[petsc-dev] uninformative docs for DMPlexGetStratumIS and DMPlexGetStratumSize

[Geoffrey Irving]

On Wed, Nov 20, 2013 at 3:49 PM, Matthew Knepley <knepley at gmail.com> wrote:
> On Tue, Nov 19, 2013 at 4:35 PM, Geoffrey Irving <irving at naml.us> wrote:
>>
>> The documentation for DMPlexGetStratumIS helpfully explains that the
>> "value" argument is
>>
>>     - value - The stratum value
>
> This is exactly right. A DMLabel is a map from value --> {set of integers}.
> There are no builtin semantics on the values.
>
>> One has to look in DMPlexStratify to find that the value is probably
>> (the word "value" isn't actually used) either the dimension or the
>> codimension.  But not which!
>
> DMPlexStratify() builds a DMLabel called "depth". The values in this label
> are point depths in the mesh graph.

Can I fix the documentation for DMPlexStatify to say this?  Currently
it says it "might" do something like that:

    "The normal association for the point grade is element dimension
(or co-dimension). For instance, all vertices would
  have depth 0, and all edges depth 1. Likewise, all cells heights
would have height 0, and all faces height 1."

>> The documentation for DMPlexStratify seems to imply that the meaning
>> of the stratum value is up to the user.  Is that right?  If so it
>> makes it hard to design robust libraries; the library would have to
>> either know how the DMPlex was created or reverse engineer the
>> dimension 0 stratum from the coordinate section (or similar
>> hideousness).
>>
>> If there is a convention, can I add it to the documentation of
>> DMPlexGetStratumIS and DMPlexGetStratumSize?
>
> These are documenting the generic DMLabel interface. These functions are
> in fact just the delegator pattern, where they dispatch to the named label. I
> am considering just eliminating all of them in favor of DMGetLabel().

Thanks for the explanations.  Is there a reason why you use the word
"Stratum" here but "Value" elsewhere in the label interface, such as
DMPlexSetLabelValue?  Stripping the extra layer of interfaces sounds
great to me.

Geoffrey