and it was a lot better than SVK, but still not quite there. I can't really put my finger on it, but it just didn't work for me.

I still prefered SVN over Mercurial. I doubt I would now, though. I'm too much in love with distributed work-flows.

I already knew about Git, but had early dismissed it because I had heard it was difficult to use (read: "ZOMG, Linux zealotry!!"). I wanted a VCS that didn't require a lot of time of getting into. Luckily, I overcame my skepticism, and gave it a shot. And I must admit, I was amazed. Not only was it really really easy to use (as opposed to what some people are saying), pretty much everything about it felt right.

I immediately fell in love with the staging concept. In fact, it was something I already had kind of "mentally built" myself in SVN (which was pain to do), and I loved that it was so accessible.

2) What were the reasons that you started working on Git? >

I'm in it for the fame, glory and women... or whatever.

Actually, I don't have a big reason other than wanting to make the world a better place (blah blah). I was bored one day, and I saw an issue posted on the mailing list that made me think "hey, I'm a programmer, I can solve this". And I did. After that, the ball just kinda started rolling.

3) What do you like most in Git? >

The development community. I think it's really focused and generally working in a constructive manner.

About git itself I'd go with the "obvious" answers; distribution, painless branching, stash, the staging area... Oh, and the performance, of course.

4) What do you hate most in Git? >

That Windows is still a second class citizen. But let's try to fix that one, huh? :)

5) What was the most surprising moment when working with Git? >

I guess when I figured out that I could just create a local repo for basically any project, and push it out somewhere if it turned into something useful. That ad-hoc way of making repos is not something I'm used to from SVN.

6) What was the most frustrating moment when working with Git? >

GWhenever I fight with line-endings. The fact that core.autocrlf=true doesn't work too well in git-svn, combined with visual studio's lack of proper newline-support causes me a lot of annoyance at my dayjob.

That's one thing that actually still SVN is better at, and it makes me a bit sad.

7) In what environment do you work? >

Windows Vista 64bit. I'm a Windows guy. Sure, go ahead an hate me -- I like Windows.

8) What other hobbies do you have? >

I murder people who ask too many questions.

No, seriously, I'm an active demo programmer. You can read more about it on http://en.wikipedia.org/wiki/Demoscene if you don't know what demos are.

Back when I was younger and in better shape I did a lot of skateboarding. I sometimes still pretend that I'm up to i, but I always end up disproving myself. But it's fun :)

9) What is your favourite movie? >

Hmm. That's a very tricky question, I don't watch that many movies these days. Can I answer with a book? If so, "Shantaram" by Gregory David Roberts. I'm sure the movie will be awesome, I heard Johnny Depp is playing the lead role.

10) What are your visions for Git? (I.e. where do you want it to go?) >

I'd like to see the Windows version being as good as the *nix-versions. Also, I don't think we should settle for less than world domination. Giant lazers on the moon? I can't see why not.

Personal tools