Tom Lane <tgl@sss.pgh.pa.us> wrote:
> This seems extremely invasive for a relatively small gain :-(
> The example you cite of an int4 index on a MAXALIGN-8 machine is
> by far the best case, and in many cases there wouldn't be anything
> bought by the extra complexity.
I also think that it has small advantage without another compression method.
But if we tries to compress index keys using prefix compression or so,
we might need to put tuples with no/small alignment.
I would like to propose it again when I find another method which will work
well with this patch.
---
ITAGAKI Takahiro
NTT Cyber Space Laboratories