Re: Writing transactions from pgtksh

From: Andreas Kretzer <andi(at)kretzer-berlin(dot)de>
To: Postgres Interface List <pgsql-interfaces(at)postgresql(dot)org>
Subject: Re: Writing transactions from pgtksh
Date: 2002-04-09 08:24:25
Message-ID: 3CB2A539.B73785CC@kretzer-berlin.de
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-interfaces

Denis Chavez schrieb:

> Hi!
>
> I'm using pgtksh to develop a Postgres client application. I need to use
> transactions, but
> the Postgres Programmer's Guide just tells me to use large objects and
> doesn't give any
> example.
>
> Any idea on how to use this pg_lo* with transactions in pgtksh?

I'm using the normal Tcl/Tk together with libpgtcl.so (or libpgtcl.dll under windoze).
It should effectively be the same thing.

The documentation states, that you must enclose your LargeObject operation
in a transaction -- and this is definetly true! So let's take a code fragment
to create a large object from a client-local file ($db is my connection handle):

proc create_lobject {db file} {
# let's open the file
if {[catch {set fd [open $file r]}]} {
... error message 'couldn't open file' ...
return -1
}
fconfigure $fd -translation binary -encoding binary -eofchar {}

# now create the object
set oid [pg_lo_create $db INV_READ|INV_WRITE]
if {[catch {set oid [expr $oid * 1]}]} { # check for numeric value
... some error message ...
return -1
}

# start with a transaction and open the object
set res [pg_exec $db "BEGIN TRANSACTION;"]
pg_result $res -clear; # don't care for result - this should always work

set lobj_fd [pg_lo_open $db $oid w]
# one should check $lobj_fd for succesfull open - I don't do this for now :-(

# now sequentially read file and write to the large object like
# you would do with a regular file
while {![eof $fd]} {
set buf [read $fd 1024]
set len [string bytelength $buf]
pg_lo_write $db $lobj_fd $buf $len
}

# finish everything
pg_lo_close $db $lobj_fd
close $fd

set res [pg_exec $db "COMMIT TRANSACTION;"]
pg_result $res -clear

return $oid
}

Remember to store the returned OID in some descriptive data set. The
large object can only be recovered by it's OID. Reading works exactly
like writing - except that you don't create the object before opening
it.

ATTENTION: This works under linux :-) but may give you some trouble
under Windows :-(
Under Windows you can't really import binary files as they are recoded
every time. Even the 'fconfigure' command can't do anything against
it (I'm still searching for the reason ...). One may be that you even
can't figure out the real length of what you read from the file and
in addition the 'buf' contains weired chars that won't make up the
real binary representation of the file (mainly because some UTF-8
chars or whatever).

There are two other methods to read and write large objects.
pg_lo_import and pg_lo_export that should do all that in one
step. I encountered the same problems with binary files on windows
so I just don't use them anymore.

Hope this helps

Andreas

In response to

Browse pgsql-interfaces by date

  From Date Subject
Next Message Bruce Momjian 2002-04-11 21:42:06 Re: libpgtcl pg_execute
Previous Message "." 2002-04-07 02:31:43 sqlbang