Geometry Boost List: Re: [geometry] Status report

Subject: Re: [geometry] Status report
From: Adam Wulkiewicz (adam.wulkiewicz_at_[hidden])
Date: 2014-11-03 13:28:14

Mateusz Loskot wrote:
> Hi,
>
> Last year, I generated modified version of the support_status report:
> https://gist.github.com/mloskot/5821937
>
> I've just generated the new reports against the current master:
> https://gist.github.com/mloskot/993c38232c88675488e8
>
> Apart from the great progress, I noticed there seem to be a regression (?)
> regarding the disjoint algorithm.
>
> The disjoint support used to have 100% coverage, but now there are gaps
> for some of the combinations.
>
> How to interpret it?

Are you sure that disjoint had 100% coverage in the past?
Which version of Boost do you have in mind?
Is this assessment based on real usage of the library or just on the 
analysis of the results of the support_status program?

According to the docs the combinations of parameters with MultiPoint 
wasn't supported in 1.56:
http://www.boost.org/doc/libs/1_56_0/libs/geometry/doc/html/geometry/reference/algorithms/disjoint.html

There is no supported geometries matrix in the docs for 1.55:
http://www.boost.org/doc/libs/1_55_0/libs/geometry/doc/html/geometry/reference/algorithms/disjoint.html
but I checked out this version and e.g. Point/MultiPoint nor 
MultiPoint/Point doesn't compile.

The same with 1.54. I didn't check earlier versions.

In the older versions (pre 1.56), instead of failing in the dispatched 
algorithm as it should it failed in some other algorithm e.g. 
for_each_xxx, sectionalize, etc. This is probably why it wasn't detected 
by the support_status since it only checks if the dispatched algorithm 
is derived from not_implemented<> and in the older versions (pre 1.56) 
it wasn't. So now it fails to compile in a "right" way.

Though indeed this change could cause a regression. However we can't say 
that after the analysis of the support_status results since it gave 
invalid results for older versions of the library. I only checked the 
Pt/Mpt combination. It's probable that for other combinations those 
internal algorithms failing e.g. for Pt/Mpt compiled before. So this 
indeed would be a regression. Do you know about such case?

Regards,
Adam