in reply to Re^8: Strawberry Perl 64bit on Win7 crashes on DBI-function
in thread Strawberry Perl 64bit on Win7 crashes on DBI-function

Back at work.

What i found out:

There's really nothing wrong with the ODBC-DSNs or the descriptions (apart from there being so many on our machines...). When i raise the buffer size of the cmd-window ("dos-box"), the whole list of the 310 DSNs actually appears. But (probably upon returning from some function) the interpreter still crashes.

To make clear: After the complete list is retrieved it crashes. Only when I try to pipe the output to a file (to post here) it does not generate the complete list (eben if i add a "$| = 1").

The longest description is 54 characters wide: /Excel Files/ (11), /Microsoft Excel Driver (*.xls, *.xlsx, *.xlsm, *.xlsb)/ (54) and should be a standard DSN on windows with office installed.

I am out of ideas for now.

Replies are listed 'Best First'.
Re^10: Strawberry Perl 64bit on Win7 crashes on DBI-function
by mje (Curate) on Apr 24, 2014 at 10:53 UTC

    If you get the full list and it segfaults (or whatever) afterwards it does sound like memory/stack corruption but assuming you are using the new version I sent you yesterday I cannot explain why.

    Not sure it will really help but you can limit the number of DSNs returned by adding something like:

    if (numDataSources > 100) {
   break;
}

    [download]

    just after fDirection = SQL_FETCH_NEXT. If you restrict the number returned and the crash goes away you might be able to identify the entry that causes corruption.

[reply]
[d/l]
      Ok. 280 is the magic number. It does not crash with 
      if (numDataSources > 280) {
   break;
}.

      [download]
      And it does crash if i change to 
      if (numDataSources > 281) {
   break;
}.

      [download]
      (The first DSN actually has numDataSources == 0).

      Update:

      There is nothing wrong with the next DSN, when i go to 281 the next one is still retrieved (like with all 310 entries) but it crashes still.

      The last lines of the output are in that case (before crashing):

      (I added "numDataSources" in the output as you can see:) 
      /(276)/, /DF203BPA/ (8), /IBM DB2 ODBC DRIVER - DB2RTCL9N/ (31)
/(277)/, /DF203BPE/ (8), /IBM DB2 ODBC DRIVER - DB2RTCL9N/ (31)
/(278)/, /DF203BS/ (7), /IBM DB2 ODBC DRIVER - DB2RTCL9N/ (31)
/(279)/, /DF203BPM/ (8), /IBM DB2 ODBC DRIVER - DB2RTCL9N/ (31)
/(280)/, /DF203CEI/ (8), /IBM DB2 ODBC DRIVER - DB2RTCL9N/ (31)
/(281)/, /DF203CMN/ (8), /IBM DB2 ODBC DRIVER - DB2RTCL9N/ (31)

      [download]
[reply]
[d/l]
[select]

      I am now thinking hard about this line in "ODBC.xs":

      ST(numDataSources++) = newSVpv(dsn, dsn_length+9 /* strlen("dbi:ODBC:") */ );

      Not sure if this always does what one expects on first glance. I am refering to the left side of the "=".

[reply]
[d/l]

        That puts the scalars on the stack. I've rewritten data_sources to not use the stack. See my other comment.

[reply]
      Ok. I'm playing with that now. It works without crash(!) for 10 and 100. I am testing further.

      I did notice that i'm getting this warning though on the "dmake"(probably have overlooked that yesterday): 

      dbdimp.c: In function 'odbc_db_STORE_attrib':
dbdimp.c:4677:22: warning: cast to pointer from integer of different s
+ize [-Wint-to-pointer-cast]
dbdimp.c:4679:22: warning: cast to pointer from integer of different s
+ize [-Wint-to-pointer-cast]
dbdimp.c:4937:18: warning: cast to pointer from integer of different s
+ize [-Wint-to-pointer-cast]
dbdimp.c:4942:18: warning: cast to pointer from integer of different s
+ize [-Wint-to-pointer-cast]
dbdimp.c:4962:27: warning: cast to pointer from integer of different s
+ize [-Wint-to-pointer-cast]

      [download]

      Could be interesting.
[reply]
[d/l]

        You can ignore those. Could I email you a new DBD::ODBC package to try? If so msg me your email address.

[reply]

In Section Seekers of Perl Wisdom