Since my alternative is using json, that is heavier (need to store keys in every row) than composite-types. Updating an element on a specific composite_type inside an array of them is done by UPDATE table SET composite[2].x = 24;
So last standing question, is it possible to insert an array of composite_types by not specifying all of the columns for each composite_type ? So if i later add other columns to the composite_type, the insert query doesn't break ? Thanks On Mon, Apr 21, 2014 at 1:46 PM, Dorian Hoxha <dorian.ho...@gmail.com>wrote: > Maybe the char array link is wrong ? I don't think an array of arrays is > good for my case. I'll probably go for json or separate table since it > looks it's not possible to use composite-types. > > > On Mon, Apr 21, 2014 at 4:02 AM, Rob Sargentg <robjsarg...@gmail.com>wrote: > >> Sorry, I should not have top-posted (Dang iPhone). Continued below: >> >> On 04/20/2014 05:54 PM, Dorian Hoxha wrote: >> >> Because i always query the whole row, and in the other way(many tables) i >> will always join + have other indexes. >> >> >> On Sun, Apr 20, 2014 at 8:56 PM, Rob Sargent <robjsarg...@gmail.com>wrote: >> >>> Why do you think you need an array of theType v. a dependent table of >>> theType. This tack is of course immune to to most future type changess. >>> >>> Sent from my iPhone >>> >>> Interesting. Of course any decent mapper will return "the whole >> row". And would it be less disk intensive as an array of "struct ( where >> struct is implemented as an array)". From other threads [1] [2] I've come >> to understand the datatype overhead per native type will be applied per >> type instance per array element. >> >> [1] 30K >> floats<http://postgresql.1045698.n5.nabble.com/Is-it-reasonable-to-store-double-arrays-of-30K-elements-td5790562.html> >> [2] char >> array<http://postgresql.1045698.n5.nabble.com/COPY-v-java-performance-comparison-tc5798389.html> >> > >