Our original evaluation of Errol against the prior work of Grisu3 was erroneous. The evaluation indicates a 2x speed improvement over Grisu3. However, corrected performance measurements show a 2x speed loss to Grisu3.

Early on in our research, we wrote benchmarks to test our algorithm in relation to the previous algorithms Dragon4 and Grisu3. The build system for Grisu3 available for download did not compile Grisu3 with optimizations turned on. We thought this may be the case and attempted to configure the Grisu3 build system to on turn compiler optimizations. Unfortunately, we got this step wrong because of our unfamiliarity with the SCons build system used in the Grisu3 build. Furthermore, our misconfiguration bug did not cause any errors, and so we thought that optimizations were already enabled in the default configuration -- when in reality it was not. The Artifact Evaluation Committee ran our benchmarks and reproduced our timing numbers, but they reproduced the same incorrect numbers that we collected. This is not surprising, since they ran our build script did not turn on optimizations, and there were no visible errors that would hint at a problem.

The mistake was discovered when Florian Loitsch, author of the Grisu3 algorithm, attempted to replicate our results. He alerted us to the discrepancy, and we immediately confirmed the corrected results. We then informed the POPL PC chair of the mistake, and the PC chair advised us to create a page like this one and update the PDF paper and corresponding talk.

An updated version of the paper is available here: https://cseweb.ucsd.edu/~mandrysc/pub/dtoa.pdf.