Date: 2011-08-24 07:53:05 +0200
From: @drstmane
To: SQL devs <>
Version: 11.5.1 (Aug2011) [obsolete]

Last updated: 2011-09-16 15:04:29 +0200

Comment 16157

Date: 2011-08-24 07:53:05 +0200
From: @drstmane

Several SQL test appear to crash the server on "algol" (Windows2008) and "gans" (WindowsXP), while working fine on "lab03" (Windows7).

Or is the pattern that "lab03" runs 64-bit with 32-bit OIDs, while "algol" & "gans" run 64-bit with 64-bit OIDs or 32-bit with 32-bit OIDs?

cf., e.g.,
http://monetdb.cwi.nl/testweb/web/testgrid.php?serial=41544:4b3e3e10e76f&module=sql&targets=Int-WindowsXP-i386-propcheck,Mic-Windows2008-i386-propcheck-installer,Int-Windows7-x86_64-propcheck-oid32,Mic-Windows7-x86_64-propcheck-oid32-installer,Int-Windows2008-x86_64-propcheck,Mic-Windows2008-x86_64-propcheck-installer&order=platform,arch,compiler

Comment 16162

Date: 2011-08-25 22:02:47 +0200
From: @drstmane

Swapping targets x86_64-propcheck & x86_64-propcheck-oid32 between algol & lab03 confirms that the pattern is indeed that the tests cause a MonetDB server crash on "algol" (Windows2008) and "gans" (WindowsXP), while working fine on "lab03" (Windows7) --- independently of BIT and OID settings; cf.,

http://monetdb.cwi.nl/testweb/web/testgrid.php?serial=41593:ab2ce99c5b12&module=sql&targets=Int-WindowsXP-i386-propcheck,Mic-Windows2008-i386-propcheck-installer,Int-Windows2008-x86_64-propcheck-oid32,Mic-Windows7-x86_64-propcheck-oid32-installer,Int-Windows7-x86_64-propcheck,Mic-Windows2008-x86_64-propcheck-installer&order=platform,arch,compiler

Comment 16164

Date: 2011-08-26 08:36:02 +0200
From: @drstmane

It appears that all tests that crash on algol (Windows2008) and gans (WindowsXP), but work fine on lab03 (Windows7),
contain a "LIKE" predicate in the query, i.e.,
there crashes might be related to evaluating LIKE predicates.
Do we use some external library like PCRE for this?

(I cannot tell, whether there are tests with a LIKE predicate that do work fine also on algol (Windows2008) and gans (WindowsXP).)

http://monetdb.cwi.nl/testweb/web/testgrid.php?serial=41593:ab2ce99c5b12&module=sql&targets=Int-WindowsXP-i386-propcheck,Mic-Windows2008-i386-propcheck-installer,Int-Windows2008-x86_64-propcheck-oid32,Mic-Windows7-x86_64-propcheck-oid32-installer,Int-Windows7-x86_64-propcheck,Mic-Windows2008-x86_64-propcheck-installer&order=platform,arch,compiler&diff

Comment 16166

Date: 2011-08-26 08:39:41 +0200
From: @drstmane

See also bug #2868.

Comment 16167

Date: 2011-08-26 11:37:27 +0200
From: @sjoerdmullender

Changeset f00e5a515b79 made by Sjoerd Mullender sjoerd@acm.org in the MonetDB repo, refers to this bug.

For complete details, see http//devmonetdborg/hg/MonetDB?cmd=changeset;node=f00e5a515b79

Changeset description:

Deal properly with "free"ing NULL pointers.
We use pcre_study() to improve matching.  When we're done, we free the
result of pcre_study() using a call to my_pcre_free().
pcre_study() may return NULL if no state is needed for optimization of
the search pattern.  Since we just pass the result of pcre_study()
around to other pcre functions, we don't really care about the result,
but we do have to free the memory by calling my_pcre_free.
However, my_pcre_free couldn't deal with NULL pointers, which caused a
crash.
We now check for NULL pointers in my_pcre_free.
This fixes bug #2869.

Comment 16168

Date: 2011-08-26 21:47:04 +0200
From: @drstmane

Testing confirms the fix.

Comment 16232

Date: 2011-09-16 15:04:29 +0200
From: @sjoerdmullender

The Aug2011 version has been released.