Re: psqlodbc.dll code questions

________________________________

From: David Vaillancourt [mailto:david(dot)vaillancourt(at)polymtl(dot)ca]
Sent: 26 October 2005 01:50
To: Dave Page
Subject: Re: psqlodbc.dll code questions


Hi Dave,

Well, followed your advice and installed the package:

http://wwwmaster.postgresql.org/download/mirrors-ftp?file=odbc%2Fversion
s%2Fsnapshots%2Fpsqlodbc-08_01_0005.zip

After which I created the ODBC source (The Unicode Driver) for
my application to connect to. But I still get a buffer ovverrun
and stack corruption that I was getting originally. Here are
some specs:

1) My machine Windows 2k server (Both app and DB are one the
same box).
2) PostgreSQL Database Server 8.0.0-beta3-dev1 ( With a valid DB
created in UNICODE).
3) My application uses ODBC through MFC's CDatabase and
CRecordset (App is UNICODE built).

Upon connection I get a buffer overrun and stack corruption.
Although when I compile my App in
MBCS, all seems to work Ok. As I've told you, I download the
source of the Driver and had been
able to do a fix for it, and it seems to work perfectly. Here's
the line where it all happens in my app:

CDatabase synsoftDb;
synsoftDb.OpenEx(

_T("DSN=PostgreSQL;DATABASE=synsoftdb;SERVER=localhost;PORT=5432;UID=som
eUser;PWD=SomePassword"), CDatabase::noOdbcDialog );

It looks like when I connect in UNICODE, the psqlodbc code
'sees' a big output buffer in:

(From psqlodbc file : dlg_specific.c)
makeConnectString(char *connect_string, const ConnInfo *ci,
UWORD len)
{
char got_dsn = (ci->dsn[0] != '\0');
char encoded_conn_settings[LARGE_REGISTRY_LEN];
UWORD hlen;
/*BOOL abbrev = (len <= 400);*/
BOOL abbrev = (len < 1024); // When my app connects
in MBCS len == 512,
// but in Unicode len ==
1024, which causes
// extra data to be
filled below

/* fundamental info */
sprintf(connect_string,
"%s=%s;DATABASE=%s;SERVER=%s;PORT=%s;UID=%s;PWD=%s",
got_dsn ? "DSN" : "DRIVER",
got_dsn ? ci->dsn : ci->drivername,
ci->database,
ci->server,
ci->port,
ci->username,
ci->password);

encode(ci->conn_settings, encoded_conn_settings);

/* extra info */
hlen = strlen(connect_string);
if (!abbrev)
// This fills the buffer way more than needed, then it's
converted
// into wide chars and busts, thus creating corruption


So for the time being I've replaced :
if(!abbrev)
with
if(0)

This way no extraneous data is written, and this seems to take
care of the problem for now. I'm afraid this is a 'patch' that
might not hold in other cases. Since you know the code and
architecture better than I do, what do you make of this? If this
isnt clear, I will try to give more detailed and structured
information if needed. Your input will be greatly apprecidated.

Hi David,

As a first step, I've fixed a bug that caused the driver to try to write
an output connection string, even if the app set the max length to zero.
This could only be seen in Unicode mode. Please try the attached patch
to the driver and let me know how you get on.

I've committed this change anyway as this is definately a bug - it just
might not be entirely what you are seeing.

Regards, Dave