Explicitly cast double values down to float in Row.GetFloat().#706
Conversation
Supports casting to float in Pomelo.EntityFramworkCore.MySql for MySQL < 8.0.17. Signed-off-by: Laurents Meyer <laucomm@gmail.com>
lauxjpn
force-pushed
the
float_conversion
branch
3 times, most recently
from
9697e5d to
5881931
Compare
|
If you are concerned, that the explicit guard against information loss could lead to an |
…ks, to ensure no significant loss of information, when casting from double to float. Signed-off-by: Laurents Meyer <laucomm@gmail.com>
lauxjpn
force-pushed
the
float_conversion
branch
from
5881931 to
4e97fee
Compare
| // Use explicit range checks to guard against that. | ||
| return value switch | ||
| { | ||
| double doubleValue => (doubleValue >= float.MinValue && doubleValue <= float.MaxValue ? (float) doubleValue : throw new InvalidCastException("The value cannot be safely cast to Single.")), |
There was a problem hiding this comment.
This test could be if doubleValue is contained in any of the following ranges: [float.MinValue, -float.Epsilon], [0, 0], [float.Epsilon, float.MaxValue]; what do you think?
There was a problem hiding this comment.
So the difference would be, that double values between -/+ float.Epsilon and 0 would throw.
The question is, if it is expected that a double between float.Epsilon and 0 is being rounded to float.Epsilon or 0. I think it is, because it's just a result of precision-loss.
(But there could probably be made a case against it as well, which would expect those values to end up as float.Epsilon. I think that is imprecise though, and they need to be rounded. And I would assume, that C#/the processor is doing exactly that in the first place.)
I would not throw in any of these cases, because this would be unexpected and a bit random.
The only potential problem I see with the current test is, that multiple arithmetic operations with a double near the float's min or max threshold, could end up pushing the value over the threshold due to precision error and thus end up throwing an InvalidCastException.
But this is an unlikely corner case and I think guarding against an accidental call of GetFloat on a double column, that then leads to a loss of information, is a much more common and severe case and is worth guarding against.
|
Fixed in 0.59.0. |
2 participants
Support casting to float for Pomelo.EntityFramworkCore.MySql for MySQL < 8.0.17.
See PomeloFoundation/Pomelo.EntityFrameworkCore.MySql#827 (comment)
Signed-off-by: Laurents Meyer laucomm@gmail.com