Allow ValueSets to be constructed with slices in addition to arrays
#2787
Conversation
|
I'm not entirely sure why CI is failing, but I don't think it's related to my change. |
|
This change makes sense to me. The CI failure is due to a yanked dependency breaking our build with |
maddiemort
force-pushed
the
valid-len-slice
branch
from
3457924 to
068fd55
Compare
|
Is there anything I can do to help get this merged? |
|
Chiming in to show support for this addition, having this |
There used to be a length restriction on the arrays usable to construct `ValueSet`s, but this was due to Rust compiler limitations that no longer exist. After that restriction was removed, the `ValidLen` trait was kept around but implemented for statically-sized arrays of all lengths. While in theory this allows `ValueSet`s to be constructed from arrays of arbitrary length, the size of the arrays must still be known at compile time. Now that there is no longer a restriction on the length of arrays, slices could in theory be used as well (since they meet the supertrait requirement of `ValidLen`), but no implementation of the trait exists. Adding an implementation of `ValidLen` for slices is straightforward. Refs: tokio-rs#2508
maddiemort
force-pushed
the
valid-len-slice
branch
from
068fd55 to
31d754f
Compare
|
Chiming in to ask for this as well. The macro limit can be worked around with |
4 participants
Motivation
I'm working on an implementation of
Subscriberthat needs to be able to store (in a form as close as possible to the original) the inputs to its methods, and reconstruct them later to be passed to anotherSubscriber. For the most part, this works really well, but the biggest sticking point by far came from trying to reconstruct aValueSetfrom a stored list of(Field, Value)pairs.The core of the problem is the
ValidLentrait and its usage byFieldSet::value_set(). In order to call that function, you have to be able to provide a statically-sized array, because that's the only type that implementsValidLen.While, in theory, constructing an array for my use case should be just about possible (although the tricks necessary to do so likely involve a macro and the reimposition of an arbitrary length limit), lifetime issues (that appear to result from a call to
<[_; N]>::try_from()) make it impossible. I'm going to construct a minimal example of this and investigate further at some point, but in the meantime:Solution
Since #2508 removed the limitation on array length here, there doesn't seem to be a drawback to simply adding an implementation of
ValidLenfor slices. They already meet theBorrowsupertrait requirement, so no additional machinery is needed.This makes this API trivially easy to call for my use case, and I can't think of any associated drawbacks.