Metadata-Version: 2.4
Name: lookprime
Version: 0.2.3
Summary: Ultra-fast prime lookup using memory-mapped odd-only sieve
Author: Your Name
License: MIT
Requires-Python: >=3.9
Description-Content-Type: text/markdown
License-File: LICENSE
Dynamic: license-file

# lookprime

**lookprime** is a high-performance prime number library for Python
built around a **memory-mapped, odd-only, bitset sieve**.

It is designed for: - extremely fast primality tests - fast prime
counting and enumeration - repeated queries across processes - large
limits with minimal RAM overhead - CLI benchmarking and inspection

The core idea:\
👉 build a prime sieve **once**, store it on disk, then reuse it via
**mmap** with near-zero startup cost.

------------------------------------------------------------------------

## Features

### Core performance

-   Odd-only sieve (½ the memory of full sieve)
-   Bitset representation (1 bit per odd number)
-   Memory-mapped cache (`mmap`) for zero-copy sharing
-   `int.from_bytes(...).bit_count()` for fast counting
-   Auto-growing cache when needed
-   Safe for multi-process use (read-only)

### Python API

-   `isprime(n)`
-   `isprime_many(iterable)`
-   `primerange(a, b)`
-   `randprime(a, b)`
-   `primepi(n)`
-   `prime(n)` (nth prime, 1-indexed)
-   `prevprime(n)`
-   `nextprime(n)`
-   `primes(n)` (first `n` primes)
-   `primes_up_to(n)`
-   `factorint(n)`
-   `cache_info()`
-   `clear_cache()`

### CLI

-   `lookprime info`
-   `lookprime build`
-   `lookprime benchmark`
-   `lookprime isprime <n>`
-   `lookprime factor <n>`
-   `lookprime clear-cache`

------------------------------------------------------------------------

## Installation

``` bash
pip install lookprime
```

Python ≥ 3.9 required.

------------------------------------------------------------------------

## Quick Start (Python)

``` python
import lookprime

lookprime.isprime(97)                 # True
lookprime.primepi(1_000_000)          # 78498
lookprime.primes_up_to(50)            # [2, 3, 5, 7, 11, ...]
lookprime.factorint(123456)           # {2: 6, 3: 1, 643: 1}
```

Bulk checks:

``` python
lookprime.isprime_many([2, 3, 4, 5, 10**9 + 7])
```

------------------------------------------------------------------------

## How the Cache Works

-   Sieve files are stored in a platform-appropriate cache directory:

    -   Linux: `~/.cache/lookprime`
    -   macOS: `~/Library/Caches/lookprime`
    -   Windows: `%LOCALAPPDATA%\lookprime\Cache`

-   Each sieve is stored as:

        lookprime_mask_<limit>.lpm

-   Files are memory-mapped (`mmap`) and shared across processes

-   Rebuilding happens **only if needed**

To inspect cache state:

``` python
lookprime.cache_info()
```

To delete all cached sieves:

``` python
lookprime.clear_cache()
```

or from CLI:

``` bash
lookprime clear-cache
```

------------------------------------------------------------------------

## CLI Usage

### Show system and cache info

``` bash
lookprime info
```

------------------------------------------------------------------------

### Build a sieve explicitly

``` bash
lookprime build --limit 100000000
```

------------------------------------------------------------------------

### Benchmark lookup speed

``` bash
lookprime benchmark
```

Options:

``` bash
lookprime benchmark \
  --limit 100000000 \
  --duration 1.0 \
  --iterations 25 \
  --chunk 65536
```

Environment override:

``` bash
export LOOKPRIME_CHUNK=65536
lookprime benchmark
```

------------------------------------------------------------------------

### Primality from CLI

``` bash
lookprime isprime 104729
```

------------------------------------------------------------------------

### Factor integers from CLI

``` bash
lookprime factor 123456
```

------------------------------------------------------------------------

## Performance Notes

-   `isprime(n)` is **O(1)** after cache load
-   `primepi(n)` uses bitset population count
-   `prime(n)` and `primes(n)` use fast bit scanning
-   Cache memory is shared between processes
-   No Python objects are created per lookup

------------------------------------------------------------------------

## Threading & Multiprocessing

-   The sieve is **read-only**
-   Safe to open in multiple processes
-   Each process memory-maps its own view
-   No locks required

------------------------------------------------------------------------

## License

MIT License.
