What actually happened
"Eviction" and "ban" make good headlines, but the record describes something narrower. Torvalds started developing Git in April 2005 after the free licence for BitKeeper, the proprietary source-control system used for kernel development since 2002, was revoked for Linux. BitMover pulled a free-of-charge arrangement. It did not throw people out of anywhere.
The background matters too. The Pro Git book says that during the early years of kernel maintenance (1991–2002), changes were passed around as patches and archived files, and that the kernel project began using a proprietary distributed version control system, BitKeeper, in 2002. In other words, the email-and-tarball era was real. It ended when BitKeeper arrived, not when Git did.
The Tridgell dispute is contested
Samba developer Andrew Tridgell wrote a tool called SourcePuller. Reports of the time describe the dispute as a collision between two views of who owned the metadata behind code submissions. The Register reported in April 2005 that kernel developers insisted the metadata generated by their submissions belonged to the community, while Larry McVoy insisted it belonged to him. The Register is paraphrased here.
Each side's account should be attributed rather than taken as settled:
- BitMover's position: McVoy claimed that Tridgell had created SourcePuller by reverse engineering the BitKeeper protocols.
- Tridgell's position: He said that, having never bought or owned BitKeeper, he had never agreed to its licence and so could not violate it. He also said his analysis began with typing "help" into a telnet session to a BitKeeper server.
- Tridgell's stated aims: In his README, he described SourcePuller as an interoperability tool. It was not a full replacement for BitKeeper and was missing a large amount of core functionality.
So the neat claim that one developer "broke the deal" is BitMover's framing. Tridgell disputed it. The Register also noted that he said losing the free licence was not the outcome he intended.
The Git timeline
Git's first commit is dated Thursday 7 April 2005 in the public repository history. It is titled "Initial revision of 'git', the information manager from hell". That is the verifiable anchor.
Some popular claims need softening:
- "11 days to build Git." The initial commit and the later self-hosting milestone are real, but this does not mean Git was a polished replacement within 11 days. Early Git was a small set of low-level commands. One guide notes that it started out very simple, with core commands that needed to be combined to complete a task.
- "Home of the Linux kernel in June 2005." This is consistent with the kernel.org archive, which carries Linux 2.6.12 and Git-based incremental snapshots from June 2005. The last pre-Git kernel release candidate, per Graphite's account, was 2.6.12-rc2 on April 3, 2005.
- The date of the licence withdrawal. The XDA piece puts it on April 5. Other accounts give slightly different dates for the surrounding events, and Torvalds' public "Kernel SCM saga" email is dated April 6 per Graphite. Treat the exact day as approximate.
Why Git looked the way it did
Torvalds had specific requirements. Pro Git lists the goals as speed, simple design, strong support for non-linear development with thousands of parallel branches, a fully distributed model, and the ability to handle large projects like the Linux kernel efficiently. Wikipedia's Git entry adds that Torvalds wanted a distributed system he could use like BitKeeper, and that none of the free systems available met his needs.
A distributed design means each developer has a full copy of the project history. Most everyday operations, such as browsing history, work locally without contacting a server. This is why Git feels fast, and why it fits kernel-scale work with thousands of contributors.
Git is not GitHub
The XDA article is right that GitHub came later, but the two are easy to conflate:
- Git is the version-control system, created in 2005 for the kernel.
- GitHub is a hosting and collaboration service built on Git repositories.
GitHub's own posts give the dates. Chris Wanstrath wrote on April 10, 2008 that GitHub was officially live, adding comment features, network graphs and Campfire and Lighthouse integrations. In a January 2009 retrospective, Scott Chacon said the site launched in beta almost exactly a year earlier from code Tom and Chris wrote as a side project.
The October 2008 anniversary post names Tom Preston-Werner, Chris Wanstrath, PJ Hyett and Scott Chacon as the four full-time employees. It also says the site had just topped 20,000 public repositories. That supports the four names in the XDA piece. It also shows the team came together over time, as Chacon joined later as a contractor and then full time.
GitHub's posts document its own launch and growth. They do not show a direct causal line from the BitKeeper dispute to GitHub's creation. A fair reading is a sequence of related developments:
- BitMover withdraws the free arrangement.
- The kernel community builds Git.
- Git becomes the kernel's tool.
- Years later, GitHub makes Git projects easier to host and discover.
The wider ripple
Git was not the only tool the dispute produced. Mercurial was announced on 19 April 2005, and Wikipedia says the impetus was BitMover's withdrawal of the free BitKeeper version. Two distributed version-control systems arriving within weeks of each other suggests the dispute exposed a real gap in open tooling. It was not simply one engineer's temper.
Wikipedia also reports that end of support for the "Free Use" BitKeeper version was officially July 1, 2005, so there was a deadline pushing the migration.
Why Windows and enterprise readers should care
Git is now part of daily Windows development. Pro Git's appendices cover Git in Visual Studio, Visual Studio Code and PowerShell. For IT admins and developers, the history explains a few design choices: why Git is local-first, why branching is cheap, and why history is treated as a core asset.
The practical takeaway is about tooling risk, and it is general industry reasoning rather than a claim from the sources above. Kernel developers built on a proprietary tool under conditional terms, and lost access when the relationship soured. Teams today depending on a vendor's free tier or hosted service face a similar, if less dramatic, exposure. Git's distributed design softens that risk, because every clone carries the full history.
Bottom line
The accurate version of the story is less cinematic than "a CEO evicted Linux". BitMover ended a free-of-charge arrangement amid a disputed claim about reverse engineering. Torvalds responded by writing a purpose-built distributed system, with its first commit on April 7, 2005. Git took over kernel development by mid-2005, and GitHub arrived three years later to help bring it to everyone else.