[racket] Additional cached index for documentation?

At Tue, 29 Apr 2014 10:16:08 -0400, Greg Hendershott wrote:
> > Probably it should be a new document along the lines of
> > "scribblings/main/[user/]search.scrbl" (in the "racket-index" package)
> > or an extra output of those documents.
> >
> > And probably you want to generate a mapping from symbols to xref tags,
> > since `xref-tag->path+anchor` can convert a tag to a URL relatively
> > efficiently (by way of "docindex.sqlite").
> 
> Did someone go ahead and do this?
> 
> Because a few weeks ago I noticed this got _much_ faster on Racket
> HEAD. Well under 1 second. Which is awesome.

I think I must have broken something in v6.0 and fixed it before the
v6.0.1 branch. The current release v6.0.1 release candidate and v5.3.6
run in about the same time for me, but v6.0 takes about 10 times as
long.

I vaguely recall noticing that cross-reference deserialization was too
slow and fixing something there, but that memory might not be right,
because I can't remember or find the details.