summaryrefslogtreecommitdiffstats
path: root/tests/tabular_output
diff options
context:
space:
mode:
Diffstat (limited to 'tests/tabular_output')
-rw-r--r--tests/tabular_output/__init__.py0
-rw-r--r--tests/tabular_output/test_delimited_output_adapter.py55
-rw-r--r--tests/tabular_output/test_output_formatter.py271
-rw-r--r--tests/tabular_output/test_preprocessors.py350
-rw-r--r--tests/tabular_output/test_tabulate_adapter.py113
-rw-r--r--tests/tabular_output/test_tsv_output_adapter.py36
-rw-r--r--tests/tabular_output/test_vertical_table_adapter.py49
7 files changed, 874 insertions, 0 deletions
diff --git a/tests/tabular_output/__init__.py b/tests/tabular_output/__init__.py
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/tests/tabular_output/__init__.py
diff --git a/tests/tabular_output/test_delimited_output_adapter.py b/tests/tabular_output/test_delimited_output_adapter.py
new file mode 100644
index 0000000..86a622e
--- /dev/null
+++ b/tests/tabular_output/test_delimited_output_adapter.py
@@ -0,0 +1,55 @@
+# -*- coding: utf-8 -*-
+"""Test the delimited output adapter."""
+
+from __future__ import unicode_literals
+from textwrap import dedent
+
+import pytest
+
+from cli_helpers.tabular_output import delimited_output_adapter
+
+
+def test_csv_wrapper():
+ """Test the delimited output adapter."""
+ # Test comma-delimited output.
+ data = [["abc", "1"], ["d", "456"]]
+ headers = ["letters", "number"]
+ output = delimited_output_adapter.adapter(iter(data), headers, dialect="unix")
+ assert "\n".join(output) == dedent(
+ '''\
+ "letters","number"\n\
+ "abc","1"\n\
+ "d","456"'''
+ )
+
+ # Test tab-delimited output.
+ data = [["abc", "1"], ["d", "456"]]
+ headers = ["letters", "number"]
+ output = delimited_output_adapter.adapter(
+ iter(data), headers, table_format="csv-tab", dialect="unix"
+ )
+ assert "\n".join(output) == dedent(
+ '''\
+ "letters"\t"number"\n\
+ "abc"\t"1"\n\
+ "d"\t"456"'''
+ )
+
+ with pytest.raises(ValueError):
+ output = delimited_output_adapter.adapter(
+ iter(data), headers, table_format="foobar"
+ )
+ list(output)
+
+
+def test_unicode_with_csv():
+ """Test that the csv wrapper can handle non-ascii characters."""
+ data = [["观音", "1"], ["Ποσειδῶν", "456"]]
+ headers = ["letters", "number"]
+ output = delimited_output_adapter.adapter(data, headers)
+ assert "\n".join(output) == dedent(
+ """\
+ letters,number\n\
+ 观音,1\n\
+ Ποσειδῶν,456"""
+ )
diff --git a/tests/tabular_output/test_output_formatter.py b/tests/tabular_output/test_output_formatter.py
new file mode 100644
index 0000000..8e1fa92
--- /dev/null
+++ b/tests/tabular_output/test_output_formatter.py
@@ -0,0 +1,271 @@
+# -*- coding: utf-8 -*-
+"""Test the generic output formatter interface."""
+
+from __future__ import unicode_literals
+from decimal import Decimal
+from textwrap import dedent
+
+import pytest
+
+from cli_helpers.tabular_output import format_output, TabularOutputFormatter
+from cli_helpers.compat import binary_type, text_type
+from cli_helpers.utils import strip_ansi
+
+
+def test_tabular_output_formatter():
+ """Test the TabularOutputFormatter class."""
