bind_mount: Return an error code, and provide a way to display it

[smcv]

• build: Error if a switch lacks a default case

  This is a useful way to detect bugs before they happen.

• Handle all enum values in switch

  This makes it clearer that we have thought about whether they need
  symlinks for source resolved, and come to the conclusion that they do not.

• build: Warn if a switch over an enum does not handle all values

  This is a useful way to catch possibilities being unhandled.

• bind_mount: Return an error code, and provide a way to display it

  This gives us better diagnostic messages on failure, particularly for
  BIND_MOUNT_ERROR_FIND_DEST_MOUNT where we previously said "Invalid
  argument".

• bind-mount: Factor out bind_mount_result_to_string()

[smcv]

Actually, it might be better to defer this until after #403 is in, because #403 adds more failure modes that should ideally be new enumeration values.

[smcv]

Now updated after merging #403.

[cgwalters]

LGTM as is - please consider my comments as optionals. You can merge as is or do a followup or not.

[cgwalters]

I think this would be more elegant with a helper method equivalent of strerror() - i.e. it returns a const char * instead of doing the printing itself. (Or analogus to impl Display for BindError in Rust).

[smcv]

One problem with doing this is that if we get an unknown/invalid bind_mount_error , we can't fall back to printing it with %d, unless the helper method returns a newly-allocated char * (or a static buffer, I suppose, but eww).

[smcv]

New version adds bind_mount_result_to_string(), does that look better to you?

[cgwalters]

Though I guess this is a wrinkle so it'd need to return two values. (Or...maybe just reset errno = 0 and don't print it if zero?)

[smcv]

I'm already not a fan of the "action at a distance" API that is errno, so having helper functions edit errno as a side-effect seems nasty to me.

I was half-tempted to reinvent the GError/DBusError pattern here, like I did in libcapsule, but held off on doing that right now.

[smcv]

I went for "return two values".

[smcv]

LGTM as is - please consider my comments as optionals

I think I'd like to land this as-is to get the better reporting, and then try refactoring as a followup and see whether it looks like a net improvement.

[smcv]

I think I'd like to land this as-is to get the better reporting, and then try refactoring as a followup and see whether it looks like a net improvement.

Hmm, actually, I should redo this, because -Wswitch-enum is provoking new warnings.