[Glass] Mixing Unicode and String instances in Indexes [WAS] Re: Further commits have been disabled for this session because: 'CorruptObj error'. This session must logout.

Wed Oct 28 11:50:56 PDT 2015

On Wed, Oct 28, 2015 at 3:31 PM, Mariano Martinez Peck <
marianopeck at gmail.com> wrote:

> ufff I could not reproduce it as a test case. However, I paste it here in
> case you see something obvious I did wrong:
>
> 1) $GS_HOME/bin/createStone -f gs_3106 3.1.0.6
>
> 2) Via tODE in gs_3106 execute:
>
> | collection |
> collection := RcIdentityBag new.
> collection createEqualityIndexOn: 'testSelector' withLastElementClass:
> String.
> collection add: (TestCase new setTestSelector: 'aString').
> collection add: (TestCase new setTestSelector: 'aUnicode7' _asUnicode7).
> UserGlobals at: #IndexesBugReport put: collection.
> Object new assert: (collection select: { :each | each.testSelector =
> 'aString' }) size = 1.
> System commitTransaction.
>
> 3) $GS_HOME/sys/local/bin/upgradeStone -f gs_3106 gs_329 3.2.9
>
> 4) Via tODE in gs_329 execute:
>
> Object new assert: ((UserGlobals at: #IndexesBugReport) select: { :each |
> each.testSelector = 'aString' }) size = 1.
>
> But it did work... so something else may be going on.
> ufff...
>

5) At least with this code I can reproduce the error I got once checking
the status of the index:

| collectionsWithBadIndexes |
IndexManager current removeAllIncompleteIndexes.
collectionsWithBadIndexes := IndexManager current nscsWithBadIndexes.
Transcript show: 'There are ' , collectionsWithBadIndexes size asString, '
collections with bad indexes.'; cr.
Transcript show: 'The following are the OOPs of such collections: '.
collectionsWithBadIndexes do: [ :each | Transcript show: ' - ' , each asOop
asString.  ].
Transcript cr.

What I cannot reproduce is the corrupt obj error from seaside...

>
>
> On Wed, Oct 28, 2015 at 3:03 PM, Mariano Martinez Peck via Glass <
> glass at lists.gemtalksystems.com> wrote:
>
>>
>>
>> On Wed, Oct 28, 2015 at 2:56 PM, Dale Henrichs <
>> dale.henrichs at gemtalksystems.com> wrote:
>>
>>>
>>>
>>> On 10/28/2015 10:48 AM, Mariano Martinez Peck wrote:
>>>
>>>
>>>
>>> On Wed, Oct 28, 2015 at 2:32 PM, Dale Henrichs <
>>> dale.henrichs at gemtalksystems.com> wrote:
>>>
>>>> I guess the exact unicode error message would be a starting point ...
>>>>
>>>>
>>> Hi Dale,
>>>
>>> Ok, I opened a new thread, as you asked.
>>> OK I will try to recover it (I must recover from backup so give me a
>>> few).
>>>
>>>
>>>
>>>> This whole thread started because we had a bug that would throw a
>>>> corruptObj message when invalid unicode was fed to a unicode primitive ...
>>>> that bug was fixed in 3.2.2 or so ...
>>>>
>>>> So this corruptObj message is occuring for a different reason
>>>> altogether ...
>>>>
>>>> In general you should not have to rebuild indexes across upgrade
>>>> boundaries (unless we document that you have to) ....
>>>>
>>>> While it is good to know that a rebuild of indexes fixed you problem,.
>>>> I am still interested in understanding what caused the error in the first
>>>> place, since it may actually be another bug ...
>>>>
>>>>
>>> Dale, a few things:
>>>
>>> 1) The original stack / code that triggered the real error was NEVER
>>> show anywhere. From the "button" I clicked in my app, which caused the
>>> problem, I can tell you 99% that the error ocurred when I performed a query
>>> using the special syntax ( {   }  ) for accessing an index of a collection.
>>> The fact of not seeing the REAL stack, WAS a problem.
>>>
>>> and that is another reason for me to get to the bottom of this ...
>>>
>>
>> Yes, this was a bit of a pain not being able to get the original error.
>>
>>
>>>
>>>
>>> 2) From what I understand, I DID have to rebuild my indexes because in
>>> the documentation it said that instances of Unicode7 cannot coexist
>>> together with isntance of String in the same index. Let's say I have an
>>> index on a instVar called 'tickerSymbol' of class XXX. I had a
>>> RcIdentitySet with instances of XXX in which some instances have a String
>>> instance in 'tickerSymbol' and others had a Unicode7 as instance. (why?  no
>>> clue...).
>>> In the documentation I read this was not possible anymore and if this
>>> were the case I would need to create a Unicode-kind of index. And that's
>>> what I did and what it worked.
>>>
>>> okay so that information was true ... and we did document it ... so we
>>> are down to trying to understand where the  error that is causing the
>>> corruptObj error on commit
>>>
>>>
>>>
>> Indeed.
>>
>>
>>> I think the bug (if it IS a bug..and if it is reproducible) should be
>>> very easy to reproduce with tODE and gsDevKit
>>>
>>> 1) create 3.1.0.6 stone called gs_3106
>>> 2) Create dummy ClassWithIndex with instVar 'instVarWithIndex'.
>>> 3) Create a equality index over instVar 'instVarWithIndex' with path
>>> element class String (old API since we are in 3.1.0.6)
>>> 4) Store in #myUserGlobals a RcIdentityBag with 2 instances of
>>> ClassWithIndex, one with a String in 'instVarWithIndex' and one with a
>>> Unicode7 instance.
>>> 5) Use `upgradeStone` (forcing a target 3.2.9 stone called gs_329) to
>>> upgrade from gs_3106 to gs_329
>>> 6) Make a #select: { } using 'instVarWithIndex' in the RcIdentityBag
>>> stored in the userglobals.
>>>
>>>
>>> This is enough for me to take a crack at reproducing the problem ...
>>> these are the details that I needed ... thanks ...
>>>
>>> I've got a full plate right now, so I will try to get this reproduced
>>> and then perhaps defer on characterizing as the fix will likely not be a
>>> simple Smalltalk-based solution ... nonetheless I want to get this one
>>> under the microscope ... thanks!
>>>
>>
>> ok, no problem. I will see if I can reproduce it too.
>>
>>
>> --
>> Mariano
>> http://marianopeck.wordpress.com
>>
>> _______________________________________________
>> Glass mailing list
>> Glass at lists.gemtalksystems.com
>> http://lists.gemtalksystems.com/mailman/listinfo/glass
>>
>>
>
>
> --
> Mariano
> http://marianopeck.wordpress.com
>

-- 
Mariano
http://marianopeck.wordpress.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gemtalksystems.com/mailman/private/glass/attachments/20151028/a46ebe76/attachment-0001.html>