Re: Arbitrary-precision numbers in patterns
- To: mathgroup at smc.vnet.net
- Subject: [mg45415] Re: [mg45388] Arbitrary-precision numbers in patterns
- From: Andrzej Kozlowski <akoz at mimuw.edu.pl>
- Date: Wed, 7 Jan 2004 01:09:05 -0500 (EST)
- References: <200401050851.DAA20153@smc.vnet.net>
- Sender: owner-wri-mathgroup at wolfram.com
It seems to me that the likely reason for this behaviour is that
Mathematica switches to a different "hash function", once the list of
hash values starts getting a bit longish. At least to my limited
understanding of such things (based only on some knowledge of Knuth,
not on practical experience) that would seem to be a natural thing to
do. That would of course involve "reordering". However, I doubt that
you can expect such things to be made public as long as the source code
of Mathematica is not.
Andrzej Kozlowski
On 5 Jan 2004, at 17:51, Maxim wrote:
> Compare
>
> In[1]:=
> Do[ f[k] = k, {k, 1., 17.} ]
> f[1.`20]
> Clear[f]
>
> Out[2]=
> 1.
>
> and
>
> In[1]:=
> Do[ f[k] = k, {k, 1., 18.} ]
> f[1.`20]
> Clear[f]
>
> Out[2]=
> f[1.0000000000000000000]
>
> -- and the user's best bet to figure out how it'll work is probably to
> flip a coin.
>
> The reason is probably just that the hashing mechanism breaks down,
> because the result returned by Mathematica changes after it re-sorts
> some internal table of DownValues for f (the 'boundary value' 17 is for
> version 5.0 on my machine; if 17. and 18. don't work, try 2. and 100.).
> But in general, my opinion is that it is only to be expected -- when we
> don't even know for sure how the definitions for f can be reordered.
>
> Maxim Rytin
> m.r at prontomail.com
>
>
>
- References:
- Arbitrary-precision numbers in patterns
- From: Maxim <dontsendhere@.>
- Arbitrary-precision numbers in patterns