+ headers = ["text", "numeric"]
+ data = [
+ ["abc", Decimal(1)],
+ ["defg", Decimal("11.1")],
+ ["hi", Decimal("1.1")],
+ ["Pablo\rß\n", 0],
+ ]
+ expected = dedent(
+ """\
+ +-------+---------+
+ | text | numeric |
+ +-------+---------+
+ | abc | 1 |
+ | defg | 11.1 |
+ | hi | 1.1 |
+ | Pablo | 0 |
+ | ß | |
+ +-------+---------+"""
+ )
+
+ print(expected)
+ print(
+ "\n".join(
+ TabularOutputFormatter().format_output(
+ iter(data), headers, format_name="ascii"
+ )
+ )
+ )
+ assert expected == "\n".join(
+ TabularOutputFormatter().format_output(iter(data), headers, format_name="ascii")
+ )
+
+
+def test_tabular_output_escaped():
+ """Test the ascii_escaped output format."""
+ headers = ["text", "numeric"]
+ data = [
+ ["abc", Decimal(1)],
+ ["defg", Decimal("11.1")],
+ ["hi", Decimal("1.1")],
+ ["Pablo\rß\n", 0],
+ ]
+ expected = dedent(
+ """\
+ +------------+---------+
+ | text | numeric |
+ +------------+---------+
+ | abc | 1 |
+ | defg | 11.1 |
+ | hi | 1.1 |
+ | Pablo\\rß\\n | 0 |
+ +------------+---------+"""
+ )
+
+ print(expected)
+ print(
+ "\n".join(
+ TabularOutputFormatter().format_output(
+ iter(data), headers, format_name="ascii_escaped"
+ )
+ )
+ )
+ assert expected == "\n".join(
+ TabularOutputFormatter().format_output(
+ iter(data), headers, format_name="ascii_escaped"
+ )
+ )
+
+
+def test_tabular_format_output_wrapper():
+ """Test the format_output wrapper."""
+ data = [["1", None], ["2", "Sam"], ["3", "Joe"]]
+ headers = ["id", "name"]
+ expected = dedent(
+ """\
+ +----+------+
+ | id | name |
+ +----+------+
+ | 1 | N/A |
+ | 2 | Sam |
+ | 3 | Joe |
+ +----+------+"""
+ )
+
+ assert expected == "\n".join(
+ format_output(iter(data), headers, format_name="ascii", missing_value="N/A")
+ )
+
+
+def test_additional_preprocessors():
+ """Test that additional preprocessors are run."""
+
+ def hello_world(data, headers, **_):
+ def hello_world_data(data):
+ for row in data:
+ for i, value in enumerate(row):
+ if value == "hello":
+ row[i] = "{}, world".format(value)
+ yield row
+
+ return hello_world_data(data), headers
+
+ data = [["foo", None], ["hello!", "hello"]]
+ headers = "ab"
+
+ expected = dedent(
+ """\
+ +--------+--------------+
+ | a | b |
+ +--------+--------------+
+ | foo | hello |
+ | hello! | hello, world |
+ +--------+--------------+"""
+ )
+
+ assert expected == "\n".join(
+ TabularOutputFormatter().format_output(
+ iter(data),
+ headers,
+ format_name="ascii",
+ preprocessors=(hello_world,),
+ missing_value="hello",
+ )
+ )
+
+
+def test_format_name_attribute():
+ """Test the the format_name attribute be set and retrieved."""
+ formatter = TabularOutputFormatter(format_name="plain")
+ assert formatter.format_name == "plain"
+ formatter.format_name = "simple"
+ assert formatter.format_name == "simple"
+
+ with pytest.raises(ValueError):
+ formatter.format_name = "foobar"
+
+
+def test_headless_tabulate_format():
+ """Test that a headless formatter doesn't display headers"""
+ formatter = TabularOutputFormatter(format_name="minimal")
+ headers = ["text", "numeric"]
+ data = [["a"], ["b"], ["c"]]
+ expected = "a\nb\nc"
+ assert expected == "\n".join(
+ TabularOutputFormatter().format_output(
+ iter(data),
+ headers,
+ format_name="minimal",
+ )
+ )
+
+
+def test_unsupported_format():
+ """Test that TabularOutputFormatter rejects unknown formats."""
