Karim BELABAS on Wed, 4 Jun 2003 21:06:32 +0200 (MEST)

[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

Re: GEN serialisation

On Wed, 4 Jun 2003, Gonzalo Tornaria wrote:
> As a first idea (to check my prototype implementation), I thought of
> using GENtostr and readseq.
> Question: is it guaranteed that
>   readseq(GENtostr(x),0) == x
> provided GEN x is well formed?

No. Rounding problems for instance (internal is base 2^BITS_IN_LONG, output
is base 10).

Besides you should use flisexpr()  [ which is documented and slightly less
cryptic ]

> I think the "lontyp" array is for this, but I'm not familiar with it. My
> understanding is the following:
> say we have a GEN x, and let  lt=lontyp[typ(x)], len=lg(x).
> * if lt=0, the type is not recursive:
> * if lt>0, the type is recursive,
>   with lt codewords, and the rest is more GENs


>   recv_GEN(GEN *data, int count)
>   {
>     for(int i=0; i<count; i++)
>     {
>       long codeword; /* a "pseudo-GEN" */
>       GEN x=data[i];
>       long lt, len;
>       recv_long(&codeword, 1); /* get the type and length */
>       lt=lontyp[typ(&codeword)];

That's dangerous, it assumes x[0] contains the typ()   [ it's documented this
way, but it's the kind of dependency I'd rather do without. ]

>       len=lg(&codeword);

Same here.

>       x[0]=codeword;       /* is this necesary ??? */

No it's not.

> Am I missing something???

If some machines are compiled with a GMP kernel, and some with the native
one, or if endianness or sizeof(long) differ, you'll run into trouble.

There's also a big problem with variable numbers. You have to make sure that
all processes have defined the same number of variables [ otherwise
transmiting and using a GEN involving an undefined variable number will very
probably trigger a SEGV ]

[ note that my previously posted solution only deals with the first problem
:-( ]

> Is it guaranteed that the first codeword has only length and type
> information??? I.e., after doing
>   x=cgetg(lg(&codeword), typ(&codeword)),
> is it guaranteed that
>   x[0]==codeword

No. You lose the "clone bit". In any case, it's very unsafe to act on words
without using the documented API ( there's been a few proposals to change the
internal representation already ).


Karim Belabas                     Tel: (+33) (0)1 69 15 57 48
Dép. de Mathématiques, Bât. 425   Fax: (+33) (0)1 69 15 60 19
Université Paris-Sud              http://www.math.u-psud.fr/~belabas/
F-91405 Orsay (France)            http://www.parigp-home.de/  [PARI/GP]