"""Needed so package data is included."""

import itertools
from pathlib import Path

from skbuild import setup

MODULE_DIR = Path(__file__).parent / "src" / "piper"
PIPER_DATA_FILES = ["py.typed", "espeakbridge.pyi"]
ESPEAK_NG_DATA_DIR = MODULE_DIR / "espeak-ng-data"
ESPEAK_NG_DATA_FILES = [
    f.relative_to(MODULE_DIR) for f in ESPEAK_NG_DATA_DIR.rglob("*") if f.is_file()
]
TASHKEEL_DATA_DIR = MODULE_DIR / "tashkeel"
TASHKEEL_DATA_FILES = [
    (TASHKEEL_DATA_DIR / f_name).relative_to(MODULE_DIR)
    for f_name in (
        "model.onnx",
        "input_id_map.json",
        "target_id_map.json",
        "hint_id_map.json",
    )
]
HEBREW_DATA_DIR = MODULE_DIR / "hebrew"
HEBREW_DATA_FILES = [
    (HEBREW_DATA_DIR / f_name).relative_to(MODULE_DIR)
    for f_name in (
        "nakdimon.onnx",
        "LICENSE",
        "SOURCE",
    )
]
LITHUANIAN_DATA_DIR = MODULE_DIR / "lithuanian"
LITHUANIAN_DATA_FILES = [
    (LITHUANIAN_DATA_DIR / f_name).relative_to(MODULE_DIR)
    for f_name in (
        "lt_kirciai.tsv",
        "lt_raides.tsv",
        "lt_kreipiniai.tsv",
        "LICENSE",
        "SOURCE",
    )
]
# Web page and images for the HTTP server
HTTP_DATA_FILES = [
    f.relative_to(MODULE_DIR)
    for f in itertools.chain(
        (MODULE_DIR / "templates").rglob("*"),
        (MODULE_DIR / "img").rglob("*"),
    )
    if f.is_file()
]

setup(
    name="piper-tts",
    version="1.8.0",
    description="Fast and local neural text-to-speech engine",
    url="http://github.com/OHF-voice/piper1-gpl",
    license="GPL-3.0-or-later",
    # g2pW is Apache-2.0; see src/piper/g2pw_onnx.py.
    license_files=["COPYING", "licenses/LICENSE.g2pW-Apache-2.0"],
    author="The Home Assistant Authors",
    author_email="hello@home-assistant.io",
    keywords=["home", "assistant", "tts", "text-to-speech"],
    classifiers=[
        "Development Status :: 3 - Alpha",
        "Intended Audience :: Developers",
        "Topic :: Multimedia :: Sound/Audio :: Speech",
        "Programming Language :: Python :: 3.9",
        "Programming Language :: Python :: 3.10",
        "Programming Language :: Python :: 3.11",
        "Programming Language :: Python :: 3.12",
        "Programming Language :: Python :: 3.13",
    ],
    python_requires=">=3.9",
    install_requires=[
        "onnxruntime>=1,<2",
        "pathvalidate>=3,<4",
    ],
    extras_require={
        "train": [
            "torch>=2,<3",
            "lightning>=2,<3",
            "tensorboard>=2,<3",
            "tensorboardX>=2,<3",
            "jsonargparse[signatures]>=4.27.7",
            "onnx>=1,<2",
            "pysilero-vad>=2.1,<3",
            "cython>=3,<4",
            "librosa<1",
        ],
        "dev": [
            "black==24.8.0",
            "flake8==7.1.1",
            "isort==5.13.2",  # used by script/lint and script/format
            "mypy==1.14.0",
            "pylint==3.2.7",
            "pytest==8.3.4",
            "build==1.2.2",
            "scikit-build<1",
            "cmake>=3.18,<4",
            "ninja>=1,<2",
            "onnx>=1,<2",  # for alignments
        ],
        "http": [
            "flask>=3,<4",
        ],
        "alignment": [
            "onnx>=1,<2",
        ],
        "zh": [
            # g2pW supplies the pinyin/bopomofo lookup tables. Its model is run
            # by piper.g2pw_onnx rather than g2pw.api, which is what keeps torch
            # (~750 MB) out of this extra: g2pw.api imports it for a DataLoader.
            "g2pW>=0.1.1,<1",
            "transformers>=4,<6",
            "sentence-stream>=1.2.1,<2",
            "unicode-rbnf>=2.4.0,<3",
        ],
        "ja": [
            "pyopenjtalk-plus>=0.4,<1",
        ],
        "th": [
            # Thai word segmentation + G2P. espeak-ng's Thai voice is a
            # placeholder (see piper.phonemize_thai) and cannot be trained on.
            # Capped below 1.11: that release pins "scikit-learn~=1.2.0", which
            # has no wheels for current Pythons and whose compiled extensions
            # predate the numpy 2 ABI, so it builds from source and then dies
            # with "numpy.dtype size changed" on import. 1.10 leaves
            # scikit-learn unpinned. Lift the cap once tltk relaxes it.
            "tltk>=1.6.8,<1.11",
            # tltk imports pandas and requests but declares neither (as of
            # 1.10). requests is pulled in by tltk/__init__.py -> tltk.corpus,
            # which runs even though we only use tltk.nlp.
            "pandas>=2,<3",
            "requests>=2,<3",
            "unicode-rbnf>=2.4.0,<3",
        ],
    },
    packages=[
        "piper",
        "piper.tashkeel",
        "piper.hebrew",
        "piper.train",
        "piper.train.vits",
        "piper.train.vits.monotonic_align",
    ],
    package_dir={"": "src"},
    include_package_data=True,
    package_data={
        "piper": [
            str(p)
            for p in itertools.chain(
                PIPER_DATA_FILES,
                ESPEAK_NG_DATA_FILES,
                TASHKEEL_DATA_FILES,
                HEBREW_DATA_FILES,
                LITHUANIAN_DATA_FILES,
                HTTP_DATA_FILES,
            )
        ],
    },
    cmake_install_dir="src/piper",
    entry_points={
        "console_scripts": [
            "piper = piper.__main__:main",
        ]
    },
)