You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
From discussions with Alex, I realized that our current strategy of storing these sequential keys as {[k1 k2] ...} is incorrect. The reason is that we want to be able to generate proposals for nested sequences of choicemaps without having to copy the whole nested structure in our proposal generators.
So tracing to k2 internal to a trace to k1 should result in the same stored values as tracing to [k1 k2].
And if you do the latter, you should NOT be able to do the former afterward (since the k1 address is already taken). This is a touch subtle, because traces to [k1 k2], [k1 k3] etc at the same level need to succeed, even though k1 is already written to at when [k1 k3] comes along.
The text was updated successfully, but these errors were encountered:
From discussions with Alex, I realized that our current strategy of storing these sequential keys as
{[k1 k2] ...}
is incorrect. The reason is that we want to be able to generate proposals for nested sequences of choicemaps without having to copy the whole nested structure in our proposal generators.So tracing to
k2
internal to a trace tok1
should result in the same stored values as tracing to[k1 k2]
.And if you do the latter, you should NOT be able to do the former afterward (since the
k1
address is already taken). This is a touch subtle, because traces to[k1 k2]
,[k1 k3]
etc at the same level need to succeed, even thoughk1
is already written to at when[k1 k3]
comes along.The text was updated successfully, but these errors were encountered: