Hideous "revdberr,invalid database type" returns in standalones

[Ben Rubinstein]

An app using the revdblibrary to communicate over ODBC.  The standalone
works on my machine, fails on my client's (and on another test machine in my
office).

Now cut down to a test stack which just attempts to open a connection
(revOpenDatabase), and either reports the error or closes the connection
again.

Rev 2.5 (with sep20 updater run): in IDE, works fine.  Standalone on my
development machine, works fine.  Standalone on any other machine, delivers
the dreaded:
    revdberr,invalid database type

AFAICT, the bundle seems to have everything it needs in it.

Development machine running OS X 10.3.5.  Client's test machine 10.2.8
(earlier version of the app, created with Rev 2.1, works fine).  Other test
machine also 10.3.5, so my suspicion is something to do with the fact that
I've got Revolution installed on my dev machine but nowhere else.

On my machine, I can move the test app anywhere and it works, ie there
aren't some nasty relative paths in it that aren't inside itself.  I've
tried stuffing it, and running it on my machine from a newly unstuffed copy,
same as on the other machines - again, it still runs ok on my machine, so I
guess it's not a transmission error.

I've tried modifying my test app so I can explicitly call
'revSetDatabaseDriverPath' while it's running (since there isn't anyway to
find out what it's been initialised to by the standard startup code). I set
it to the full path to the "database_drivers" folder inside the bundle.
Same result as if I set it to anything else.

Can there be some weird set up in the library or similar on my machine?
Something that gets installed when you run Rev?  I'm really at a loss here.

FWIW: building the test app under Rev 2.2.1, when the button is pressed the
it (I think) simply hits an error and the script stops.   This is the case
even on my dev machine.  Building the test app under Rev 2.1.2 works fine,
on all machines on which tested... which is fine except that the real app is
using the encryption features of Rev 2.5.

[Apologies for the cross-posting, but I'm on the edge here with an important
client, who is facing an immovable deadline of their own, and at the last
minute this issue has come up.  It's arguably both use and improve, since
there's clearly a bug, but there may be a workaround!]

I am in pain.  Any help, or suggestions, gratefully received.
 
  Ben Rubinstein               |  Email: benr_mc at cogapp.com
  Cognitive Applications Ltd   |  Phone: +44 (0)1273-821600
  http://www.cogapp.com        |  Fax  : +44 (0)1273-728866