[REVIEW 3-5] fdo#47644 performance regression on largish .doc

[Caolán McNamara]

On Fri, 2012-05-11 at 10:09 +0100, Michael Meeks wrote:
>  we actually precisely know our offset from the beginning of the chain/array
> ( which is nNew/nPageSize ).

hmm, that escaped me entirely. Been trying to find a loophole that
invalidates that e.g. 10 20 30 40 where 20 is the first entry. But
anything like that seems to be covered in that the ::Inits pass the
first entry which ends up as nStart and we build the chain only from
there. So I can't see any reason it wouldn't work.

> 	So - how about the attached diff, hopefully rather easier to review &
> back-port :-)

Seems sane, though it probably makes sense to tweak the trigger for
calling buildPageChainCache as the units for nNew there are in bytes
while the define is in "pages". And we might want to avoid calling it in
the case of a small relative jump in pages forward.

C.