Re: [Firebird-devel] Problem ... looking for suggestions. LONG

[Mike N. <ta...@al...>]

Ann W. Harrison wrote:


> My copy of Firebird and earlier versions
> of both Firebird and InterBase get a "wrong length record (183)"
> error on the second or third record in the RDB$FUNCTIONS table
> of one particular V5.6 database.

First, do you know if this works using the original tarball, or does that
one also fail?

Try adding a static size_t and increment it everytime you enter VIO_data. It
will probably break when the counter reaches 7352. It gives you a nice
counter to break on in the debugger. Using Win32, if you're lazy, you can
even add

 static size_t nEntries = 0;
 ++nEntries;
 if (nEntries == 0x1cb8) {
  __asm int 3;
 }

What's wrong here is that he record format doesn't match the length of the
decompressed block. SQZ is fine, it's bad earlier.

The function it tries to write is probably "F_DEFAULTFLOAT" (last one
printed in the gbak console), while the particular relation is
"RDB$FUNCTION_ARGUMENTS".

From tracing into MET_format() (via VIO_record), it already here display
that this relation.rel_cur_format.fmt_length is 0x34. What we get from
SQZ_decompress later on is 0x32. It's bad input even earlier.

The compressed buffer that breaks, len 0x1f, looks like

01 80 FD 00 0E F_DEFAULTFLOAT EF 20 FB 00 01 1B FB 00 01 08 FB 00


The closest I get back in the call stack before becoming completely lost is
rse.c get_record() on the rsb_indexed case. Well, if you don't count how it
gets the RPB* in get_record():

  request->req_rpb + rsb->rsb_stream

rsb->rsb_stream is zero, why it's the plain request->req_rpb that it gets.


> The certified InterBase does
> not, though I can't find any changes on sourceforge that might
> account for the difference.

Then we definitely have to dig deeper. Have you tried the *hrm* mers build
*hrm*?

>    Could someone who has a recent Firebird build (blush) try
>    the same?

It breaks.


> The values from -3 to -128 indicate that the byte
> that follows is to be replicated (-1 times )that many times.

Are -1 and -2 illegal values? If so, I think it should be asserted.


I even ran the server under Purify to rule out buffer overflows. No detected
problems.

Not much of a help, but I hope it can speed you on your way.

/Mike