Design of persistent tag cache

[Greg Ward]

Related question.  Right now, I have a failing test because my
tag-caching code has changed the behaviour of

  echo "crap" >> .hgtags
  hg commit "add crap"
  hg tags
  hg tags

Without caching, both runs of "hg tags" print warnings.  With caching
as currently implemented, only the first run does.

I *could* bend over backwards to avoid caching info from .hgtags
revisions with crap in them.  But I'd rather not.  Does anything think
the repeated warning is worth the bother?

Greg