Re: [Tcllib-devel] Regarding SourceForge, the Tcllib CVS, and transitioning somewhere.

[Lars H. <Lar...@re...>]

Will Duquette skrev 2011-02-15 01.32:
> Lars,
>
> Provable immutability is one of Fossil's requirements; in the development
> arena that Richard was (and might still be) targetting, it was required that
> a file with a given revision UUID could simply not be changed.

But that's true for git too! The rebase command *cannot* change the data in 
an existing commit. It creates a new commit, which frequently has the same 
contents (tree) as the original, but since it has a different history it will 
also have a different ID.

> You can read
> more about that at the fossil website. This is not a requirement for most
> users. In short, I don't think Richard is saying that git is substandard in
> this area, but rather that Fossil provides a stronger form of immutability if
> you happen to need that.

The point where I fear Richard is making an unfounded claim is rather the 
point before, where he says "Single file database (fossil): good, safe. Pile 
of files on disk like git: bad, unsafe." The thing he worries about seems to 
be corruption of data due to interrupted updates of it (and as a database 
expert, it is his job to be concerned about this). However, because of the 
immutability of objects (blobs, trees, and commits) in a git repository, the 
files which store these /aren't/ updated; "modified" contents will, by virtue 
of being different, receive a different ID and therefore end up in a 
different file.

The whole thing is actually quite similar to the way Tcl handles values and 
variables. Commits, trees, and blobs are immutable like a value (indeed, this 
is very much because EIAS), while "branch heads" and tags are mutable like a 
variable. You can change what value is stored in a variable, but you cannot 
alter the value itself.

Lars Hellström