Patch for DBI dataloss problem

Hello

As reported in the Debian bug tracking system. Can you comment if this
patch is OK?

bye,

-christian-



From: Tim Cutts
To: ch@debian.org
Subject: libdbi-perl 1.55
Date: Mon, 14 May 2007 11:48:26 +0100
X-Mailer: Apple Mail (2.752.2)

Hi Christian,

One of the changes introduced in DBI 1.55 is a very small fix which
fixes a bug which can really cause data loss (a problem with
selectall_arrayref). The bug was spotted by one of my colleagues
here (Rob Davies, whom Tim Bunce credits in the changelog). We have
actually seen it cause real data loss here, so it's not just a
theoretical problem.

What I was wondering was whether there is any way we can get that fix
into Debian on etch's next point release? The patch to fix that
particular issue is as follows:

diff -Naur DBI-1.54/DBI.xs DBI-1.54.patched/DBI.xs
--- DBI-1.54/DBI.xs 2007-02-22 01:36:15.000000000 +0000
+++ DBI-1.54.patched/DBI.xs 2007-04-26 09:38:00.667000000 +0100
@@ -9,7 +9,7 @@

#define IN_DBI_XS 1 /* see DBIXS.h */
#define PERL_NO_GET_CONTEXT
-
+#include
#include "DBIXS.h" /* DBI public interface for DBD's written in C
*/

# if (defined(_WIN32) && (! defined(HAS_GETTIMEOFDAY)))
@@ -3104,6 +3104,9 @@
outitems = perl_call_sv(isGV(imp_msv) ? (SV*)GvCV(imp_msv) :
imp_msv,
(is_DESTROY ? gimme | G_EVAL | G_KEEPERR : gimme) );
}
+
+ assert(outitems >= 0);
+
SPAGAIN;

/* XXX restore local vars so ST(n) works below */
diff -Naur DBI-1.54/Driver.xst DBI-1.54.patched/Driver.xst
--- DBI-1.54/Driver.xst 2007-02-22 01:36:15.000000000 +0000
+++ DBI-1.54.patched/Driver.xst 2007-04-26 15:15:47.084000000 +0100
@@ -121,6 +121,7 @@
}
else {
sth = dbixst_bounce_method("prepare", 3);
+ SPAGAIN; SP -= items; /* because stack might have been realloc'd */
if (!SvROK(sth))
XSRETURN_UNDEF;
}
@@ -157,6 +158,7 @@
else {
/* --- prepare --- */
sth = dbixst_bounce_method("prepare", 3);
+ SPAGAIN; SP -= items; /* because stack might have been realloc'd */
if (!SvROK(sth)) {
if (is_selectrow_array) { XSRETURN_EMPTY; } else { XSRETURN_UNDEF; }
}

Re: Patch for DBI dataloss problem

Looks fine, though the DBI.xs change (assert) isn't needed and isn't in the DBI code.

Tim.

On Tue, Jun 19, 2007 at 12:43:53AM +0200, Christian Hammers wrote:
> Hello
>
> As reported in the Debian bug tracking system. Can you comment if this
> patch is OK?
>
> bye,
>
> -christian-
>
>
>
> From: Tim Cutts
> To: ch@debian.org
> Subject: libdbi-perl 1.55
> Date: Mon, 14 May 2007 11:48:26 +0100
> X-Mailer: Apple Mail (2.752.2)
>
> Hi Christian,
>
> One of the changes introduced in DBI 1.55 is a very small fix which
> fixes a bug which can really cause data loss (a problem with
> selectall_arrayref). The bug was spotted by one of my colleagues
> here (Rob Davies, whom Tim Bunce credits in the changelog). We have
> actually seen it cause real data loss here, so it's not just a
> theoretical problem.
>
> What I was wondering was whether there is any way we can get that fix
> into Debian on etch's next point release? The patch to fix that
> particular issue is as follows:
>
> diff -Naur DBI-1.54/DBI.xs DBI-1.54.patched/DBI.xs
> --- DBI-1.54/DBI.xs 2007-02-22 01:36:15.000000000 +0000
> +++ DBI-1.54.patched/DBI.xs 2007-04-26 09:38:00.667000000 +0100
> @@ -9,7 +9,7 @@
>
> #define IN_DBI_XS 1 /* see DBIXS.h */
> #define PERL_NO_GET_CONTEXT
> -
> +#include
> #include "DBIXS.h" /* DBI public interface for DBD's written in C
> */
>
> # if (defined(_WIN32) && (! defined(HAS_GETTIMEOFDAY)))
> @@ -3104,6 +3104,9 @@
> outitems = perl_call_sv(isGV(imp_msv) ? (SV*)GvCV(imp_msv) :
> imp_msv,
> (is_DESTROY ? gimme | G_EVAL | G_KEEPERR : gimme) );
> }
> +
> + assert(outitems >= 0);
> +
> SPAGAIN;
>
> /* XXX restore local vars so ST(n) works below */
> diff -Naur DBI-1.54/Driver.xst DBI-1.54.patched/Driver.xst
> --- DBI-1.54/Driver.xst 2007-02-22 01:36:15.000000000 +0000
> +++ DBI-1.54.patched/Driver.xst 2007-04-26 15:15:47.084000000 +0100
> @@ -121,6 +121,7 @@
> }
> else {
> sth = dbixst_bounce_method("prepare", 3);
> + SPAGAIN; SP -= items; /* because stack might have been realloc'd */
> if (!SvROK(sth))
> XSRETURN_UNDEF;
> }
> @@ -157,6 +158,7 @@
> else {
> /* --- prepare --- */
> sth = dbixst_bounce_method("prepare", 3);
> + SPAGAIN; SP -= items; /* because stack might have been realloc'd */
> if (!SvROK(sth)) {
> if (is_selectrow_array) { XSRETURN_EMPTY; } else { XSRETURN_UNDEF; }
> }