Failed test rt85886.t after DBI update

[lemrouch]

Hi,
we have updated Perl module DBI from 1.646 to 1.647 and t/rt85886.t started to fail with:

t/rt85886.t ............... 1/2 
#   Failed test 'ivok'
#   at t/rt85886.t line 49.
SV = PV(0x55efb9d71230) at 0x55efb9ee5068
  REFCNT = 1
  FLAGS = (POK,pPOK)
  PV = 0x55efb96b3430 "1"\0
  CUR = 1
  LEN = 16
# Looks like you failed 1 test of 2.
t/rt85886.t ............... Dubious, test returned 1 (wstat 256, 0x100)
Failed 1/2 subtests 

Does anyone have any idea what's going on?

[lemrouch]

Failed test is caused by changes in Oracle.so caused by:

--- a/DBI/dbixs_rev.h      2025-01-09 15:35:23.000000000 +0000
+++ b/DBI/dbixs_rev.h        2025-01-17 14:09:59.000000000 +0000
@@ -1,3 +1,2 @@
-/* Fri Jul 13 13:32:02 2012 */
-/* Mixed revision working copy (15349:15353) */
-#define DBIXS_REVISION 15349
+/* Fri Jan 17 15:09:59 2025 */
+#define DBIXS_REVISION 1698

which is used in DBD::Oracle:

./dbdimp.c:#if DBIXS_REVISION >= 13590
./dbdimp.c:#endif  /* DBIXS_REVISION >= 13590 */
./oci8.c:#if DBIXS_REVISION > 13590

[Tux]

Guilty as charged

I need to have a look then, and come up with a fix (or you could)

nice catch, but 15349 makes no sense anymore (svn != git), so reverting is not my preference. That "number" has not changed since below-mentioned commit, so it had extremly low value as it was before I changed its meaning.
I completely overlooked the fact that DBD's could use that for conditional code. Mea culpa.

This condition is based on a change from 2012-07-13 (released in DBI-1.623 on 2013-01-02), so if this conditional code could be changed to (currently non-existent) (DBI_VERSION >= 1.623)

The current meaning is "number of git commits till now". I could add 14159 to that to have 15349 match 1190 (the number of that git commit), but that would still break other code that might check on other values of DBIXS_REVISION

[Tux]

What if I add info to dbixs_rev.h?

🐧 cat dbixs_rev.h
/* Tue Mar 11 10:51:15 2025 */
/*  M dbixs_rev.pl */
/*  M lib/DBI/Changes.pm */
#define DBIXS_VERSION  1
#define DBIXS_RELEASE  648
#define DBIXS_REVISION 1703

Then the #if in DBD::Oracle could be

#if defined(DBIXS_RELEASE) || DBIXS_REVISION == 1698 || DBIXS_REVISION >= 13590
...
#endif

[lemrouch]

That's not up to me to decide but it looks just a little bit confusing without context.

[Tux]

diff --git a/dbdimp.c b/dbdimp.c
index 06ade41..5d08e70 100644
--- a/dbdimp.c
+++ b/dbdimp.c
@@ -766,7 +766,7 @@ int dbd_st_bind_col(SV *sth, imp_sth_t *imp_sth, SV *col, SV *ref, IV type, SV *
         imp_sth->fbh[field-1].bind_flags = 0; /* default to none */
     }

-#if DBIXS_REVISION >= 13590
+#if defined(DBIXS_RELEASE) || DBIXS_REVISION == 1698 || DBIXS_REVISION >= 13590
        /* DBIXS 13590 added StrictlyTyped and DiscardString attributes */
        if (attribs) {
                HV *attr_hash;
diff --git a/oci8.c b/oci8.c
index b01dbe9..9eb5b1a 100644
--- a/oci8.c
+++ b/oci8.c
@@ -4189,7 +4189,7 @@ dbd_st_fetch(SV *sth, imp_sth_t *imp_sth){
                                                        --datalen;
                                        }
                                        sv_setpvn(sv, p, (STRLEN)datalen);
-#if DBIXS_REVISION > 13590
+#if defined(DBIXS_RELEASE) || DBIXS_REVISION == 1698 || DBIXS_REVISION > 13590
                /* If a bind type was specified we use DBI's sql_type_cast
                        to cast it - currently only number types are handled */
                                        if ((fbh->req_type != 0) && (fbh->bind_flags != 0)) {

I am still trying to trace an error I get in t/30long.t, which might be related to the database in use

I also found warnings that need suppression in DBI.pm

[Tux]

perl5-dbi/dbi@b60d376 is in, so that should/could be a new baseline

[twata1]

FYI:
rt85886 itself appears to have been moved to #86 .

[Tux]

#186 was merged
I think this ticket should be closable now

[djzort]

Please try out 1.91_1 from cpan https://metacpan.org/release/ZARQUON/DBD-Oracle-1.91_1

Hopefully it has fixed more problems than it solved.

As I always recommend, don't just blindly update DBD::Oracle in your application!

Please mirror the github repo in to your org's local git and set up some CI around it to test it against your specific configuration (Linux or other OS, Oracle DB version, SDK versions etc)

Whilst we have github actions with some actual testing against oracle - there is just too many versions and combinations to do better than provide indicative test results (albeit very strongly indicative) - enhancing the github actions to cater to your needs is entirely appreciated and PR will gladly be welcomed.