Re: Review Request 42688: Fixed equality semantics for Labels.

[Neil Conway <ne...@gmail.com>]

-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/42688/#review119248
-----------------------------------------------------------



Per comments in MESOS-4445, the current plan is to not change equality behavior right now, and just to document that duplicates in `Labels` are not supported.

- Neil Conway


On Jan. 23, 2016, 8:04 p.m., Neil Conway wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/42688/
> -----------------------------------------------------------
> 
> (Updated Jan. 23, 2016, 8:04 p.m.)
> 
> 
> Review request for mesos and Michael Park.
> 
> 
> Bugs: MESOS-4445
>     https://issues.apache.org/jira/browse/MESOS-4445
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> The previous implementation was buggy in two scenarios: (1) the label sets
> contained different #s of duplicate elements (2) the "left" label set contained
> duplicates and the "right" label set contained a value that didn't appear in the
> "left" set.
> 
> 
> Diffs
> -----
> 
>   src/Makefile.am 19bf3a7c2e43ca04ed6e6d506e052de5537f7c2f 
>   src/common/type_utils.cpp 76f48f6a1f5467db032ded8acd296d03353b4172 
>   src/tests/labels_tests.cpp PRE-CREATION 
> 
> Diff: https://reviews.apache.org/r/42688/diff/
> 
> 
> Testing
> -------
> 
> make check
> 
> REVIEW NOTES:
> 
> * This is a WIP.
> * Rather than implement a correct equality algorithm myself, I decided to just use `std::unordered_multiset`. Computing equality correctly is a bit involved.
> * This requires a hash implementation for `Label`. I supplied this in new function object in the `mesos` namespace, but we could also supply a specialization of `std::hash<Label>`. Not sure which is better.
> * The same bug occurs in other equality operators for repeated fields (`CommandInfo.uris`, `Environment`, `DockerInfo`, etc.). I suppose we should probably fix these as well? Note that the bugs in the old approach are a lot less likely to occur when the struct has a lot of fields.
> 
> 
> Thanks,
> 
> Neil Conway
> 
>