+ formatter = TabularOutputFormatter()
+
+ with pytest.raises(ValueError):
+ formatter.format_name = "foobar"
+
+ with pytest.raises(ValueError):
+ formatter.format_output((), (), format_name="foobar")
+
+
+def test_tabulate_ansi_escape_in_default_value():
+ """Test that ANSI escape codes work with tabulate."""
+
+ data = [["1", None], ["2", "Sam"], ["3", "Joe"]]
+ headers = ["id", "name"]
+
+ styled = format_output(
+ iter(data),
+ headers,
+ format_name="psql",
+ missing_value="\x1b[38;5;10mNULL\x1b[39m",
+ )
+ unstyled = format_output(
+ iter(data), headers, format_name="psql", missing_value="NULL"
+ )
+
+ stripped_styled = [strip_ansi(s) for s in styled]
+
+ assert list(unstyled) == stripped_styled
+
+
+def test_get_type():
+ """Test that _get_type returns the expected type."""
+ formatter = TabularOutputFormatter()
+
+ tests = (
+ (1, int),
+ (2.0, float),
+ (b"binary", binary_type),
+ ("text", text_type),
+ (None, type(None)),
+ ((), text_type),
+ )
+
+ for value, data_type in tests:
+ assert data_type is formatter._get_type(value)
+
+
+def test_provide_column_types():
+ """Test that provided column types are passed to preprocessors."""
+ expected_column_types = (bool, float)
+ data = ((1, 1.0), (0, 2))
+ headers = ("a", "b")
+
+ def preprocessor(data, headers, column_types=(), **_):
+ assert expected_column_types == column_types
+ return data, headers
+
+ format_output(
+ data,
+ headers,
+ "csv",
+ column_types=expected_column_types,
+ preprocessors=(preprocessor,),
+ )
+
+
+def test_enforce_iterable():
+ """Test that all output formatters accept iterable"""
+ formatter = TabularOutputFormatter()
+ loremipsum = (
+ "lorem ipsum dolor sit amet consectetur adipiscing elit sed do eiusmod".split(
+ " "
+ )
+ )
+
+ for format_name in formatter.supported_formats:
+ formatter.format_name = format_name
+ try:
+ formatted = next(formatter.format_output(zip(loremipsum), ["lorem"]))
+ except TypeError:
+ assert False, "{0} doesn't return iterable".format(format_name)
+
+
+@pytest.mark.parametrize(
+ "extra_kwargs",
+ [
+ {},
+ {"style": "default"},
+ {"style": "colorful"},
+ ],
+)
+def test_all_text_type(extra_kwargs):
+ """Test the TabularOutputFormatter class."""
+ data = [[1, "", None, Decimal(2)]]
+ headers = ["col1", "col2", "col3", "col4"]
+ output_formatter = TabularOutputFormatter()
+ for format_name in output_formatter.supported_formats:
+ for row in output_formatter.format_output(
+ iter(data), headers, format_name=format_name, **extra_kwargs
+ ):
+ assert isinstance(row, text_type), "not unicode for {}".format(format_name)
diff --git a/tests/tabular_output/test_preprocessors.py b/tests/tabular_output/test_preprocessors.py
new file mode 100644
index 0000000..e428bfa
--- /dev/null
+++ b/tests/tabular_output/test_preprocessors.py
@@ -0,0 +1,350 @@
+# -*- coding: utf-8 -*-
+"""Test CLI Helpers' tabular output preprocessors."""
+
+from __future__ import unicode_literals
+from decimal import Decimal
+
+import pytest
+
+from cli_helpers.compat import HAS_PYGMENTS
+from cli_helpers.tabular_output.preprocessors import (
+ align_decimals,
+ bytes_to_string,
+ convert_to_string,
+ quote_whitespaces,
+ override_missing_value,
+ override_tab_value,
+ style_output,
+ format_numbers,
+)
+
+if HAS_PYGMENTS:
+ from pygments.style import Style
+ from pygments.token import Token
+
+import inspect
+import cli_helpers.tabular_output.preprocessors
+import types
+
+
+def test_convert_to_string():
+ """Test the convert_to_string() function."""
