[9fans] APE or not for PostgreSQL

[9fans] APE or not for PostgreSQL

[Stephen C Harris]

I'm considering having a go at porting PostgreSQL database to Plan 9.

The goal is to provide a SQL query service where one writes a query
to a "newquery" file and gets back a resultset directory, much 
like one obtains a tcp connection directory.  The directory will have a 
control file for moving in the result set and doing transaction control, 

and a file for each field of the resultset under a "data" directory, 
through 
which one can read and write the current record in the result set.

My question is whether or not to just try and port it to the APE 
emulation
environment or to go "full Plan 9"?  Does the APE environment limit the
ways in which one can provide a 9P service in any way? In other words,
can I link to whatever other Plan9 libraries I'd like to and use them
with impunity if I compile under APE?

There are only a handful of select() calls in the code, and no
mmap() calls, so the native Plan 9 port may not be too
bad.  It seems to use a process-per-connection model and not
pthreads, which is another win. Anybody know any other calls to watch 
out for that might influence my decision?

Steve Harris

Re: [9fans] APE or not for PostgreSQL

[Matt]

> I'm considering having a go at porting PostgreSQL database to Plan 9.
>
> The goal is to provide a SQL query service where one writes a query
> to a "newquery" file and gets back a resultset directory, much
> like one obtains a tcp connection directory.  The directory will have a
> control file for moving in the result set and doing transaction control,
>
> and a file for each field of the resultset under a "data" directory,
> through
> which one can read and write the current record in the result set.
>

i can't say anything about ape but this approach is what I would expect for
a plan9 database

I had been wondering if there was an implemention around that did such a
thing

good luck

Matt

Re: [9fans] APE or not for PostgreSQL

[Anthony Sorace]

i had started working on a PostgreSQL port several
weeks ago, but got sidetracked. sounds like your
target is more involved than mine; i needed a more
"stock" PostgreSQL port for the project i'm doing
this for.

along the lines of what you're doing, i'd like to
see a service that can connect to arbitrary SQL
servers (either remote or local) and provide a
9-ish interface to that, much like /net. that way
applications could be written using the fs
interface, leaving the back-end up to the user
(which would be especially useful in heterogeneous
environments).
: anothy;

Re: [9fans] APE or not for PostgreSQL

[James A. Robinson]

> along the lines of what you're doing, i'd like to
> see a service that can connect to arbitrary SQL
> servers (either remote or local) and provide a
> 9-ish interface to that, much like /net. that way
> applications could be written using the fs
> interface, leaving the back-end up to the user
> (which would be especially useful in heterogeneous
> environments).

This is what I was thinking of awhile ago. For example, I use isql on
my linux box to talk to the big sybase server running on our solaris
machines.  We'd never consider moving sybase to linux at this point,
but having a client to connect is very nice.


Jim

Re: [9fans] APE or not for PostgreSQL

[Boyd Roberts]

sql/ape, turns my stomach.

anyone thought of making the sql schema a fs?

plan 9 is not unix.

--
Boyd Roberts                            boyd@psycho-basket-case.org

    ``I come over here to kill them cocksuckers, not work for 'em''

          -- Moon Dog, _Pettibone's Law_, John Keene

Re: [9fans] APE or not for PostgreSQL

[Steve Kilbane]

> anyone thought of making the sql schema a fs?

can't see how that would help, actually. schemas in themselves are both
numbingly dull and mostly useless. it's how they can be combined with
relational algebra that makes them interesting and useful. and it's
how they _can't_ be combined using pure relational algebra that makes
existing databases such ugly hacks. :-)

i can't help but feel that if you tried to design a fs algebra
to replace relational algebra, you'd end up with flat files and
ye olde unix command set.

steve