Michael Fuhr wrote:
> On Fri, Dec 02, 2005 at 04:30:54PM -0500, Tom Lane wrote:
> > Bruce Momjian <pgman@candle.pha.pa.us> writes:
> > > Wow, check this out:
> > >   test=> SELECT CAST (pow(10::numeric, 10000) + 1 AS TEXT)
> > > It works fine!  I have all the digits, and the trailing 1.0:
> > >   000001.0000000000000000
> > > while SELECT pow(10::numeric, 10000) fails.
> > 
> > That's just about as wacky as can be, because numeric_text() is
> > implemented on top of numeric_out() ... there's no way that numeric_out
> > can be delivering the wrong answer if the cast produces the right text.
> > So somewhere between numeric_out and the delivery to the client,
> > something's getting confused.  I think it's time you got out your
> > debugger and started tracing through the backend ...
> Bruce, have you run a process trace on the backend to see if write()
> (or whatever) is writing the correct number of characters?  What
> exactly is your output device and how are you connected to the
> machine that runs the backend (ssh to a remote box from an xterm,
> sitting in front of the box's VT52 serial console, etc.)?
> If you run the query that fails in a standalone backend, do you get
> something like "(typeid = 1700, len = -1, typmod = -1, byval = f)"
> at the end of the line, or is that part truncated too?

I found the cause.  I traced into printf then realized I was not in libc
but port/snprintf.c, and I see 4096 defined for those buffers.  I will
work on a patch to make it dynamic.  At the time I think there was
thought that 4096 was as large as it ever needed to be, but obviously
this was wrong. I think Win32 would see the same failure because it used
port/snprintf.c too.

  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073

---------------------------(end of broadcast)---------------------------
TIP 1: if posting/reading through Usenet, please send an appropriate
       subscribe-nomail command to [EMAIL PROTECTED] so that your
       message can get through to the mailing list cleanly

Reply via email to