Re: bug in simplify_shallow

[Bill Allombert <Bill.Allombert@math.u-bordeaux.fr>]

On Sat, Aug 26, 2023 at 03:47:09PM +0200, Karim Belabas wrote:
> You'll see it with valgrind.
> 
> >From the GP code, I can guess what's happening: the lexical variable L
> will be destroyed (gunclone) when we leave the user function scope. But
> that user function still returns the return value of listput() which is
> the element stored in the list. The latter is a 'clone' (as all list
> entries), so supposedly safe. The lexical variable destructor breaks
> that asumption.
> 
> I don't think the reference count machinery currently in place can
> deal with this. Adding a gcopy somewhere should be enough (but slow down
> all 'listput' operations...). Don't have time to delve in the C code
> right now to confirm this; I'll have a look later if Bill doesn't beat
> me to it :-)

Well, we could probably change listput  (and listinsert) to use the prototype
code m.  But to be honest I did not remember that listput returned x and it is
not explicitly documented, and the implementation is awkward.
(the simple, correct implementation would return gcopy of the original x, not a clone 
of it).

Cheers,
Bill.