vcc: Internal type for the reserved 'default' symbol

[dridi]

A new DEFAULT type identifies symbols called 'default' and types that can have a default symbol carry a pseudo symbol. This approach can reconcile the overloaded default symbols that all share the generic type DEFAULT but carry their respective kinds.

This helps remove some of the special casing for default probes and backends whilst offering a sentinel DEFAULT type to safely deal with the special casing where it is actually needed, simplifying the code a wee bit in those areas.

Fixes #4177

[nigoroll]

I appreciate the fix, but as a vcc-noob I would be like to learn why a DEFAULT type is the better option to having multiple symbols named default with different types.
Do we not create a mess with type methods etc?

[dridi]

I appreciate the fix, but as a vcc-noob I would be like to learn why a DEFAULT type is the better option to having multiple symbols named default with different types.

When I started using Varnish it was perfectly legal to have multiple symbols of the same type, see for example 9962996 that introduced coverage for a corner case. The test case contains 3 distinct debug symbols.

It was a deliberate choice to only allow one symbol for a given name in bd30219 and as you can see from the explanation in the commit message that is the turning point where symbols became unique. The default symbol and its special meaning turned into a pretend BACKEND symbol that was either half wrong or half correct.

The introduction of a DEFAULT type here makes it completely wrong or completely correct.

As a side note on symbol uniqueness modulus default, another change happened afterwards where symbols became unique per namespace in #3344. A perfect segue into the next point you raise.

Do we not create a mess with type methods etc?

Yes but not really. Because default.method() becomes ambiguous depending on the underlying type, I'm more than willing to disallow calling methods or attributes directly from the default symbol.

[bsdphk]

I think I'm OK with this handling of it.

[nigoroll]

I'm more than willing to disallow calling methods or attributes directly from the default symbol.

Not sure if this is a good idea, but under all circumstances something like

set bereq.backend = default;
set bereq.http.healthy = bereq.backend.healthy()

should continue to work.

[dridi]

This is fine because bereq.backend is a different symbol of the actual BACKEND type. This change does not affect other symbols of the types BACKEND and PROBE, but only the actual default pseudo-symbol.

[nigoroll]

I think this can be merged, phk agreed already

[dridi]

As long as there were questions or remarks I would still consider the review to be in progress. Between here and #4177 you raised interesting points.

[nigoroll]

late correction:

have multiple symbols of the same type

I think you meant "multiple symbols of the name name (and different types)" and also read this, even though this is not was was written.