+ data = [[1, "John"], [2, "Jill"]]
+ headers = [0, "name"]
+ expected = ([["1", "John"], ["2", "Jill"]], ["0", "name"])
+ results = convert_to_string(data, headers)
+
+ assert expected == (list(results[0]), results[1])
+
+
+def test_override_missing_values():
+ """Test the override_missing_values() function."""
+ data = [[1, None], [2, "Jill"]]
+ headers = [0, "name"]
+ expected = ([[1, "<EMPTY>"], [2, "Jill"]], [0, "name"])
+ results = override_missing_value(data, headers, missing_value="<EMPTY>")
+
+ assert expected == (list(results[0]), results[1])
+
+
+@pytest.mark.skipif(not HAS_PYGMENTS, reason="requires the Pygments library")
+def test_override_missing_value_with_style():
+ """Test that *override_missing_value()* styles output."""
+
+ class NullStyle(Style):
+ styles = {Token.Output.Null: "#0f0"}
+
+ headers = ["h1", "h2"]
+ data = [[None, "2"], ["abc", None]]
+
+ expected_headers = ["h1", "h2"]
+ expected_data = [
+ ["\x1b[38;5;10m<null>\x1b[39m", "2"],
+ ["abc", "\x1b[38;5;10m<null>\x1b[39m"],
+ ]
+ results = override_missing_value(
+ data, headers, style=NullStyle, missing_value="<null>"
+ )
+
+ assert (expected_data, expected_headers) == (list(results[0]), results[1])
+
+
+def test_override_tab_value():
+ """Test the override_tab_value() function."""
+ data = [[1, "\tJohn"], [2, "Jill"]]
+ headers = ["id", "name"]
+ expected = ([[1, " John"], [2, "Jill"]], ["id", "name"])
+ results = override_tab_value(data, headers)
+
+ assert expected == (list(results[0]), results[1])
+
+
+def test_bytes_to_string():
+ """Test the bytes_to_string() function."""
+ data = [[1, "John"], [2, b"Jill"]]
+ headers = [0, "name"]
+ expected = ([[1, "John"], [2, "Jill"]], [0, "name"])
+ results = bytes_to_string(data, headers)
+
+ assert expected == (list(results[0]), results[1])
+
+
+def test_align_decimals():
+ """Test the align_decimals() function."""
+ data = [[Decimal("200"), Decimal("1")], [Decimal("1.00002"), Decimal("1.0")]]
+ headers = ["num1", "num2"]
+ column_types = (float, float)
+ expected = ([["200", "1"], [" 1.00002", "1.0"]], ["num1", "num2"])
+ results = align_decimals(data, headers, column_types=column_types)
+
+ assert expected == (list(results[0]), results[1])
+
+
+def test_align_decimals_empty_result():
+ """Test align_decimals() with no results."""
+ data = []
+ headers = ["num1", "num2"]
+ column_types = ()
+ expected = ([], ["num1", "num2"])
+ results = align_decimals(data, headers, column_types=column_types)
+
+ assert expected == (list(results[0]), results[1])
+
+
+def test_align_decimals_non_decimals():
+ """Test align_decimals() with non-decimals."""
+ data = [[Decimal("200.000"), Decimal("1.000")], [None, None]]
+ headers = ["num1", "num2"]
+ column_types = (float, float)
+ expected = ([["200.000", "1.000"], [None, None]], ["num1", "num2"])
+ results = align_decimals(data, headers, column_types=column_types)
+
+ assert expected == (list(results[0]), results[1])
+
+
+def test_quote_whitespaces():
+ """Test the quote_whitespaces() function."""
+ data = [[" before", "after "], [" both ", "none"]]
+ headers = ["h1", "h2"]
+ expected = ([["' before'", "'after '"], ["' both '", "'none'"]], ["h1", "h2"])
+ results = quote_whitespaces(data, headers)
+
+ assert expected == (list(results[0]), results[1])
+
+
+def test_quote_whitespaces_empty_result():
+ """Test the quote_whitespaces() function with no results."""
