Decide on handling path starting with / on *nix

[davidfowl]

It's a broken assumption since absolute paths on *nix can start with /

[Eilon]

@davidfowl we talked in theory about swapping these two checks: https://github.com/aspnet/FileSystem/blob/dev/src/Microsoft.AspNet.FileProviders.Physical/PhysicalFileProvider.cs#L85-L94

However, I don't think that's right either. The real problem is that on Windows it's easy to tell the difference between a volume-rooted path (C:\...) and an absolute path (\foo\bar), and on *nix there's no way to differentiate.

Honestly not quite sure how we could fix this even in theory...

[davidfowl]

I think we have to drop this behavior completely.

[cesarblum]

This somehow affects aspnet/Universe#304. I had to disable a unit test on *nix which ideally should be run on that platform. However the test gets incorrect results because of the way "/" is being treated.

[muratg]

Let's reach a decision here. @davidfowl do you suggest we remove support for relative paths?

Can we just enforce a . in front of relative paths? (i.e. ./foo/bar/ instead of /foo/bar) to differentiate it from absolute paths?

[davidfowl]

@davidfowl do you suggest we remove support for relative paths?

No.

Can we just enforce a . in front of relative paths? (i.e. ./foo/bar/ instead of /foo/bar) to differentiate it from absolute paths?

That's ugly. We can just not support paths that begin with /. Relative paths should just start with nothing. .e.g foo/bar.

[muratg]

That sounds good to me.

[muratg]

@pakrym this also prevents CoreCLR x-plat test runs. Please look into this after your current item.

[pranavkm]

It's a broken assumption since absolute paths on *nix can start with /

Doesn't the IFileProvider root you at the rooted path? It's not meant to be a general purpose representation of the native file system.

[davidfowl]

It's about what we allow in the methods. We want to fail for absolute paths.

[muratg]

So do we fail on Windows for paths like C:\foo\bar as well?

[davidfowl]

Yes I believe we already do

[pakrym]

Decided that current beheviour is okay, just changing supress message.
#150

[muratg]

Cool, please update the bug title to reflect the decision before checking in.