##// END OF EJS Templates
capture_output does not respect trailing semicolon (#13940)...
capture_output does not respect trailing semicolon (#13940) This change tests whether the Python code inside the `%%capture` magic ends with a `';'`. If it does, the code no longer creates the object that is supposed to store the cell output. If such object already exists, it is deleted. Please review and suggest changes if you think they can improve the solution. Thanks.

File last commit:

r27764:aefe51c6
r28128:75ecfe93 merge
Show More
test_text.py
207 lines | 6.9 KiB | text/x-python | PythonLexer
# encoding: utf-8
"""Tests for IPython.utils.text"""
#-----------------------------------------------------------------------------
# Copyright (C) 2011 The IPython Development Team
#
# Distributed under the terms of the BSD License. The full license is in
# the file COPYING, distributed as part of this software.
#-----------------------------------------------------------------------------
#-----------------------------------------------------------------------------
# Imports
#-----------------------------------------------------------------------------
import os
import math
import random
from pathlib import Path
import pytest
from IPython.utils import text
#-----------------------------------------------------------------------------
# Globals
#-----------------------------------------------------------------------------
def test_columnize():
"""Basic columnize tests."""
size = 5
items = [l*size for l in 'abcd']
out = text.columnize(items, displaywidth=80)
assert out == "aaaaa bbbbb ccccc ddddd\n"
out = text.columnize(items, displaywidth=25)
assert out == "aaaaa ccccc\nbbbbb ddddd\n"
out = text.columnize(items, displaywidth=12)
assert out == "aaaaa ccccc\nbbbbb ddddd\n"
out = text.columnize(items, displaywidth=10)
assert out == "aaaaa\nbbbbb\nccccc\nddddd\n"
out = text.columnize(items, row_first=True, displaywidth=80)
assert out == "aaaaa bbbbb ccccc ddddd\n"
out = text.columnize(items, row_first=True, displaywidth=25)
assert out == "aaaaa bbbbb\nccccc ddddd\n"
out = text.columnize(items, row_first=True, displaywidth=12)
assert out == "aaaaa bbbbb\nccccc ddddd\n"
out = text.columnize(items, row_first=True, displaywidth=10)
assert out == "aaaaa\nbbbbb\nccccc\nddddd\n"
out = text.columnize(items, displaywidth=40, spread=True)
assert out == "aaaaa bbbbb ccccc ddddd\n"
out = text.columnize(items, displaywidth=20, spread=True)
assert out == "aaaaa ccccc\nbbbbb ddddd\n"
out = text.columnize(items, displaywidth=12, spread=True)
assert out == "aaaaa ccccc\nbbbbb ddddd\n"
out = text.columnize(items, displaywidth=10, spread=True)
assert out == "aaaaa\nbbbbb\nccccc\nddddd\n"
def test_columnize_random():
"""Test with random input to hopefully catch edge case """
for row_first in [True, False]:
for nitems in [random.randint(2,70) for i in range(2,20)]:
displaywidth = random.randint(20,200)
rand_len = [random.randint(2,displaywidth) for i in range(nitems)]
items = ['x'*l for l in rand_len]
out = text.columnize(items, row_first=row_first, displaywidth=displaywidth)
longer_line = max([len(x) for x in out.split('\n')])
longer_element = max(rand_len)
assert longer_line <= displaywidth, (
f"Columnize displayed something lager than displaywidth : {longer_line}\n"
f"longer element : {longer_element}\n"
f"displaywidth : {displaywidth}\n"
f"number of element : {nitems}\n"
f"size of each element : {rand_len}\n"
f"row_first={row_first}\n"
)