+ data = []
+ headers = ["h1", "h2"]
+ expected = ([], ["h1", "h2"])
+ results = quote_whitespaces(data, headers)
+
+ assert expected == (list(results[0]), results[1])
+
+
+def test_quote_whitespaces_non_spaces():
+ """Test the quote_whitespaces() function with non-spaces."""
+ data = [["\tbefore", "after \r"], ["\n both ", "none"]]
+ headers = ["h1", "h2"]
+ expected = ([["'\tbefore'", "'after \r'"], ["'\n both '", "'none'"]], ["h1", "h2"])
+ results = quote_whitespaces(data, headers)
+
+ assert expected == (list(results[0]), results[1])
+
+
+@pytest.mark.skipif(not HAS_PYGMENTS, reason="requires the Pygments library")
+def test_style_output_no_styles():
+ """Test that *style_output()* does not style without styles."""
+ headers = ["h1", "h2"]
+ data = [["1", "2"], ["a", "b"]]
+ results = style_output(data, headers)
+
+ assert (data, headers) == (list(results[0]), results[1])
+
+
+@pytest.mark.skipif(HAS_PYGMENTS, reason="requires the Pygments library be missing")
+def test_style_output_no_pygments():
+ """Test that *style_output()* does not try to style without Pygments."""
+ headers = ["h1", "h2"]
+ data = [["1", "2"], ["a", "b"]]
+ results = style_output(data, headers)
+
+ assert (data, headers) == (list(results[0]), results[1])
+
+
+@pytest.mark.skipif(not HAS_PYGMENTS, reason="requires the Pygments library")
+def test_style_output():
+ """Test that *style_output()* styles output."""
+
+ class CliStyle(Style):
+ default_style = ""
+ styles = {
+ Token.Output.Header: "bold ansibrightred",
+ Token.Output.OddRow: "bg:#eee #111",
+ Token.Output.EvenRow: "#0f0",
+ }
+
+ headers = ["h1", "h2"]
+ data = [["观音", "2"], ["Ποσειδῶν", "b"]]
+
+ expected_headers = ["\x1b[91;01mh1\x1b[39;00m", "\x1b[91;01mh2\x1b[39;00m"]
+ expected_data = [
+ ["\x1b[38;5;233;48;5;7m观音\x1b[39;49m", "\x1b[38;5;233;48;5;7m2\x1b[39;49m"],
+ ["\x1b[38;5;10mΠοσειδῶν\x1b[39m", "\x1b[38;5;10mb\x1b[39m"],
+ ]
+ results = style_output(data, headers, style=CliStyle)
+
+ assert (expected_data, expected_headers) == (list(results[0]), results[1])
+
+
+@pytest.mark.skipif(not HAS_PYGMENTS, reason="requires the Pygments library")
+def test_style_output_with_newlines():
+ """Test that *style_output()* styles output with newlines in it."""
+
+ class CliStyle(Style):
+ default_style = ""
+ styles = {
+ Token.Output.Header: "bold ansibrightred",
+ Token.Output.OddRow: "bg:#eee #111",
+ Token.Output.EvenRow: "#0f0",
+ }
+
+ headers = ["h1", "h2"]
+ data = [["观音\nLine2", "Ποσειδῶν"]]
+
+ expected_headers = ["\x1b[91;01mh1\x1b[39;00m", "\x1b[91;01mh2\x1b[39;00m"]
+ expected_data = [
+ [
+ "\x1b[38;5;233;48;5;7m观音\x1b[39;49m\n\x1b[38;5;233;48;5;7m"
+ "Line2\x1b[39;49m",
+ "\x1b[38;5;233;48;5;7mΠοσειδῶν\x1b[39;49m",
+ ]
+ ]
+ results = style_output(data, headers, style=CliStyle)
+
+ assert (expected_data, expected_headers) == (list(results[0]), results[1])
+
+
+@pytest.mark.skipif(not HAS_PYGMENTS, reason="requires the Pygments library")
+def test_style_output_custom_tokens():
+ """Test that *style_output()* styles output with custom token names."""
