In theory this speeds things up by avoiding allocation. In practice, grisu is the bottleneck, at least on 0.4. I will comment further on #25, but this should be an improvement regardless.
Coverage decreased (-0.12%) when pulling e67b1daff999ee160cced5285520be995f6ce976 on simonster:use-printf into 8b2e55545117225023f164ac11a2fc9f0c3a148a on twadleigh:master.
In theory this speeds things up by avoiding allocation. In practice, grisu is the bottleneck, at least on 0.4. I will comment further on #25, but this should be an improvement regardless.