first big restruct

bench/ctable/ctable_v_panda.py

@@ -0,0 +1,100 @@
+import time
+import numpy as np
+import blosc2
+import pandas as pd
+from pydantic import BaseModel, Field
+from typing import Annotated
+import psutil
+
+# --- 1. Tu RowModel COMPLEJO ---
+class NumpyDtype:
+    def __init__(self, dtype):
+        self.dtype = dtype
+
+class RowModel(BaseModel):
+    id: Annotated[int, NumpyDtype(np.int64)] = Field(ge=0)
+    c_val: Annotated[complex, NumpyDtype(np.complex128)] = Field(default=0j)
+    score: Annotated[float, NumpyDtype(np.float64)] = Field(ge=0, le=100)
+    active: Annotated[bool, NumpyDtype(np.bool_)] = True
+
+# --- 2. Parámetros ---
+N = 10_000_000  # 1M filas
+print(f"=== BENCHMARK: 1M Filas COMPLEJAS (Listas de Listas) ===\n")
+
+# ==========================================
+# 0. GENERAR DATOS (Lista de listas COMPLEJA)
+# ==========================================
+print("--- Generando 1M filas complejas ---")
+t0 = time.time()
+data_list = []
+for i in range(N):
+    data_list.append([
+        i,                                    # id: int64
+        complex(i*0.1, i*0.01),              # c_val: complex128
+        10.0 + np.sin(i*0.001)*50,           # score: float64
+        (i % 3 == 0)                         # active: bool
+    ])
+t_gen = time.time() - t0
+print(f"Tiempo generación: {t_gen:.4f} s")
+print(f"Lista ocupa: {len(data_list):,} filas\n")
+
+# ==========================================
+# 1. PANDAS: Lista compleja -> DataFrame
+# ==========================================
+print("--- 1. PANDAS (Creación) ---")
+gc_pandas = psutil.Process().memory_info().rss / (1024**2)
+t0 = time.time()
+
+df = pd.DataFrame(data_list, columns=['id', 'c_val', 'score', 'active'])
+
+t_pandas_create = time.time() - t0
+gc_pandas_after = psutil.Process().memory_info().rss / (1024**2)
+mem_pandas = gc_pandas_after - gc_pandas
+print(f"Tiempo creación:  {t_pandas_create:.4f} s")
+print(f"Memoria usada:    {mem_pandas:.2f} MB")
+
+# Pandas head(1000)
+t0 = time.time()
+df_head = df.head(N)
+t_pandas_head = time.time() - t0
+print(f"Tiempo head(1000): {t_pandas_head:.6f} s\n")
+
+# ==========================================
+# 2. BLOSC2 Oficial: extend() con conversión
+# ==========================================
+print("--- 2. BLOSC2 Oficial (extend + conversión Pydantic) ---")
+gc_blosc = psutil.Process().memory_info().rss / (1024**2)
+t0 = time.time()
+
+# ❌ Blosc2 oficial REQUIERE conversión a modelos
+ctable = blosc2.CTable(RowModel, expected_size=N)
+ctable.extend(data_list)
+
+t_blosc_create = time.time() - t0
+gc_blosc_after = psutil.Process().memory_info().rss / (1024**2)
+mem_blosc = gc_blosc_after - gc_blosc
+mem_compressed = sum(col.schunk.nbytes for col in ctable._cols.values()) / (1024**2)
+print(f"Tiempo creación:  {t_blosc_create:.4f} s")
+total_comprimido = sum(col.cbytes for col in ctable._cols.values()) + ctable._valid_rows.cbytes
+total_sin_comprimir = sum(col.nbytes for col in ctable._cols.values()) + ctable._valid_rows.nbytes
+
+print(f"Comprimido: {total_comprimido / 1024 ** 2:.2f} MB")
+print(f"Sin comprimir: {total_sin_comprimir / 1024 ** 2:.2f} MB")
+print(f"Ratio: {total_sin_comprimir/total_comprimido:.2}x")
+
+t0 = time.time()
+ctable_head = ctable.head(N)
+t_blosc_head = time.time() - t0
+print(f"Tiempo head(1000): {t_blosc_head:.6f} s\n")
+
+
+
+# ==========================================
+# 🏆 RESUMEN COMPLETO
+# ==========================================
+print("═" * 80)
+print("🥇 BENCHMARK 1M FILAS COMPLEJAS (int64+complex128+float64+bool)")
+print("═" * 80)
+print(f"{'MÉTRICA':<22} {'PANDAS':>12} {'BLOsc2*':>10} {'TU CTable':>12}")
+print(f"{'':<22} {'':>12} {'*+Pydantic':>10} {'¡Directo!':>12}")
+print("-" * 80)

