GENTLE REMINDER of RFR CSR JDK-8202555: Double.toString(double) sometimes produces incorrect results

Raffaello Giulietti raffaello.giulietti at
Fri Dec 27 22:25:07 UTC 2019

On 2019-12-27 22:16, Andrew Haley wrote:
 > On 12/27/19 11:43 AM, Raffaello Giulietti wrote:
 >> The latest patches of [1] have been submitted more than 8 months ago 
 >> I'm still looking for motivated reviewers.
 > I am sorely aware that we have not performed well. All I can say in our
 > defence is that we have looked at it, and a thorough review is hard,
 > and there is significant risk touching this code. I know that Andrew
 > Dinn would love to review it but he keeps getting dragged away to do
 > other work. We have not forgotten you.

Hi Andrew & Andrew

thanks for your kind reply.

I'm aware that a review is harder than usual because of the size of the 
writing [1]. Of course, I can justify and explain every single line of 
the code if so requested. Most of it is based on details in the writing 
and is often a mere translation in Java.

To my knowledge, the algorithm is quite novel, so I cannot simply refer 
to existing, well-established and well-known work (except for some 
limited aspects on Adams' Ryu work [2], which are pretty coincidental).

Brain Burkhalter (Oracle) read the writing until section 8 included and 
stopped with section 9 which discusses the efficient implementation. I 
don't know if he read further in the meantime.

Except in answering specific questions about my contribution, I guess I 
can do very little to help those interested in finding the time to dive 
into the details. My estimate is that it takes between 0.5 and 1 day to 
understand the paper and less than a couple of hours to check that the 
code is a direct translation in Java. As usual, YMMV.



[2] Adams, "Ryu: Fast Float-to-String Conversion", PLDI, 2018

More information about the core-libs-dev mailing list