KeyManager: Perform all key verification during New* functions about trillian HOT 6 CLOSED

Probably OK as long as you can guarantee it won't break with more realistic implementations.

Also it's potentially losing the detailed information in the error as to why the key was invalid if done via New() .

from trillian.

I don't think it will break realistic implementations.
I expect that key material should be available on server start and errors should be caught fairly early. Also, if we need to distinguish between multiple error cases in New we can define specific errors.

from trillian.

Not sure. Constructors really shouldn't be doing significant work (like reading files) and having constructors that can fail is usually an anti-pattern.

What might work is separating out all the code that decodes keys and having the PEM key manager constructor just take the keys. I don't know if this is better without trying it.

from trillian.

from trillian.

FYI, the one example of a constructor in Effective Go is one that can fail:

https://golang.org/doc/effective_go.html#composite_literals

The pattern of "constructors shouldn't be doing significant work" is mostly for languages like C++, where they can get invoked implicitly, but just about nothing is implicit in Go, so no problem, significant work and failure are totally fine, especially if it can reduce the state space of the returned object!

from trillian.

I sometimes think Go was designed purely to contradict every other existing language.

from trillian.