FixupMultiply10: Why do you add delta_plus to numerator about double-conversion HOT 3 CLOSED

Sorry this reply took so long...
You are probably right that it's not necessary. I wrote the code in accordance to the PLDI paper which proves that this approach is correct. A few weeks back I tried to prove that the algorithm is still correct without adding delta_plus. It didn't finish (although I admittedly didn't spend much time on it).
Given that the adding shouldn't be expensive I prefer not to change it. (It just has been too long since I wrote it, and I would need significant effort to get completely back into the algorithm).
If there are real downsides to adding it, I could give it a bit more attention, though.

from double-conversion.

I was wrong, adding it is necessary.

I'm working on a project, and today I found that testcase "1e-23" (vs debugger represents it as "9.9999999999999996e-24") will fail if I change the algorithm. In this case, the first generated digit is 0, then it reached the boundary (in delta plus room), round up, 1.

I'm sorry for this.

from double-conversion.

Thanks a lot for giving an update.
I will make sure I add this as a test-case.

from double-conversion.