+
+ class CliStyle(Style):
+ default_style = ""
+ styles = {
+ Token.Results.Headers: "bold ansibrightred",
+ Token.Results.OddRows: "bg:#eee #111",
+ Token.Results.EvenRows: "#0f0",
+ }
+
+ headers = ["h1", "h2"]
+ data = [["1", "2"], ["a", "b"]]
+
+ expected_headers = ["\x1b[91;01mh1\x1b[39;00m", "\x1b[91;01mh2\x1b[39;00m"]
+ expected_data = [
+ ["\x1b[38;5;233;48;5;7m1\x1b[39;49m", "\x1b[38;5;233;48;5;7m2\x1b[39;49m"],
+ ["\x1b[38;5;10ma\x1b[39m", "\x1b[38;5;10mb\x1b[39m"],
+ ]
+
+ output = style_output(
+ data,
+ headers,
+ style=CliStyle,
+ header_token=Token.Results.Headers,
+ odd_row_token=Token.Results.OddRows,
+ even_row_token=Token.Results.EvenRows,
+ )
+
+ assert (expected_data, expected_headers) == (list(output[0]), output[1])
+
+
+def test_format_integer():
+ """Test formatting for an INTEGER datatype."""
+ data = [[1], [1000], [1000000]]
+ headers = ["h1"]
+ result_data, result_headers = format_numbers(
+ data, headers, column_types=(int,), integer_format=",", float_format=","
+ )
+
+ expected = [["1"], ["1,000"], ["1,000,000"]]
+ assert expected == list(result_data)
+ assert headers == result_headers
+
+
+def test_format_decimal():
+ """Test formatting for a DECIMAL(12, 4) datatype."""
+ data = [[Decimal("1.0000")], [Decimal("1000.0000")], [Decimal("1000000.0000")]]
+ headers = ["h1"]
+ result_data, result_headers = format_numbers(
+ data, headers, column_types=(float,), integer_format=",", float_format=","
+ )
+
+ expected = [["1.0000"], ["1,000.0000"], ["1,000,000.0000"]]
+ assert expected == list(result_data)
+ assert headers == result_headers
+
+
+def test_format_float():
+ """Test formatting for a REAL datatype."""
+ data = [[1.0], [1000.0], [1000000.0]]
+ headers = ["h1"]
+ result_data, result_headers = format_numbers(
+ data, headers, column_types=(float,), integer_format=",", float_format=","
+ )
+ expected = [["1.0"], ["1,000.0"], ["1,000,000.0"]]
+ assert expected == list(result_data)
+ assert headers == result_headers
+
+
+def test_format_integer_only():
+ """Test that providing one format string works."""
+ data = [[1, 1.0], [1000, 1000.0], [1000000, 1000000.0]]
+ headers = ["h1", "h2"]
+ result_data, result_headers = format_numbers(
+ data, headers, column_types=(int, float), integer_format=","
+ )
+
+ expected = [["1", 1.0], ["1,000", 1000.0], ["1,000,000", 1000000.0]]
+ assert expected == list(result_data)
+ assert headers == result_headers
+
+
+def test_format_numbers_no_format_strings():
+ """Test that numbers aren't formatted without format strings."""
+ data = ((1), (1000), (1000000))
+ headers = ("h1",)
+ result_data, result_headers = format_numbers(data, headers, column_types=(int,))
+ assert list(data) == list(result_data)
+ assert headers == result_headers
+
+
+def test_format_numbers_no_column_types():
+ """Test that numbers aren't formatted without column types."""
