Re: tdf#109097 - mdds accelerating random lookups

On 25/08/17 16:42, Eike Rathke wrote:
> Does not sound very appealing to me. A better lookup time is certainly
> desirable by means of interpolation or binary search, but at the cost of
> insertion/deletion time that's not gonna fly.

        Sure - unfortunately its not possible to do a good interpolation lookup
without the positions in there.

> We still have bottlenecks
> where inserting and especially changing type of a cell, which leads to
> shrinkage of one block and enlargement or insertion of another, can
> build up to looong waiting times. I don't have the bug number at hand
> right now, but the worst case is a Find&Replace that changes string
> cells to numeric (or vice versa) in a column of data.

        Last I looked at that - we were doing find/replace horizontally which
was very wasteful - and switching it to doing things in a columnar
direction seemed to help in
https://bugs.documentfoundation.org/show_bug.cgi?id=92160

> I think these should be addressed first.

        Fair cop; ultimately - re-building the column as it is replaced by
moving and appending the items to a new clean column should have a more
linear cost there I'd hope.

        Any other concerns there ? ultimately the old code must have had this
linear cost too - for insertions / deletions etc. (but not for cell type
changes).

        ATB,

                Michael.

-- 
michael.meeks@collabora.com <><, Pseudo Engineer, itinerant idiot