Those all look like rounding errors (probably due to x87 excess
precision, since they only happen on i386). This suggests that the >appropriate response is either to ignore these tests on i386, or
possibly (the graph object output by gdf_to_nx has (x: float, y:
float) node keys, so rounding issues can potentially break retrieving >individual nodes/edges) to remove this package on i386.
Those all look like rounding errors (probably due to x87 excessYour assessment of the situation is very likely correct, since the errors are very small.
precision, since they only happen on i386). This suggests that the >>appropriate response is either to ignore these tests on i386, or
possibly (the graph object output by gdf_to_nx has (x: float, y:
float) node keys, so rounding issues can potentially break
retrieving individual nodes/edges) to remove this package on i386.
The (IMHO) correct fix should be to replace all exact comparisons with approximative ones. I don't think subtle precision errors warrant removal of the package on i386, since this a very common and well-known issue of floating point math.
Downgrading the bug to wishlist and disabling autopkgtest for 32 bits is possible though.
Sysop: | Keyop |
---|---|
Location: | Huddersfield, West Yorkshire, UK |
Users: | 493 |
Nodes: | 16 (2 / 14) |
Uptime: | 159:08:56 |
Calls: | 9,700 |
Files: | 13,732 |
Messages: | 6,179,717 |