[Debian-science-sagemath] fpylll: dependency or dependencee of Sage ?: fplll 5.0.3

[Julien Puydt]

Hi,

On 20/10/2016 16:02, Jerome BENOIT wrote:

> On 19/10/16 21:07, Julien Puydt wrote:
>>
>> I had a look and I'm wondering if the ABI wasn't changed too:
>> modifying the interface of a class will change the size of the
>> structures and the storage order.
>
> Is the new ABI compatible with the former one ?

I would say it depends of the use:
(1) if you always do
A *a = api_call_create(args);
a->method1(args);
a->method2(args);
api_call_destroy(a);

or some such, then things should be ok, since you're not managing the 
size of the mallocs yourself, and not poking in the data members yourself.

(2) but if you do something like:
A a;
or
a->member;
then chances are you won't take enough room on the stack for the added 
variables, or read/write at the wrong place in memory for the member 
variables.

But the short version of this discussion is of course : no, it's not 
compatible.

> Is it envisaged to introduce any `version script' to fplll ?

Well, I see the FPLLL_LT_* variables were updated 
(https://github.com/fplll/fplll/blame/master/configure.ac) (notice that 
these were put in place in august because of me: 
https://github.com/fplll/fplll/issues/185 ), but point (1) of the 
instructions was applied, while I'm sure point (2) should have : 
probably the soname version should have changed.

I reported to upstream as: https://github.com/fplll/fplll/issues/217

Cheers,

Snark on #debian-science