+ data = ((1), (1000), (1000000))
+ headers = ("h1",)
+ result_data, result_headers = format_numbers(
+ data, headers, integer_format=",", float_format=","
+ )
+ assert list(data) == list(result_data)
+ assert headers == result_headers
+
+
+def test_enforce_iterable():
+ preprocessors = inspect.getmembers(
+ cli_helpers.tabular_output.preprocessors, inspect.isfunction
+ )
+ loremipsum = (
+ "lorem ipsum dolor sit amet consectetur adipiscing elit sed do eiusmod".split(
+ " "
+ )
+ )
+ for name, preprocessor in preprocessors:
+ preprocessed = preprocessor(zip(loremipsum), ["lorem"], column_types=(str,))
+ try:
+ first = next(preprocessed[0])
+ except StopIteration:
+ assert False, "{} gives no output with iterator data".format(name)
+ except TypeError:
+ assert False, "{} doesn't return iterable".format(name)
+ if isinstance(preprocessed[1], types.GeneratorType):
+ assert False, "{} returns headers as iterator".format(name)
diff --git a/tests/tabular_output/test_tabulate_adapter.py b/tests/tabular_output/test_tabulate_adapter.py
new file mode 100644
index 0000000..6e7c7db
--- /dev/null
+++ b/tests/tabular_output/test_tabulate_adapter.py
@@ -0,0 +1,113 @@
+# -*- coding: utf-8 -*-
+"""Test the tabulate output adapter."""
+
+from __future__ import unicode_literals
+from textwrap import dedent
+
+import pytest
+
+from cli_helpers.compat import HAS_PYGMENTS
+from cli_helpers.tabular_output import tabulate_adapter
+
+if HAS_PYGMENTS:
+ from pygments.style import Style
+ from pygments.token import Token
+
+
+def test_tabulate_wrapper():
+ """Test the *output_formatter.tabulate_wrapper()* function."""
+ data = [["abc", 1], ["d", 456]]
+ headers = ["letters", "number"]
+ output = tabulate_adapter.adapter(iter(data), headers, table_format="psql")
+ assert "\n".join(output) == dedent(
+ """\
+ +---------+--------+
+ | letters | number |
+ |---------+--------|
+ | abc | 1 |
+ | d | 456 |
+ +---------+--------+"""
+ )
+
+ data = [["abc", 1], ["d", 456]]
+ headers = ["letters", "number"]
+ output = tabulate_adapter.adapter(iter(data), headers, table_format="psql_unicode")
+ assert "\n".join(output) == dedent(
+ """\
+ ┌─────────┬────────┐
+ │ letters │ number │
+ ├─────────┼────────┤
+ │ abc │ 1 │
+ │ d │ 456 │
+ └─────────┴────────┘"""
+ )
+
+ data = [["{1,2,3}", "{{1,2},{3,4}}", "{å,魚,текст}"], ["{}", "<null>", "{<null>}"]]
+ headers = ["bigint_array", "nested_numeric_array", "配列"]
+ output = tabulate_adapter.adapter(iter(data), headers, table_format="psql")
+ assert "\n".join(output) == dedent(
+ """\
+ +--------------+----------------------+--------------+
+ | bigint_array | nested_numeric_array | 配列 |
+ |--------------+----------------------+--------------|
+ | {1,2,3} | {{1,2},{3,4}} | {å,魚,текст} |
+ | {} | <null> | {<null>} |
+ +--------------+----------------------+--------------+"""
+ )
+
+
+def test_markup_format():
+ """Test that markup formats do not have number align or string align."""
+ data = [["abc", 1], ["d", 456]]
+ headers = ["letters", "number"]
+ output = tabulate_adapter.adapter(iter(data), headers, table_format="mediawiki")
+ assert "\n".join(output) == dedent(
+ """\
+ {| class="wikitable" style="text-align: left;"
+ |+ <!-- caption -->
+ |-
+ ! letters !! number
+ |-
+ | abc || 1
+ |-
+ | d || 456
+ |}"""
+ )
+
+
+@pytest.mark.skipif(not HAS_PYGMENTS, reason="requires the Pygments library")
+def test_style_output_table():
+ """Test that *style_output_table()* styles the output table."""
