Placement protocol should allow failure in place creation.

[rphmeier]

The only way to indicate failure in in-place allocation is to have your InPlace implementer return a Result from finalize(). However, most failures in in-place allocation will occur when there is insufficient space in the Placer to make a place. The implementation of Placer for ExchangeHeapSingleton in std sidesteps this issue by panicking when there isn't sufficient heap space, but that is hardly an acceptable solution for every implementation of in-place allocation.

Error handling for in-place allocation could become a little weird if every allocation can fail unexpectedly.

[pnkfelix]

(We recently discussed this amongst the lang team.)

My standard response is: Why not in try!(alloc_place()) { value_expr } ?

(Or in the more recent proposed syntax, try!(alloc_place()) <- value_expr ?)

[huonw]

After we discussed on IRC (to late to bring up, unfortunately!), I realised this was encodable in the manner @pnkfelix said. See "Fallible Placement" for more details and a working example.

I'm inclined to close this issue, but maybe it could cover improving the support/smoothing out some of the strangeness.

[rphmeier]

That's the solution I came up with for my Allocator trait, and it seems to work well enough. My only issue with it is one that you already raised in your Fallible Placement section: you have to do the allocation in the actual creation of the Placer, which makes the various trait implementations feel redundant.

[Centril]

This is moot now.