[Tarantool-patches] [PATCH] build: disable LUAJIT_ENABLE_PAIRSMM

[Igor Munkin]

Kostja,

On 20.02.20, Konstantin Osipov wrote:
> * Igor Munkin <imun at tarantool.org> [20/02/20 18:27]:
> 
> This is pathetic. Don't you realize that once a version is
> released and declared stable, it's a legacy you have to live with?

As we announced the patch was applied by mistake. If you found the place
we missed, please show it. If you want to complain and blame someone,
feel free to choose me, since I initially proposed this amiss and
breaking solution in #4560[1]. Unfortunately this doesn't make the
situation easier for both of us.

The fact we applied the patch makes me sorry (but I'll try to omit
emotions from the ml). Moreover, such breakage can't be introduced
within a minor release. Originally it was needed to implement the PoC
for #4521[2] and wasn't considered as a feature per se. It formally
breaks the semantics of the language currently being used for out
platform. As we can see it really breaks the existing code.

> 
> Now you have *3* broken behaviours to support, instead of just
> two. Some people will still use a version which they believe is
> stable, and encounter strange failures.

I'm bad in math here, since I see only two behaviours: the old one and
the broken one. Three would be if we settle the issue and save the
introduced behaviour the way you proposed below. Then there will be:
* working Tarantool w/o __pairs
* broken Tarantool w/ __pairs
* working Tarantool w/ __pairs

But again: I might miscount it, so it would be great if you clarify your
math.

Yes, there are stable releases poisoned with the introduced breakage. If
you think I'm careless about it, you're wrong. I guess we can ask users
to upgrade like many times you used to say "obnovites" with apoligies. I
wish we could avoid it this time. But I see more problems with the
platform infrastructure and environment as a result than "blacklisting"
several Tarantool versions.

> 
> Eventually you'll have to enable 5.2 compatibility anyway (are you
> going to stay with 5.1 forever??), and then break it again.

I can't answer it now. You are the visionary, I'm even not a mainainer.
We discussed it in tg chat some time ago but I haven't been answered the
question about your vision of this question. I welcome you to proceed
the discussion and you can choose the most convenient way for it.

> 
> luafun is a builtin and table.deepcopy are builtins. What prevented you from fixing them 
> instead?
> That would fix 99% of cases. 

I see no guarantees here, and guess you can provide nothing.

> 
> 
> -- 
> Konstantin Osipov, Moscow, Russia

[1]: https://github.com/tarantool/tarantool/issues/4560
[2]: https://github.com/tarantool/tarantool/issues/4521

-- 
Best regards,
IM