About the 3.5/3.6 decision, sounds reasonable to me.
About why we got to the state where one of us says a bug critical enough to stop the release from going out, when we're in gamma - I think we included the fix with the implicit goal of "fixing the class builder" but with the explicit SUnit test of "classes with instance variables and subclasses can be reshaped properly".
Nobody said "this fix does X, which is enough. It doesn't do Y, but we can live with that", where Y would be "allow the whole hierarchy to get rebuilt", for example.
We need more explicit reviews.
Daniel
Doug Way dway@riskmetrics.com wrote:
Daniel Vainsencher wrote:
<shifting to sqf list>
That's an interesting point. Does that mean we actually delay 3.5, when the 3.6 update stream has already started? another question is, how did we get into gamma, when apparently, one of our two goals for the release isn't actually achieved?
Tim did mention this as a bug on March 14. There wasn't any mention of whether it was a serious enough problem to try to fix in 3.5, so I ignored it for the moment, hoping that someone would come up with a fix.
It did appear to be possibly related to the ClassBuilder problem, but that wasn't certain, either. The problem occured with or without the ClassBuilder fix.
Just now I did a quick check of when the bug was introduced. I was guessing that maybe Andreas introduced it with the ClassBuilder refactoring/cleanup in 3.4alpha, which was when the other ClassBuilder bug was introduced. However, that was not the case... I tested 3.2 and the bug is there, too. Turns out this bug is as old as the hills... the bug exists back in 2.6! It does not exist in 2.4, though. (It's always nice to have old images lying around. :-) I didn't have a 2.5 image handy, though.)
So, given that the bug is 3+ years old, and we don't yet have a fix that we agree on, I think it's pretty safe to say that we don't need to address this in 3.5. As soon as someone comes up with a good fix, we can include it in 3.6alpha. (Maybe Brent's fix is sufficient, I don't know, but it hasn't gotten any feedback yet.)
- Doug Way
Daniel
Tim Rowledge tim@sumeru.stanford.edu wrote:
Daniel Vainsencher danielv@netvision.net.il wrote:
AFAICT, this fix is KCP territory. Unless there is a very good reason otherwise, I would await their recommendation (and not delay the release).
Well since the 3.5 release was purported to be mostly to include a fix for a very similar bug and the bug in question has pretty similar effects - cannot recompile a number of classes - I'd say that without a fix we really shouldn't even consider 3.5 in beta in any meaningful sense.
tim
Tim Rowledge, tim@sumeru.stanford.edu, http://sumeru.stanford.edu/tim Spellchecker not found. Press -- to continue ...
Squeakfoundation mailing list Squeakfoundation@lists.squeakfoundation.org http://lists.squeakfoundation.org/listinfo/squeakfoundation
Squeakfoundation mailing list Squeakfoundation@lists.squeakfoundation.org http://lists.squeakfoundation.org/listinfo/squeakfoundation
We need more explicit reviews.
Or more explicit goals ;-) The fix fixed exactly the bug which was reported, namely that the meta class hierarchy doesn't get recompiled correctly. Nowhere the fix claimed that it "fixes class builder" because there is no definition for it unless you identify the bugs. A word about the SUnit tests I wrote - I wanted this to be a start for more elaborate tests and therefore I wrote a bunch of tests which ensure critical invariants of ClassBuilder.
To be more explicit: Those tests actually validate if the reshaping machinery maps instance state correctly. It is not (never has been, never was intended to) a test for recompiling the entire class hierarchy. The original reason for this was that on a (non-fixed) system, failing the test of recompiling the meta class hierarchy would mean your system is @!#^ed. Nothing I would expect from a test - and therefore this test can only be added once the bug is fixed.
Cheers, - Andreas
-----Original Message----- From: squeakfoundation-bounces@lists.squeakfoundation.org [mailto:squeakfoundation-bounces@lists.squeakfoundation.org] On Behalf Of Daniel Vainsencher Sent: Saturday, April 05, 2003 3:14 AM To: Discussing the Squeak Foundation Subject: [Squeakfoundation]WeakArray bug (was Re: [UPDATES] 3.5gamma)
About the 3.5/3.6 decision, sounds reasonable to me.
About why we got to the state where one of us says a bug critical enough to stop the release from going out, when we're in gamma - I think we included the fix with the implicit goal of "fixing the class builder" but with the explicit SUnit test of "classes with instance variables and subclasses can be reshaped properly".
Nobody said "this fix does X, which is enough. It doesn't do Y, but we can live with that", where Y would be "allow the whole hierarchy to get rebuilt", for example.
We need more explicit reviews.
Daniel
Doug Way dway@riskmetrics.com wrote:
Daniel Vainsencher wrote:
<shifting to sqf list>
That's an interesting point. Does that mean we actually
delay 3.5, when
the 3.6 update stream has already started? another
question is, how did
we get into gamma, when apparently, one of our two goals
for the release
isn't actually achieved?
Tim did mention this as a bug on March 14. There wasn't
any mention of
whether it was a serious enough problem to try to fix in
3.5, so I ignored it
for the moment, hoping that someone would come up with a fix.
It did appear to be possibly related to the ClassBuilder
problem, but that
wasn't certain, either. The problem occured with or
without the ClassBuilder
fix.
Just now I did a quick check of when the bug was
introduced. I was guessing
that maybe Andreas introduced it with the ClassBuilder
refactoring/cleanup in
3.4alpha, which was when the other ClassBuilder bug was
introduced. However,
that was not the case... I tested 3.2 and the bug is there,
too. Turns out
this bug is as old as the hills... the bug exists back in
2.6! It does not
exist in 2.4, though. (It's always nice to have old images
lying around. :-)
I didn't have a 2.5 image handy, though.)
So, given that the bug is 3+ years old, and we don't yet
have a fix that we
agree on, I think it's pretty safe to say that we don't
need to address this
in 3.5. As soon as someone comes up with a good fix, we
can include it in
3.6alpha. (Maybe Brent's fix is sufficient, I don't know,
but it hasn't
gotten any feedback yet.)
- Doug Way
Daniel
Tim Rowledge tim@sumeru.stanford.edu wrote:
Daniel Vainsencher danielv@netvision.net.il wrote:
AFAICT, this fix is KCP territory. Unless there is a
very good reason
otherwise, I would await their recommendation (and
not delay the
release).
Well since the 3.5 release was purported to be mostly
to include a fix
for a very similar bug and the bug in question has
pretty similar
effects - cannot recompile a number of classes - I'd
say that without a
fix we really shouldn't even consider 3.5 in beta in
any meaningful
sense.
tim
Tim Rowledge, tim@sumeru.stanford.edu,
http://sumeru.stanford.edu/tim
Spellchecker not found.
Press -- to continue ...
Squeakfoundation mailing list Squeakfoundation@lists.squeakfoundation.org http://lists.squeakfoundation.org/listinfo/squeakfoundation
Squeakfoundation mailing list Squeakfoundation@lists.squeakfoundation.org http://lists.squeakfoundation.org/listinfo/squeakfoundation
Squeakfoundation mailing list Squeakfoundation@lists.squeakfoundation.org http://lists.squeakfoundation.org/listinfo/squeakfoundation
squeakfoundation@lists.squeakfoundation.org