m230900
oMiser Articles
HashFlags Technique
Notes & Work Items
|
|
m230900
oMiser Articles |
|
Status |
Date |
Description |
|
| in progress | 2026-02-12 | Add m230900c on Object Comparison. Also find where that's talked about and link to m230900c. | |
| in progress | 2026-02-12 | Clean up m230900b on invariance, and mention the MOb there since it figure into how names are recovered for certain cases of individuals. | |
| in progress | 2026-01-25 |
If there is a bit on ob-cells to indicate possible value
of a binding, that bit cannot be part of the Hash tag construction,
because it is not inherent in the structure and origin of the ob-cell.
So it can only be a hint, and it doesn't even indicate when it might be
the value of more than one binding as well as no binding (any longer).
While this might be solvable by another level of indirection.
That's a scary prospect. [dh:2026-02-12: I thought it might work by some
finess with the hash of such an object. Each binding of the same
object would require an entry in the MOb, a kind of pseudo-individual
that ties an individual to a particular binding name. But the
bound ob-cell has no way to get back to any of those unless the ob-cell
hashflag is somehow useable to get "back up hill." We also need to cope with false positives when the ob-cell is no longer so bound. This might simply be a bridge too far. It's still interesting. The flag bit, if there is one, cannot be part of the HashFlag because it's not structural and invariant.] |
|
|
in progress |
2025-10-24 | Look into pseudo-random generators as ways to generate hashes from other hashes. This would have the HashFlag be treated in some manner as (part of) the current state of an RNG with significant cyclicity. One of the diehard RNGs that have easy next-random procedures might work. GB_FLIP is not suitable in this situation. Also check Knuth on this. This will influence how big we want the HashFlag to be and how we place the Flag bits. [dh:2025-11-04: I an digging around in the Stanford GraphBase and the dieharder offered good RNGs and see how the hashflags can be taken as seeds to generate new ones, especially for pairs. For enclosures and synthetics, different RNG methods could be used. For named individuals, we have the hashing on names to be of assistance.][dh:2026-02-12 because the reference count needs to be a 32-bit word, the HashFlags field only has 32 bit available. How many flag bits are needed determins how large the Hash portion can be. I am thinking of 8 bits for the flag, but that might not be essential. I know of 3 bits so far. I need to know how good a 24 bit hash can be before shrinking the 8-bit flag allowance though. | |
| done | 2025-09-14 |
Create this this page from the hybridForm examplar |
|
| done | 2024-02-18 |
|
|
 |
|
created 2025-09-14T19:48Z by orcmid |