DBI.Connection call yields "Object: nil error: did not understand #atEnd" in ST 3.2.91-b98173d

>> On 29 Aug 2015, at 23:53, Mark Bratcher <[hidden email]> wrote:
>>
>> The exact same connection works fine under Smalltalk 3.2.5. So it's not an authentication issue in my case. I tried it on a couple of different databases, each time working in 3.2.5, but the same error using 3.2.91-b98173d.
>
> that is odd. The difference from 3.2.5 to master in dbd-mysql itself is that I removed
> commitTransaction/rollbackTransaction because the super class already had these
> selectors.
>
> If it isn’t too much work for you could you do the following (once fro 3.2.5 and once
> for master):
>
> * Use wireshark/tcpdump to record a packet trace of the GST/MySQL communication
> (maybe change the PW to something different..)
> * Show the code that was used
> * And the exception?
>
>
> kind regards
> holger

>> The socket is connecting fine. But digging into the connection code, I found that a `class new` operation now automatically calls the `initialize` method in 3.2.91-b98173d, but it does not do so in 3.2.5. This causes a problem in the MySQLPacket and MySQLInputPacket connection code and results in `MySQLInputPacket >> initialize` being called too early in `^(self new) stream: aStream ; initialize` inside of ` MySQLPacket class >> on: aStream`. It's called automatically on `self new` before the stream is set.
> I tried to align this semantic to Pharo (as porting from/to Pharo gets easier). Thank
> you for your analysis! The question is why does the CI pass? Did I by accident not
> go through authentication?
>
> I will try to reproduce and integrate a fix this week.
>
> holger
>
>

>
> > On 31 Aug 2015, at 12:28, Mark Bratcher <[hidden email]> wrote:
> >
> > Hi Holger
>
> Hi,
>
> > Thanks again for being so responsive.
> >
> > After doing some admittedly light research on the `new` method versus
> `initialize`, it looks like Pharo (and probably, therefore, Squeak) is the
> only variant of Smalltalk that automatically calls an instance initializer
> (`initialize`) on `new`. If GNU Smalltalk is following this to be easier to
> port from Pharo, that raises a philosophical question for GNU Smalltalk: is
> intended for it to align as much as possible with the Pharo implementation
> (and then perhaps, ultimately, become a "Pharo variant"), or to attempt to
> remain more "pure" (whatever that might mean :)) relative to Smalltalk-80?
> I noticed in various texts discussion Smalltalk class instance creation,
> specifically show a pattern something like:
>
>
> it is a pragmatic decision. The GNU Smalltalk community is not very large
> and
> there are not many projects that get created for GNU Smalltalk (e.g. Iliad
> was a
> notable exception). With gst-convert we have a tool to convert from other
> dialects
> but in recent times I think I/we only ported from Pharo.
>
> I don’t think there is intention to be “Pharo compatible”. E.g.
> String/Symbol will
> not be considered equal. There is no plans to introduce a ProtoObject and
> maybe
> not even the “MetaLink”.. at least not to kernel/
>
> At the same time I started to use the >>#new/#initialize pattern as well
> so it fellt
> like a natural progress.
>
> kind regards
>         holger
>
>
>
>
>
>
>

>
> > On 31 Aug 2015, at 14:08, Mark Bratcher <[hidden email]> wrote:
> >
> > Thank you, Holger.
> >
> > I understand the pragmatics of the decision. One could argue that, in
> the current bug case, the `(self new) stream: aStream ; initialize`
> statement is a bit "off-pattern". :) I think it's a little unfortunate that
> Pharo deviated from Smalltalk-80 in a small, but fundamental way. I think
> it means that if I want to write more portable Smalltalk code, I might
> avoid using "initialize" since its behavior will be different in Pharo (and
> now GNU) versus other implementations. Such avoidance would be necessary in
> cases where one wants instance initialization to be deferred after instance
> creation.
>
> Yes, it is unfortunate. But if you look at the diff that introduced this
> behavior then
> you see that it was used a lot already. In both Pharo and GST there is
> >>#basicNew
> as well but I don’t know if this is in an ANSI protocol.
>
> So e.g. this could fix the issue you have debugged:
>
> diff --git a/packages/dbd-mysql/Connection.st
> b/packages/dbd-mysql/Connection.st
> index b97dc89..ef5ac65 100644
> --- a/packages/dbd-mysql/Connection.st
> +++ b/packages/dbd-mysql/Connection.st
> @@ -237,7 +237,7 @@ Object subclass: MySQLPacket [
>
>      MySQLPacket class >> on: aStream [
>         <category: 'instance creation'>
> -       ^(self new)
> +       ^(self basicNew)
>             stream: aStream;
>             initialize
>      ]
>
>
>