From: cross@spitfire.i.gajendra.net
In article <10o9jdr$2v1fr$2@dont-email.me>,
Arne Vajhøj wrote:
>On 3/4/2026 10:01 AM, Dan Cross wrote:
>> I used to work with one of the SVN developers. When he joined
>> our team, he apologized. :-D
>>
>> But seriously, there's nothing inherently wrong with subversion.
>> It was a decent implementation, but the world passed it by while
>> it was still in incubation.
>>
>> Torvalds famously hates it, but not all projects fit into the
>> `git` model of shipping the entire repository around. At the
>> time he presented at Google, for example, google3 contained
>> about a terabyte of source code: it would not be reasonable to
>> ship that to every engineer's workstation and keep it
>> synchronized. A monorepo just has different requirements than a
>> relatively small project like the Linux kernel. The Linux
>> people have a real tendency to overstate the size and scope of
>> their project.
>
>I have no practical Git experience.
>
>But aren't people supposed to use partial clone
>and sparse checkout to deal with the size problem
>of monorepos?
That all came _much_ later; he presented Git at Google in the
very early days. Indeed, much of that work was done at the
hyperscalars because their engineers wanted to use git to manage
their monorepos.
The effort was only marginally successful. Most never migrated
succesfully; Google did its own internal Perforce replacement;
Meta did its own thing; MSFT has their own tools; I don't
actually know what Amazon uses, but I suspect they've got a
similar story. The early momentum building around Jujutsu is a
direct result of the hyperscalar experiences (jj is essentially
a rewritten and better open source version of Google's `fig`,
which sits on top of CitC, SrcFS, and interacts with the
builder clusters and ObjFS).
- Dan C.
--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)
|
