Re: genomic locus

> I think I can wrangle this type into GiST just by tweaking consistent(),
> union(), and picksplit(), if I manage to express my needs in C without breaking
> too many things. My first attempt segfaulted.
Actually, consistent() can determ actual query data type by strategy number. See
examples in ltree, intarray

> If all goes to plan, I will end up with an index tree partitioned by contig at
> the top level and geometrically down from there. That will be as close as I can
> get to an array of config-specific indices, without having to store data in
> separate tables.
>
> What do you think of that?

I have some doubt that you can distinguish root page, but it's possible to
distinguish leaf pages, intarray and tsearch do that.

Reading your plan, I found an idea for GIN: key for GIN is a pair of (contig,
one genome position). So, any search for interset operation with be actually a
range search from (contig, start) to (contig, end)

> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> I have a low-level technical question. Because I can’t anticipate the maximum
> length of contig names (and do not want to waste space), I have made the new
> locus type a varlena, like this:
>
> #include "utils/varlena.h"
>
> typedef struct LOCUS
> {
>   int32 l_len_; /* varlena header (do not touch directly!) */
>   int32 start;
>   int32 end;
>   char  contig[FLEXIBLE_ARRAY_MEMBER];
> } LOCUS;
>
> #define LOCUS_SIZE(str) (offsetof(LOCUS, contig) + sizeof(str))
sizeof? or strlen ?

>
> That flexible array member messes with me every time I need to copy it while
> deriving a new locus object from an existing one (or from a pair). What I ended
> up doing is this:
>
>   LOCUS  *l = PG_GETARG_LOCUS_P(0);
>   LOCUS  *new_locus;
>   char   *contig;
>   int    size;
>   new_locus = (LOCUS *) palloc0(sizeof(*new_locus));
>   contig = pstrdup(l->contig); // need this to determine the length of contig
l->contig should be null-terminated for pstrdup, but if so, you don't need to
pstrdup() it - you could use l->contig directly below. BTW, LOCUS_SIZE should
add 1 byte for '\0' character in this case.

> name at runtime
>   size = LOCUS_SIZE(contig);
>   SET_VARSIZE(new_locus, size);
>   strcpy(new_locus->contig, contig);
>
> Is there a more direct way to clone a varlena structure (possibly assigning an
> differently-sized contig to it)? One that is also memory-safe?
Store length of contig in LOCUS struct.

--
Teodor Sigaev E-mail: teodor(at)sigaev(dot)ru
WWW: http://www.sigaev.ru/