Re: update on TOAST status

Jeffery Collins wrote:
> Jan Wieck wrote:
>
> > FYI,
> >
> >     For now, "lztext" is the  only  test  candidate  datatype  to
> >     invoke  the  toaster.  It can hold up to multi-megabytes now.
> >     But be warned, this datatype will disappear as soon as "text"
> >     is toastable.
> >
>
> I have not been following the TOAST discussion, but why would lztext
> disappear?  It seems like a useful datatype independent of TOAST or not
> TOAST?
   The  "lztext" type was something I developed before TOAST was   born. It's was a "text" type that tried to compress
thevalue   at input time.
 
   In the TOAST world, each input value will be passed around as   is.  Only when it gets down to be stored in a table
and the   resulting  heap  tuple  exceeds  2K,  the toaster will try to   compress toastable attributes and/or move off
attributes.The   behaviour  will  be  configurable  on  a per tables attribute   base. So someone can specify "don't
trycompression", "ignore   this  attribute until all others are toasted" or "never toast   this, instead fail and abort
-unwise but possible".
 
   In the current CVS sources,  "lztext"  already  doesn't  know   anything   about  compression  anymore.  It's  more
or less   equivalent to "text"  now,  where  it's  lztextin()  function   produces  a plain varlena structure like
textin()does.  Only   that all it's other functions are aware that the values  they   recieve might be toasted ones.
It'sthe toaster that does the   compression/move-off for it now.
 
   So as soon as "text" is toastable,  there  is  absolutely  no   need for "lztext" anymore. We will add an alias to
theparser   for  7.1,  which  will  disappear  in  7.2  again.   If   you   pg_dump/restore  your  databases during the
7.0->7.1upgrade,   all your table schemas will  automatically  be  changed  from   "lztext" to "text".
 


Jan

--

#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me.                                  #
#================================================== JanWieck@Yahoo.com #