[PATCH] Some infrastructure work for virtual file system

[Rafael Espíndola]

On 11 February 2014 05:16, Manuel Klimek <klimek at google.com> wrote:
> Very cool. A few high level questions:
> - shouldn't the virtual file system go into llvm, not clang?

Not until we have a concrete need for one. In another tool.

> - seems to me like there is only one AbstractFileSystem that should be owned
> pretty top-level, so why use an IntrusiveRefCntPtr?
> - why not make all the function virtual? Doesn't seem to have real
> downsides, and would immediately enable fully virtualized file system
> implementations (seems fine to have default implementations for all the
> functions); if the plan is to have the functions state now to provide an
> easier migration path, I think it'd make sense to add a comment describing
> the end-state...

I disagree with having unused features. We should really have only the
functions that are needed and make virtual only the ones that actually
need to be overridden.

Cheers,
Rafael