+
+ class CliStyle(Style):
+ default_style = ""
+ styles = {
+ Token.Output.TableSeparator: "ansibrightred",
+ }
+
+ headers = ["h1", "h2"]
+ data = [["观音", "2"], ["Ποσειδῶν", "b"]]
+ style_output_table = tabulate_adapter.style_output_table("psql")
+
+ style_output_table(data, headers, style=CliStyle)
+ output = tabulate_adapter.adapter(iter(data), headers, table_format="psql")
+ PLUS = "\x1b[91m+\x1b[39m"
+ MINUS = "\x1b[91m-\x1b[39m"
+ PIPE = "\x1b[91m|\x1b[39m"
+
+ expected = (
+ dedent(
+ """\
+ +----------+----+
+ | h1 | h2 |
+ |----------+----|
+ | 观音 | 2 |
+ | Ποσειδῶν | b |
+ +----------+----+"""
+ )
+ .replace("+", PLUS)
+ .replace("-", MINUS)
+ .replace("|", PIPE)
+ )
+
+ assert "\n".join(output) == expected
diff --git a/tests/tabular_output/test_tsv_output_adapter.py b/tests/tabular_output/test_tsv_output_adapter.py
new file mode 100644
index 0000000..9249d87
--- /dev/null
+++ b/tests/tabular_output/test_tsv_output_adapter.py
@@ -0,0 +1,36 @@
+# -*- coding: utf-8 -*-
+"""Test the tsv delimited output adapter."""
+
+from __future__ import unicode_literals
+from textwrap import dedent
+
+import pytest
+
+from cli_helpers.tabular_output import tsv_output_adapter
+
+
+def test_tsv_wrapper():
+ """Test the tsv output adapter."""
+ # Test tab-delimited output.
+ data = [["ab\r\nc", "1"], ["d", "456"]]
+ headers = ["letters", "number"]
+ output = tsv_output_adapter.adapter(iter(data), headers, table_format="tsv")
+ assert "\n".join(output) == dedent(
+ """\
+ letters\tnumber\n\
+ ab\r\\nc\t1\n\
+ d\t456"""
+ )
+
+
+def test_unicode_with_tsv():
+ """Test that the tsv wrapper can handle non-ascii characters."""
+ data = [["观音", "1"], ["Ποσειδῶν", "456"]]
+ headers = ["letters", "number"]
+ output = tsv_output_adapter.adapter(data, headers)
+ assert "\n".join(output) == dedent(
+ """\
+ letters\tnumber\n\
+ 观音\t1\n\
+ Ποσειδῶν\t456"""
+ )
diff --git a/tests/tabular_output/test_vertical_table_adapter.py b/tests/tabular_output/test_vertical_table_adapter.py
new file mode 100644
index 0000000..359d9d9
--- /dev/null
+++ b/tests/tabular_output/test_vertical_table_adapter.py
@@ -0,0 +1,49 @@
+# -*- coding: utf-8 -*-
+"""Test the vertical table formatter."""
+
+from textwrap import dedent
+
+from cli_helpers.compat import text_type
+from cli_helpers.tabular_output import vertical_table_adapter
+
+
+def test_vertical_table():
+ """Test the default settings for vertical_table()."""
+ results = [("hello", text_type(123)), ("world", text_type(456))]
+
+ expected = dedent(
+ """\
+ ***************************[ 1. row ]***************************
+ name | hello
+ age | 123
+ ***************************[ 2. row ]***************************
+ name | world
+ age | 456"""
+ )
+ assert expected == "\n".join(
+ vertical_table_adapter.adapter(results, ("name", "age"))
+ )
+
+
+def test_vertical_table_customized():
+ """Test customized settings for vertical_table()."""
+ results = [("john", text_type(47)), ("jill", text_type(50))]
+
+ expected = dedent(
+ """\
+ -[ PERSON 1 ]-----
+ name | john
+ age | 47
+ -[ PERSON 2 ]-----
+ name | jill
+ age | 50"""
+ )
+ assert expected == "\n".join(
+ vertical_table_adapter.adapter(
+ results,
+ ("name", "age"),
+ sep_title="PERSON {n}",
+ sep_character="-",
+ sep_length=(1, 5),
+ )
+ )