import abc
from collections import OrderedDict

import numpy as np
import pytest

from astropy.io import fits
from astropy.utils import metadata
from astropy.utils.metadata import (
    MergeConflictError,
    MetaData,
    common_dtype,
    enable_merge_strategies,
    merge,
)


class OrderedDictSubclass(OrderedDict):
    pass


class MetaBaseTest:
    __metaclass__ = abc.ABCMeta

    def test_none(self):
        d = self.test_class(*self.args)
        assert isinstance(d.meta, OrderedDict)
        assert len(d.meta) == 0

    @pytest.mark.parametrize(
        "meta",
        ([dict([("a", 1)]), OrderedDict([("a", 1)]), OrderedDictSubclass([("a", 1)])]),
    )
    def test_mapping_init(self, meta):
        d = self.test_class(*self.args, meta=meta)
        assert type(d.meta) == type(meta)
        assert d.meta["a"] == 1

    @pytest.mark.parametrize("meta", (["ceci n'est pas un meta", 1.2, [1, 2, 3]]))
    def test_non_mapping_init(self, meta):
        with pytest.raises(TypeError):
            self.test_class(*self.args, meta=meta)

    @pytest.mark.parametrize(
        "meta",
        ([dict([("a", 1)]), OrderedDict([("a", 1)]), OrderedDictSubclass([("a", 1)])]),
    )
    def test_mapping_set(self, meta):
        d = self.test_class(*self.args, meta=meta)
        assert type(d.meta) == type(meta)
        assert d.meta["a"] == 1

    @pytest.mark.parametrize("meta", (["ceci n'est pas un meta", 1.2, [1, 2, 3]]))
    def test_non_mapping_set(self, meta):
        with pytest.raises(TypeError):
            d = self.test_class(*self.args, meta=meta)

    def test_meta_fits_header(self):
        header = fits.header.Header()
        header.set("observer", "Edwin Hubble")
        header.set("exptime", "3600")

        d = self.test_class(*self.args, meta=header)

        assert d.meta["OBSERVER"] == "Edwin Hubble"


class ExampleData:
    meta = MetaData()

    def __init__(self, meta=None):
        self.meta = meta


class TestMetaExampleData(MetaBaseTest):
    test_class = ExampleData
    args = ()


def test_metadata_merging_conflict_exception():
    """Regression test for issue #3294.

    Ensure that an exception is raised when a metadata conflict exists
    and ``metadata_conflicts='error'`` has been set.
    """
    data1 = ExampleData()
    data2 = ExampleData()
    data1.meta["somekey"] = {"x": 1, "y": 1}
    data2.meta["somekey"] = {"x": 1, "y": 999}
    with pytest.raises(MergeConflictError):
        merge(data1.meta, data2.meta, metadata_conflicts="error")


def test_metadata_merging():
    # Recursive merge
    meta1 = {
        "k1": {
            "k1": [1, 2],
            "k2": 2,
        },
        "k2": 2,
        "k4": (1, 2),
    }
    meta2 = {
        "k1": {"k1": [3]},
        "k3": 3,
        "k4": (3,),
    }
    out = merge(meta1, meta2, metadata_conflicts="error")
    assert out == {
        "k1": {
            "k2": 2,
            "k1": [1, 2, 3],
        },
        "k2": 2,
        "k3": 3,
        "k4": (1, 2, 3),
    }

    # Merge two ndarrays
    meta1 = {"k1": np.array([1, 2])}
    meta2 = {"k1": np.array([3])}
    out = merge(meta1, meta2, metadata_conflicts="error")
    assert np.all(out["k1"] == np.array([1, 2, 3]))

    # Merge list and np.ndarray
    meta1 = {"k1": [1, 2]}
    meta2 = {"k1": np.array([3])}
    assert np.all(out["k1"] == np.array([1, 2, 3]))

    # Can't merge two scalar types
    meta1 = {"k1": 1}
    meta2 = {"k1": 2}
    with pytest.raises(MergeConflictError):
        merge(meta1, meta2, metadata_conflicts="error")

    # Conflicting shape
    meta1 = {"k1": np.array([1, 2])}
    meta2 = {"k1": np.array([[3]])}
    with pytest.raises(MergeConflictError):
        merge(meta1, meta2, metadata_conflicts="error")

    # Conflicting array type
    meta1 = {"k1": np.array([1, 2])}
    meta2 = {"k1": np.array(["3"])}
    with pytest.raises(MergeConflictError):
        merge(meta1, meta2, metadata_conflicts="error")

    # Conflicting array type with 'silent' merging
    meta1 = {"k1": np.array([1, 2])}
    meta2 = {"k1": np.array(["3"])}
    out = merge(meta1, meta2, metadata_conflicts="silent")
    assert np.all(out["k1"] == np.array(["3"]))


def test_metadata_merging_new_strategy():
    original_merge_strategies = list(metadata.MERGE_STRATEGIES)

    class MergeNumbersAsList(metadata.MergeStrategy):
        """
        Scalar float or int values are joined in a list.
        """

        types = ((int, float), (int, float))

        @classmethod
        def merge(cls, left, right):
            return [left, right]

    class MergeConcatStrings(metadata.MergePlus):
        """
        Scalar string values are concatenated
        """

        types = (str, str)
        enabled = False

    # Normally can't merge two scalar types
    meta1 = {"k1": 1, "k2": "a"}
    meta2 = {"k1": 2, "k2": "b"}

    # Enable new merge strategy
    with enable_merge_strategies(MergeNumbersAsList, MergeConcatStrings):
        assert MergeNumbersAsList.enabled
        assert MergeConcatStrings.enabled
        out = merge(meta1, meta2, metadata_conflicts="error")
    assert out["k1"] == [1, 2]
    assert out["k2"] == "ab"
    assert not MergeNumbersAsList.enabled
    assert not MergeConcatStrings.enabled

    # Confirm the default enabled=False behavior
    with pytest.raises(MergeConflictError):
        merge(meta1, meta2, metadata_conflicts="error")

    # Enable all MergeStrategy subclasses
    with enable_merge_strategies(metadata.MergeStrategy):
        assert MergeNumbersAsList.enabled
        assert MergeConcatStrings.enabled
        out = merge(meta1, meta2, metadata_conflicts="error")
    assert out["k1"] == [1, 2]
    assert out["k2"] == "ab"
    assert not MergeNumbersAsList.enabled
    assert not MergeConcatStrings.enabled

    metadata.MERGE_STRATEGIES = original_merge_strategies


def test_common_dtype_string():
    u3 = np.array(["123"])
    u4 = np.array(["1234"])
    b3 = np.array([b"123"])
    b5 = np.array([b"12345"])
    assert common_dtype([u3, u4]).endswith("U4")
    assert common_dtype([b5, u4]).endswith("U5")
    assert common_dtype([b3, b5]).endswith("S5")


def test_common_dtype_basic():
    i8 = np.array(1, dtype=np.int64)
    f8 = np.array(1, dtype=np.float64)
    u3 = np.array("123")

    with pytest.raises(MergeConflictError):
        common_dtype([i8, u3])

    assert common_dtype([i8, i8]).endswith("i8")
    assert common_dtype([i8, f8]).endswith("f8")