On February 21, 2019 10:08:00 AM PST, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>Robert Haas <robertmhaas@gmail.com> writes:
>> Perhaps a better approach still would be to do what Andres proposed
>> back in March:
>
>> #> Is there any chance we can instead can convert dictionaries into a
>form
>> #> we can just mmap() into memory? That'd scale a lot higher and
>more
>> #> dynamicallly?
>
>That seems awfully attractive. I was about to question whether we
>could
>assume that mmap() works everywhere, but it's required by SUSv2 ... and
>if anybody has anything sufficiently lame for it not to work, we could
>fall back on malloc-a-hunk-of-memory-and-read-in-the-file.
>
>We'd need a bunch of work to design a position-independent binary
>representation for dictionaries, and then some tool to produce disk
>files
>containing that, so this isn't exactly a quick route to a solution.
>On the other hand, it isn't sounding like the current patch is getting
>close to committable either.
>
>(Actually, I guess you need a PI representation of a dictionary to
>put it in a DSM either, so presumably that part of the work is
>done already; although we might also wish for architecture independence
>of the disk files, which we probably don't have right now.)
That's what I was pushing for ages ago...
--
Sent from my Android device with K-9 Mail. Please excuse my brevity.