de underscoring and the SqueakMap Package Loader

Want to guess how many SqueakMap Packages broke when the underscore assignment was nixed?

Want to guess how many load just fine after executing

        Preferences enable: #allowUnderscoreAssignment.

?-)

Cheers,
   - Andreas

[hidden email] wrote:
> Want to guess how many SqueakMap Packages broke when the underscore
> assignment was nixed?
>
>

There's a lot of lore to communicate.  I know from direct experience  
that it costs on the order of $1M to document this quantity of  
material professionally.

I've started a page on the wiki to collect such information: http://
croquet.doit.wisc.edu/wiki/tiki-index.php?page=Lore

It now has one entry: UnderScore.  I encourage folks to add more.


On May 2, 2006, at 8:48 PM, Andreas Raab wrote:

Andreas Raab wrote:

Want to guess how many load just fine after executing
    Preferences enable: #allowUnderscoreAssignment.
?-)
Cheers,
  - Andreas

[hidden email] wrote:

Want to guess how many SqueakMap Packages broke when the underscore assignment was nixed?

Thanks! [ sincerely :-) ]  I appreciate all the hard work that went into getting Croquet out, but no, I don't want to guess.  I want to know that they will load just fine without me doing anything.

Here is my reasoning:

1. Upward compatibility is a solid working principle in any good environment.
   
2. Consider of the consequences of all those out there that we would like to invite in  who will stumble on the problem first time they use the SqueakMap.  Many will turn around a go somewhere else...especially those that have little time to check closer and who "know better".

My discovery of it came about from having started on David Shafer's Squeak/Seaside Tutor which requires 5 packages from the SqueakMap.  I decided to bring these with me to the Croquet image.  Of the 5 packages only 1 made it.  GOODS, BBTree, Refactoring Browser, and OPProcess all failed because of the assign problem.  Shout succeeded.  That suggests that about 80% of the packages may fail on import to Croquet from SqueakMap.

The odds are terrible and we are going to lose a lot of newbees the first time they go into the Sqeak Package Loader.  Its a good idea to do something about that problem.

Hi John -

> Thanks! [ sincerely :-) ]

You're welcome (sincerely; I was just making fun of the form of your
question :)

> I appreciate all the hard work that went into
> getting Croquet out, but no, I don't want to guess.  I want to know that
> they will load just fine without me doing anything.

You mean besides explicitly abandoning the warning that "this package
has no published release for your Squeak version"? ;-) Isn't that ample
hint that there may be problems with loading it? ;-)

> Here is my reasoning:
>
>    1. Upward compatibility is a solid working principle in any good
>       environment.

Certainly. That's why the preference is in there. Whether it should be
on by default or not is a different question altogether. David and I
both felt strongly about this issue in such that leaving the preference
on would reintroduce the underscore through the back door since there is
no forcing function any longer and by the end of the day we'd end up
with the same mess that we were trying to avoid.

>    2. Consider of the consequences of all those out there that we would
>       like to invite in  who will stumble on the problem first time they
>       use the SqueakMap.  Many will turn around a go somewhere
>       else...especially those that have little time to check closer and
>       who "know better".

You have a point here. I thought about whether to include SM in the
first place and hoped that it would be mostly a good thing; perhaps that
is a mistake (easy enough to fix though)

> My discovery of it came about from having started on David Shafer's
> Squeak/Seaside Tutor which requires 5 packages from the SqueakMap.  I
> decided to bring these with me to the Croquet image.  Of the 5 packages
> only 1 made it.  GOODS, BBTree, Refactoring Browser, and OPProcess all
> failed because of the assign problem.  Shout succeeded.  That suggests
> that about 80% of the packages may fail on import to Croquet from SqueakMap.

The sample size seems a bit small ;-)

> The odds are terrible and we are going to lose a lot of newbees the
> first time they go into the Sqeak Package Loader.  Its a good idea to do
> something about that problem.

Well, there are a few possible solutions I can see off-hand. For one
thing, we can ask people not to ignore the warnings they're getting and
to load packages that were actually published for Croquet (which
includes underscore-fixing). Second, we can simply remove the SqueakMap
loader which avoids the problem once and for all and reintroduce it when
we have a CroquetMap which only keeps Croquet packages. Third, we can
try to do something about loading these packages (e.g., rewrite the code
on the fly or somesuch). Depending on how much time you want to spend to
either lobby/hack/document the problem it's your pick ;-)

Cheers,
   - Andreas

Andreas Raab wrote:

Hi John -

[SNIP]

The sample size seems a bit small ;-)

hmmm...the sample is small, but it isn't intentionally biased, but 4 out of 5...how much are you willing to put down that most of the other packages aren't broken for the same reason?

The odds are terrible and we are going to lose a lot of newbees the first time they go into the Sqeak Package Loader.  Its a good idea to do something about that problem.

Well, there are a few possible solutions I can see off-hand. For one thing, we can ask people not to ignore the warnings they're getting and to load packages that were actually published for Croquet (which includes underscore-fixing). Second, we can simply remove the SqueakMap loader which avoids the problem once and for all and reintroduce it when we have a CroquetMap which only keeps Croquet packages. Third, we can try to do something about loading these packages (e.g., rewrite the code on the fly or somesuch). Depending on how much time you want to spend to either lobby/hack/document the problem it's your pick ;-)

I apologize my first note.  It was a little flippant, I have to catch myself sometimes. 

There is no inherent problem with SqueakMap being around as long as Monticello is the official single repository source.  This being the case, SqueakMap will have no effect on the Croquet version thread except inform it about other parts of the Smalltalk world.  In the mean time, SqueakMap has a number tools which are not yet available in Monticello and are potentially useful.  When it is worth the effort, they will port.

Cheers,
  - Andreas