@pytest.mark.parametrize("row_first", [True, False])
def test_columnize_medium(row_first):
"""Test with inputs than shouldn't be wider than 80"""
size = 40
items = [l*size for l in 'abc']
out = text.columnize(items, row_first=row_first, displaywidth=80)
assert out == "\n".join(items + [""]), "row_first={0}".format(row_first)
@pytest.mark.parametrize("row_first", [True, False])
def test_columnize_long(row_first):
"""Test columnize with inputs longer than the display window"""
size = 11
items = [l*size for l in 'abc']
out = text.columnize(items, row_first=row_first, displaywidth=size - 1)
assert out == "\n".join(items + [""]), "row_first={0}".format(row_first)
def eval_formatter_check(f):
ns = dict(n=12, pi=math.pi, stuff='hello there', os=os, u=u"café", b="café")
s = f.format("{n} {n//4} {stuff.split()[0]}", **ns)
assert s == "12 3 hello"
s = f.format(" ".join(["{n//%i}" % i for i in range(1, 8)]), **ns)
assert s == "12 6 4 3 2 2 1"
s = f.format("{[n//i for i in range(1,8)]}", **ns)
assert s == "[12, 6, 4, 3, 2, 2, 1]"
s = f.format("{stuff!s}", **ns)
assert s == ns["stuff"]
s = f.format("{stuff!r}", **ns)
assert s == repr(ns["stuff"])
# Check with unicode:
s = f.format("{u}", **ns)
assert s == ns["u"]
# This decodes in a platform dependent manner, but it shouldn't error out
s = f.format("{b}", **ns)
pytest.raises(NameError, f.format, "{dne}", **ns)
def eval_formatter_slicing_check(f):
ns = dict(n=12, pi=math.pi, stuff='hello there', os=os)
s = f.format(" {stuff.split()[:]} ", **ns)
assert s == " ['hello', 'there'] "
s = f.format(" {stuff.split()[::-1]} ", **ns)
assert s == " ['there', 'hello'] "
s = f.format("{stuff[::2]}", **ns)
assert s == ns["stuff"][::2]
pytest.raises(SyntaxError, f.format, "{n:x}", **ns)
def eval_formatter_no_slicing_check(f):
ns = dict(n=12, pi=math.pi, stuff="hello there", os=os)
s = f.format("{n:x} {pi**2:+f}", **ns)
assert s == "c +9.869604"
s = f.format("{stuff[slice(1,4)]}", **ns)
assert s == "ell"
s = f.format("{a[:]}", a=[1, 2])
assert s == "[1, 2]"
def test_eval_formatter():
f = text.EvalFormatter()
eval_formatter_check(f)
eval_formatter_no_slicing_check(f)
def test_full_eval_formatter():
f = text.FullEvalFormatter()
eval_formatter_check(f)
eval_formatter_slicing_check(f)
def test_dollar_formatter():
f = text.DollarFormatter()
eval_formatter_check(f)
eval_formatter_slicing_check(f)
ns = dict(n=12, pi=math.pi, stuff='hello there', os=os)
s = f.format("$n", **ns)
assert s == "12"
s = f.format("$n.real", **ns)
assert s == "12"
s = f.format("$n/{stuff[:5]}", **ns)
assert s == "12/hello"
s = f.format("$n $$HOME", **ns)
assert s == "12 $HOME"
s = f.format("${foo}", foo="HOME")
assert s == "$HOME"
def test_strip_email():
src = """\
>> >>> def f(x):
>> ... return x+1
>> ...
>> >>> zz = f(2.5)"""
cln = """\
>>> def f(x):
... return x+1
...
>>> zz = f(2.5)"""
assert text.strip_email_quotes(src) == cln
def test_strip_email2():
src = "> > > list()"
cln = "list()"
assert text.strip_email_quotes(src) == cln
def test_LSString():
lss = text.LSString("abc\ndef")
assert lss.l == ["abc", "def"]
assert lss.s == "abc def"
lss = text.LSString(os.getcwd())
assert isinstance(lss.p[0], Path)
def test_SList():
sl = text.SList(["a 11", "b 1", "a 2"])
assert sl.n == "a 11\nb 1\na 2"
assert sl.s == "a 11 b 1 a 2"
assert sl.grep(lambda x: x.startswith("a")) == text.SList(["a 11", "a 2"])
assert sl.fields(0) == text.SList(["a", "b", "a"])
assert sl.sort(field=1, nums=True) == text.SList(["b 1", "a 2", "a 11"])