| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
| |
(#15108)
avoid future implementation mischief. (Maybe not. Sometimes, general
distrust of theory leads people to distrust simple reasoning over times
from CPUs trying as hard as possible to mask DRAM latency via pre-fetch.)
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
(#15104)
request. This can be conceived as an alternate, more capable resolution of
https://github.com/nim-lang/Nim/issues/12200
than
https://github.com/nim-lang/Nim/pull/12208
The code re-org idea here is to upgrade tablimpl.nim:`delImpl`/`delImplIdx`
to abstract client code conventions for cell emptiness & cell hashing via
three new template arguments - `makeEmpty`, `cellEmpty`, `cellHash` which
all take a single integer argument and clear a cell, test if clear or
produce the hash of the key stored at that index in `.data[]`.
Then we update the 3 call sites (`Table`, `CountTable`, `SharedTable`) of
`delImpl`/`delImplIdx` by defining define those arguments just before the
first invocation as non-exported templates.
Because `CountTable` does not save hash() outputs as `.hcode`, it needs a
new tableimpl.nim:`delImplNoHCode` which simply in-lines the hash search
when no `.hcode` field is available for "prefix compare" acceleration.
It is conceivable this new template could be used by future variants, such
as one optimized for integer keys where `hash()` and `==` are fast and
`.hcode` is both wasted space & time (though a small change to interfaces
there for a sentinel key meaning "empty" is needed for maximum efficiency).
We also eliminate the old O(n) `proc remove(CountTable...)` in favor of
simply invoking the new `delImpl*` templates and take care to correctly
handle the case where `val` is either zero for non-existent keys in `inc`
or evolves to zero over time in `[]=` or `inc`.
The only user-visible changes from the +-42 delta here are speed, iteration
order post deletes, and relaxing the `Positive` constraint on `val` in
`proc inc` again, as indicated in the `changelog.md` entry.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* remove a condition that table size must be passed as power of 2
* remove power-of-2 condition from sets and sharedtables
* remove power-of-2 condition from deques
* use 'correctSize' for both branches
* prettify changelog.md and fix typos
* add a changelog entry
* fix double-call of 'right-size'
* fix the same thing in sets.nim
* introduce a new internal proc `slotsNeeded`
Deprecate the public proc `rightSize`, which is not needed anymore.
Now it is an identity function, allowing the old code to work
correctly and without extra allocations.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
(#13816)
* Unwind just the "pseudorandom probing" (whole hash-code-keyed variable
stride double hashing) part of recent sets & tables changes (which has
still been causing bugs over a month later (e.g., two days ago
https://github.com/nim-lang/Nim/issues/13794) as well as still having
several "figure this out" implementation question comments in them (see
just diffs of this PR).
This topic has been discussed in many places:
https://github.com/nim-lang/Nim/issues/13393
https://github.com/nim-lang/Nim/pull/13418
https://github.com/nim-lang/Nim/pull/13440
https://github.com/nim-lang/Nim/issues/13794
Alternative/non-mandatory stronger integer hashes (or vice-versa opt-in
identity hashes) are a better solution that is more general (no illusion
of one hard-coded sequence solving all problems) while retaining the
virtues of linear probing such as cache obliviousness and age-less tables
under delete-heavy workloads (still untested after a month of this change).
The only real solution for truly adversarial keys is a hash keyed off of
data unobservable to attackers. That all fits better with a few families
of user-pluggable/define-switchable hashes which can be provided in a
separate PR more about `hashes.nim`.
This PR carefully preserves the better (but still hard coded!) probing
of the `intsets` and other recent fixes like `move` annotations, hash
order invariant tests, `intsets.missingOrExcl` fixing, and the move of
`rightSize` into `hashcommon.nim`.
* Fix `data.len` -> `dataLen` problem.
|
| |
|
|
|
|
|
|
|
|
|
| |
tests (#13498) [backport]
* fix #13496 handle tombstones
* add test
* more tests
* fix #13504; add SharedTable tests
* fix #https://github.com/nim-lang/Nim/issues/13505 intsets.missingOrExcl silently gave wrong results sometimes
* add test for tintsets
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
| |
(#11385)
* make fullpaths the default in error messages and stack traces for more convenient development
* split up -d:release into -d:release and -d:danger flags
* workaround a Nim config parser bug
* fixes an old nim config parser bug
* make megatest green again
* make nimpretty tests work again
* make nimsuggest green
|
| |
|
|
|
|
|
|
|
|
|
|
| |
* tables: initialized by default
* sets: initialized by default
* DRY: extract shared functionality
* add a changelog entry
* fix errors
* don't test include files
* make it work for sharedtables
* fix discovered bugs
* add exhaustive tests
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |\
| |
| | |
Improvements to tables.clear()
|
| | |
| |
| |
| |
| |
| | |
The record tuples used in CountData.data don't contain an 'hcode' member,
unlike Table and OrderedTable, causing the existing clearImpl() implementation
to break when attempting to assign to t.data[i].hcode.
|
| | | |
|
| |/ |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
| |
Conflicts:
doc/tut1.txt
lib/core/locks.nim
lib/pure/collections/tables.nim
lib/pure/selectors.nim
|
| |
|