Re: [gentoo-dev] [RFC] Dropping PyPy(3) back to ~arch

2021-03-03 Thread Michał Górny
On Wed, 2021-03-03 at 23:00 +, Sam James wrote:
> > On 2 Mar 2021, at 15:05, Michał Górny  wrote:
> > 
> > Hi,
> > 
> > I think I've made a mistake by stabilizing PyPy(3), and I would like
> > to
> > drop it back to ~arch (and stable-mask the relevant flags).
> > 
> > Roughly, there are 4 problems with it:
> > 
> > [snip]
> 
> > Honestly, I've tried to improve PyPy in the past but I don't really
> > have
> > time or motivation to do this continuously.  PyPy is an interesting
> > project, and it has its isolated uses.  However, I don't think it's
> > ready as a general-purpose Python interpreter for production
> > environments.
> > 
> > I don't really want to remove it entirely or revert all the work
> > we've
> > put into testing packages with it -- but I think we should move it
> > to
> > ~arch at the very least.
> 
> It’s unfortunate but no objections. News item may be worthwhile
> to explain to users why & how for posterity, but not necessary.

Yeah, I'm still not sure what to do.

What is certain is that we need to do something because current PyPy
versions are vulnerable, and two-slots-in-one-target approach is not
sustainable.

Option 1. is to stabilize the new versions, including alpha pypy3.7
(test-restricted).  This will make it possible to delay the final
decision.  However, it would imply that stable users rebuild everything
to upgrade pypy3 and that would suck if it would mean it goes ~arch
anyway.

Option 2. is to drop it to ~arch.  This feels better given its quality
but kinda sucks for users.

When we remove py3.7, having pypy3.7 as the newest pypy will be the same
kind of problem whether it's stable or not.

-- 
Best regards,
Michał Górny





Re: [gentoo-dev] [RFC] Dropping PyPy(3) back to ~arch

2021-03-03 Thread Sam James

> On 2 Mar 2021, at 15:05, Michał Górny  wrote:
> 
> Hi,
> 
> I think I've made a mistake by stabilizing PyPy(3), and I would like to
> drop it back to ~arch (and stable-mask the relevant flags).
> 
> Roughly, there are 4 problems with it:
> 
> [snip]

> Honestly, I've tried to improve PyPy in the past but I don't really have
> time or motivation to do this continuously.  PyPy is an interesting
> project, and it has its isolated uses.  However, I don't think it's
> ready as a general-purpose Python interpreter for production
> environments.
> 
> I don't really want to remove it entirely or revert all the work we've
> put into testing packages with it -- but I think we should move it to
> ~arch at the very least.

It’s unfortunate but no objections. News item may be worthwhile
to explain to users why & how for posterity, but not necessary.



signature.asc
Description: Message signed with OpenPGP


Re: [gentoo-dev] [RFC] Dropping PyPy(3) back to ~arch

2021-03-02 Thread Ultrabug

On 02/03/2021 16:05, Michał Górny wrote:

Hi,



Hi



I don't really want to remove it entirely or revert all the work we've
put into testing packages with it -- but I think we should move it to
~arch at the very least.

WDYT?



I agree with you, I'd drop it back to ~arch too.

Thanks for your work on keeping pypy up.



[gentoo-dev] [RFC] Dropping PyPy(3) back to ~arch

2021-03-02 Thread Michał Górny
Hi,

I think I've made a mistake by stabilizing PyPy(3), and I would like to
drop it back to ~arch (and stable-mask the relevant flags).

Roughly, there are 4 problems with it:

1. The current stable (both Gentoo and upstream) is PyPy3.6 that is
equivalent to CPython 3.6.  Yes, the one we removed already.  People
using this version are already hitting issues because of upstreams
dropping py3.6 support or us failing to provide proper compatibility.

2. The next branch, PyPy3.7 is still in alpha state upstream.  It has
a few major bugs, e.g. the regex engine is sometimes slightly broken. 
The stdlib tests are mostly broken (compared to moderately broken
in 3.6).

3. To the best of my knowledge, there's no work hapenning to have
PyPy3.8, so it will keep being behind CPython.

4. PyPy upstream doesn't merge CPython stdlib changes timely, nor takes
care of vulnerabilities inherited from CPython.

5. PyPy3 doesn't give the same ABI stability guarantees as CPython,
and our choice not to slot between PyPy3 versions doesn't help (and it's
not worth changing considering all the other points).  Upgrading PyPy3
involves intermittent breakage, and people using it with Portage (which
is my fault) experience fatal breakage.


Honestly, I've tried to improve PyPy in the past but I don't really have
time or motivation to do this continuously.  PyPy is an interesting
project, and it has its isolated uses.  However, I don't think it's
ready as a general-purpose Python interpreter for production
environments.

I don't really want to remove it entirely or revert all the work we've
put into testing packages with it -- but I think we should move it to
~arch at the very least.

WDYT?

-- 
Best regards,
Michał Górny