Use better temporary files mechanism in install_unpacked_wheel

[sbidoul]

install_unpacked_wheel creates files such as INSTALLER or RECORD with a .pip extension then renames them to the intended file in order to have an atomic file creation.

Example:

pip/src/pip/_internal/operations/install/wheel.py

Lines 589 to 603 in 9e24b58

	# Record details of all files installed
	record = os.path.join(dest_info_dir, 'RECORD')
	temp_record = os.path.join(dest_info_dir, 'RECORD.pip')
	with open_for_csv(record, 'r') as record_in:
	with open_for_csv(temp_record, 'w+') as record_out:
	reader = csv.reader(record_in)
	outrows = get_csv_rows_for_installed(
	reader, installed=installed, changed=changed,
	generated=generated, lib_dir=lib_dir,
	)
	writer = csv.writer(record_out)
	# Sort to simplify testing.
	for row in sorted_outrows(outrows):
	writer.writerow(row)
	shutil.move(temp_record, record)

The goal of this issue is to use a better and safer approach.

[rudolfovic]

adjacent_tmp_file opens files as binary while the csvwriter insists on writing strings no matter what is passed to it. A possible solution would be to add a parameter to adjacent_tmp_file so that it passes it accordingly to NamedTemporaryFile.
NamedTemporaryFile uses mode='w+b' by default and adjacent_tmp_file does not override this.
This fix passes the tests.
Mode can either be added to adjacent_tmp_file as a named parameter so that it does not have to be specified in other cases (of which there are currently two).
From what I found it is in line with the pip conventions
e.g. def decode(input, fallback_encoding, errors='replace'): in webencodings/__init__.py.
Let me know what you think.

[uranusjr]

Would it be enough to keep csv.writer happy if we wrap the file handle in an io.TextIOWrapper?

[siayush]

Hey, @uranusjr can you please clear it more. About this "if we wrap the filehandle in an io.TextIOWrapper" .
Thanks

[uranusjr]

So the problem here is that csvwriter expects a str-based file object, but on Python 3 the caller (adjacent_tmp_file) writes file in binary mode. So I was wondering whether it would be possible to wrap the file object opened by adjacent_tmp_file with an io.TextIOWrapper on Python 3 to turn it into a text file-like object, and pass that to csvwriter instead. The wrapper would perform the necessariy encoding when csvwriter writes to it, to turn the written data into bytes for the NamedTemporaryFile file object.

[pradyunsg]

@uranusjr I suggest avoiding complicating this issue by discussing trying to do this in-memory (since it's a good first issue).

[siayush]

Thank you @uranusjr for simplifying it. Let me try it.

[McSinyx]

@siayush, may I pick this up?