Re: Scala 2.7.6 final

On Mon, Sep 14, 2009 at 06:37:00PM +0200, martin odersky wrote:
> I still stand by the withdraw advice. The patch looks highly risky to
> me. I would feel uncomfortable to have this in the open.

Only in the interests of clarifying and not blame-shifting (and I do
commit plenty of code which might break something and for which I'd
assume full responsibility as the only person ever to have looked at it)
in THIS case I was well aware I didn't understand the issue, and so we
sent it to you for review:

https://lampsvn.epfl.ch/trac/scala/ticket/1896

Wherein we see rytz comment "OK, Martin agrees that r17719 is the
correct fix." paulp is very much in favor of not committing broken code,
but when it comes to the more challenging parts of the compiler he is to
some extent at the mercy of wiser folks.

Also, as the author of that patch I would have very strongly objected to
rolling it into a release without far more intense scrutiny, had I known
that was happening before it happened. Of course I would say that about
any patch (because I know the scala build process too well to believe
there is any such thing as a totally "safe" patch) but this particular
patch would have earned 10x the emphasis.

> 1) Trunk is by now too far removed from 2.7 to be able to make
> backpatches with confidence.

I don't think you should exactly withdraw that; I mean it's possible,
but not without much more serious review and testing than such patches
have received in the past.

On Mon, Sep 14, 2009 at 6:54 PM, Paul Phillips wrote:
> On Mon, Sep 14, 2009 at 06:37:00PM +0200, martin odersky wrote:
>> I still stand by the withdraw advice. The patch looks highly risky to
>> me. I would feel uncomfortable to have this in the open.
>
> Only in the interests of clarifying and not blame-shifting (and I do
> commit plenty of code which might break something and for which I'd
> assume full responsibility as the only person ever to have looked at it)
> in THIS case I was well aware I didn't understand the issue, and so we
> sent it to you for review:
>
>  https://lampsvn.epfl.ch/trac/scala/ticket/1896
>
> Wherein we see rytz comment "OK, Martin agrees that r17719 is the
> correct fix." paulp is very much in favor of not committing broken code,
> but when it comes to the more challenging parts of the compiler he is to
> some extent at the mercy of wiser folks.
>
> Also, as the author of that patch I would have very strongly objected to
> rolling it into a release without far more intense scrutiny, had I known
> that was happening before it happened.  Of course I would say that about
> any patch (because I know the scala build process too well to believe
> there is any such thing as a totally "safe" patch) but this particular
> patch would have earned 10x the emphasis.
>
>> 1) Trunk is by now too far removed from 2.7 to be able to make
>> backpatches with confidence.
>
> I don't think you should exactly withdraw that; I mean it's possible,
> but not without much more serious review and testing than such patches
> have received in the past.
>
All agreed. I should have scrutinized the patch more closely.
I'll need to find the time to get to the bottom of that issue.

Cheers

I'm not a tooling expert, but I'd caution against "withdrawing". There's all manner of tools (Maven, Eclipse plugin update site, etc) that are likely to be unhappy about disappearing versions. It'd probably be safer to release 2.7.6.1 or 2.7.7

--j

On Mon, Sep 14, 2009 at 9:37 AM, martin odersky <martin.odersky@epfl.ch> wrote:

On Mon, Sep 14, 2009 at 6:09 PM, martin odersky <martin.odersky@epfl.ch> wrote:
> I draw the following conclusions from this debacle:
>
> 1) Trunk is by now too far removed from 2.7 to be able to make
> backpatches with confidence.

Actually, I have to withdraw that. Talking with Toni and having a look
at the patch
it looks likely to me that lift would have the same problems with
trunk. In fact, the patch does not look sound to me. It clearly
violates the intent written in the comments.
So, we should withdraw the patch, get to the root of the problem, and
once the real patch is found, try again.

> 2) We should withdraw 2.7.6. immediately and let 2.7.5 be the last in its line.
>
I still stand by the withdraw advice. The patch looks highly risky to
me. I would feel uncomfortable to have this in the open.

> 3) We should instead concentrate in getting 2.8 out of the door.
>
That's still true, but if we have the existential types patch we could
try lift with it.
Let's do that before pushing out a release.

Cheers

Yes.  It's actually already  too late. 2.7.6 is on maven central and won't be updated.

On Mon, Sep 14, 2009 at 6:26 PM, Jorge Ortiz <jorge.ortiz@gmail.com> wrote:

I'm not a tooling expert, but I'd caution against "withdrawing". There's all manner of tools (Maven, Eclipse plugin update site, etc) that are likely to be unhappy about disappearing versions. It'd probably be safer to release 2.7.6.1 or 2.7.7

--j

On Mon, Sep 14, 2009 at 9:37 AM, martin odersky <martin.odersky@epfl.ch> wrote:

On Mon, Sep 14, 2009 at 6:09 PM, martin odersky <martin.odersky@epfl.ch> wrote:
> I draw the following conclusions from this debacle:
>
> 1) Trunk is by now too far removed from 2.7 to be able to make
> backpatches with confidence.

Actually, I have to withdraw that. Talking with Toni and having a look
at the patch
it looks likely to me that lift would have the same problems with
trunk. In fact, the patch does not look sound to me. It clearly
violates the intent written in the comments.
So, we should withdraw the patch, get to the root of the problem, and
once the real patch is found, try again.

> 2) We should withdraw 2.7.6. immediately and let 2.7.5 be the last in its line.
>
I still stand by the withdraw advice. The patch looks highly risky to
me. I would feel uncomfortable to have this in the open.

> 3) We should instead concentrate in getting 2.8 out of the door.
>
That's still true, but if we have the existential types patch we could
try lift with it.
Let's do that before pushing out a release.

Cheers

On 14 Sep 2009, at 23:44, Josh Suereth wrote:

Yes.  It's actually already  too late. 2.7.6 is on maven central and won't be updated.

On Mon, Sep 14, 2009 at 6:26 PM, Jorge Ortiz <jorge.ortiz@gmail.com> wrote:

I'm not a tooling expert, but I'd caution against "withdrawing". There's all manner of tools (Maven, Eclipse plugin update site, etc) that are likely to be unhappy about disappearing versions. It'd probably be safer to release 2.7.6.1 or 2.7.7

It would probably make sense to roll forward, and release 2.7.7 even if it's the same as 2.7.5 to execute the roll-back of the patch. That way, any auto-updating (or forward looking) dependencies can handle the fact if they've already downloaded 2.7.6.
Alex