[fix] overlay threshold for sort_by_side #1189
Conversation
There was a problem hiding this comment.
Thanks!
It would help a lot if you comment on the geometric intuition of the value 100 that is used as threshold.
Regarding #1183 is there some other combination of values that fixes that or you believe it is of different nature?
| @@ -325,7 +325,7 @@ public : | |||
| double | |||
| >::type; | |||
|
|
|||
| ct_type const tolerance = 1000000000; | |||
| static auto const tolerance = common_approximately_equals_epsilon<ct_type>::value(); | |||
There was a problem hiding this comment.
this is the actual change right?
There was a problem hiding this comment.
I'll document the 100 later. It's not new, it's moved.
It is recently (earlier this year) adapted from a larger value to this value, considering another issue and all current tests.
It is used for clustering turns, and now also for sort by side. In both cases the exact value is not that important, but it should not be 1 (for floating point comparisons) and not be too large (obviously). The tolerance which was now replaced was very large, I don't remember exactly why.
There was a problem hiding this comment.
See also #1108
Fixes: issue boostorg#1186
barendgehrels
force-pushed
the
fix/issue_1186_epsilon
branch
from
0d5f1a0
to
1a0c287
Compare
Fixes: issue #1186