9049 - update stops working with very large attribute lists

Status: RESOLVED FIXED

(Directory Graveyard :: PerLDAP, defect, P1)

Description

[Kevin McCarthy]

Posted in newsgroup:

I am using Perl 5.004_04, PerLDAP 1.22 on Solaris 2.51.  I've written
some administrative programs to manipulate groupOfUniqueNames
objectclasses in LDAP.  [These are used to restrict access to parts of a
web server using Enterprise Server].

My programs have been working, until today.  It seems I've hit a size
limit of some kind.  I was trying to remove a uniquemember entry from an
object with 4085 entries in it.  I tracked the problem down to a line in
Entry.pm, removeValue(), line 368:
   @{$self->{"_self_obj_"}->{"_${attr}_save_"}} = @{$self->{$attr}}
unless
      defined($self->{"_${attr}_save_"});

This line was failing.  The _${attr}_save_ did not receive the values
from $attr.  It has a size of 0 after the assignment (and should have
been 4085)

I changed the line to read:
   $self->{"_self_obj_"}->{"_${attr}_save_"} = [ @{$self->{$attr}} ]
unless
      defined($self->{"_${attr}_save_"});

This version worked.  I'm assuming this is a Perl bug, something to do
with array assignment as opposed to populating a reference to an array
can hold more values.  Since this second version is working for me, I
would like to get some feedback, and if favorable, merge the change into
the code.  This change would need to be made in about 5 places in
Entry.pm.

Thanks,

Kevin McCarthy
[e-mail]

Comment 1

[Kevin McCarthy]

Attachment: Patch file against 1.2.3

Comment 2

[Leif Hedstrom]

Leif Hedstrom <[e-mail]> writes:
>Nick Ing-Simmons wrote:
>
>> I'd suggest changing the default in the Config.pm that you ship to
>> default to -Od (or whatever the "safest" value is).
>> Reasoning here is that those that have the expertise to dork with
>> this stuff can still dork it back to -O2, while those that don't are
>> more likely to get something that works "out of the box".
>
>Ahh, makes a lot of sense. Thanks for all the suggestions, I'm going to
>try it again as soon as I possibly can. Does anyone know if this
>optimization bug is in VC++ 5.0 only, or is it in 6.0 as well?

There is more than one bug. For ages VC++ could not build perl with -O2.
Then some service pack to VC5 allowed that, but still could not build
Tk.

FWIW (and it ain't worth what I paid for it) VC6.0 Professional _does_
seem to be able to build Tk800.014 with -O2.


>I remember
>having a similar problem when compiling Perl on NT with VC++ v5.0, and
>setting optimization to just /O (I think) helped (otherwise the build
>wouldn't even pass Perl's regression tests...).
>
>Btw, how hard is it to create a binary install package that I can send to
>my ActivePerl users?

Quite easy. Tinker with top-level Makefile.PL then tar up the blib
attached is mail that helped me.

--
Nick Ing-Simmons <[e-mail]>
Via, but not speaking for: Texas Instruments Ltd.

Comment 3

[Leif Hedstrom]

IGNORE THE LAST COMMENT, I PUT IT IN THE WRONG BUG...

Comment 4

[Leif Hedstrom]

Fixed as suggested.