bench/ctable/print.py

@@ -0,0 +1,81 @@
+import time
+import numpy as np
+import pandas as pd
+import blosc2
+from pydantic import BaseModel, Field
+
+# --- 1. Definir el Modelo ---
+class RowModel(BaseModel):
+    id: int = Field(ge=0)
+    name: bytes = Field(default=b"unknown", max_length=10)
+    score: float
+
+# --- 2. Parámetros ---
+N = 100_000
+row_data = {"id": 1, "name": b"benchmark", "score": 3.14}
+
+print(f"=== BENCHMARK: Ingestión Iterativa ({N} filas) ===\n")
+
+# ==========================================
+# TEST PANDAS (Baseline)
+# ==========================================
+print("--- 1. PANDAS (Lista -> DataFrame) ---")
+t0 = time.time()
+
+buffer_list = []
+for _ in range(N):
+    buffer_list.append(row_data)
+
+df = pd.DataFrame(buffer_list)
+t_pandas = time.time() - t0
+
+print(f"Tiempo Total: {t_pandas:.4f} s")
+mem_pandas = df.memory_usage(deep=True).sum() / (1024**2)
+print(f"Memoria RAM:  {mem_pandas:.2f} MB")
+
+print("\n--- PANDAS: Primeras 1000 líneas ---")
+t0_print = time.time()
+print(df.head(1000).to_string())
+t_print_pandas = time.time() - t0_print
+print(f"\nTiempo de impresión: {t_print_pandas:.4f} s")
+
+
+# ==========================================
+# TEST BLOSC2 (Estrategia: extend() con lista)
+# ==========================================
+print("\n" + "="*60)
+print("--- 2. BLOSC2 (extend con lista de dicts) ---")
+t0 = time.time()
+
+# Acumular en lista de diccionarios
+buffer_list_2 = []
+for _ in range(N):
+    buffer_list_2.append(row_data)
+
+# Crear CTable vacía e insertar todo de golpe
+ctable = blosc2.CTable(RowModel)
+ctable.extend(buffer_list_2)
+
+t_blosc_extend = time.time() - t0
+print(f"Tiempo Total: {t_blosc_extend:.4f} s")
+
+mem_blosc_extend = sum(col.schunk.nbytes for col in ctable._cols.values()) / (1024**2)
+print(f"Memoria (Compr): {mem_blosc_extend:.2f} MB")
+
+print("\n--- BLOSC2: Primeras 1000 líneas ---")
+t0_print = time.time()
+ctable_head = ctable.head(1000)
+print(ctable_head)
+t_print_blosc = time.time() - t0_print
+print(f"\nTiempo de impresión: {t_print_blosc:.4f} s")
+
+# ==========================================
+# CONCLUSIONES
+# ==========================================
+print("\n" + "="*60)
+print("--- RESUMEN ---")
+print(f"Pandas (lista->df):       {t_pandas:.4f} s")
+print(f"Blosc2 (extend):          {t_blosc_extend:.4f} s ({t_pandas/t_blosc_extend:.2f}x {'más rápido' if t_blosc_extend < t_pandas else 'más lento'})")
+print(f"\nImpresión Pandas:         {t_print_pandas:.4f} s")
+print(f"Impresión Blosc2:         {t_print_blosc:.4f} s")
+print(f"\nCompresión Blosc2 vs Pandas: {mem_blosc_extend / mem_pandas * 100:.2f}% del tamaño")

pyproject.toml

@@ -39,15 +39,12 @@ dependencies = [
     "numexpr>=2.14.1; platform_machine != 'wasm32'",
     "requests",
 ]
-version = "4.1.3.dev0"
+version = "4.0.0-b2.dev0"
 [project.entry-points."array_api"]
 blosc2 = "blosc2"
 
-[project.urls]
-homepage = "https://github.com/Blosc/python-blosc2"
-documentation = "https://www.blosc.org/python-blosc2/python-blosc2.html"
 
-[dependency-groups]
+[project.optional-dependencies]
 dev = [
     "dask",
     "h5py",
@@ -82,6 +79,10 @@ doc = [
     "numba",
 ]
 
+[project.urls]
+homepage = "https://github.com/Blosc/python-blosc2"
+documentation = "https://www.blosc.org/python-blosc2/python-blosc2.html"
+
 [tool.cibuildwheel]
 build-verbosity = 1
 # Skip unsupported python versions as well as 32-bit platforms, which are not supported anymore.
@@ -120,17 +121,14 @@ ignore = [
     "PT011",
     "RET505",
     "RET508",
-    "RUF001",
-    "RUF002",
-    "RUF003",
     "RUF005",
     "RUF015",
     "RUF059",
     "SIM108",
 ]
 
 [tool.ruff.lint.extend-per-file-ignores]
-"tests/**" = ["F841", "C901"]
+"tests/**" = ["F841"]
 
 [tool.ruff.lint.mccabe]
 # Raise complexity from the default 10 to 13

src/blosc2/__init__.py

@@ -589,6 +589,7 @@ def _raise(exc):
 """
 
 # Delayed imports for avoiding overwriting of python builtins
+from .ctable import CTable
 from .ndarray import (
     abs,
     acos,
@@ -796,6 +797,7 @@ def _raise(exc):
     "count_nonzero",
     "cparams_dflts",
     "cpu_info",
+    "ctable",
     "cumulative_prod",
     "cumulative_sum",
     "decompress",