gh-58451: Add optional delete_on_close parameter to NamedTemporaryFile

[Ev2geny]

This is a reincarnation of the already marked as awaiting mergePR 22431, which was subsequently closed due to some git mistake.

This PR fixes issue 58451: tempfile.NamedTemporaryFile not particularly useful on Windows

tempfile.NamedTemporaryFile is too hard to use portably when you need to open the file by name after writing it. To do that, you need to close the file first (on Windows), which means you have to pass delete=False, which in turn means that you get no help in cleaning up the actual file resource,

Hence at the moment there is no out of the box solution to use tempfile.NamedTemporaryFile on Windows in such scenario (which is often used in unit testing):

• in test module:

1. create and open temporary file
2. write data to it
3. pass name of the temporary file to the operational code

• In operational code, being tested

1. open file, using name of the temporary file
2. read data from this temporary file

In this Pull Request the issue is solved by adding an additional optional argument to NamedTemporaryFile() 'delete_on_close' (default is True). It works in combination with already existing argument 'delete', and determines the deletion behaviour.

If delete is true (the default) and delete_on_close is true (the default), the file is deleted as soon as it is closed. If delete is true and delete_on_close is false, the file is deleted on context manager exit only, if no context manager was used, then the file is deleted only when the file-like object is finalized. If delete is false, the value of delete_on_close is ignored.

So, the change shall be fully backwards compatible.

• Issue: tempfile.NamedTemporaryFile not particularly useful on Windows #58451

[Ev2geny]

The changes done to the code since it left the PR 22431

• Merge conflicts are resolved

• A new functionality is added based on the review feedback from @eryksun to cover the case if delete is true and delete_on_close is false, but the file is never used in a with context. The new functionality is that in this case the file is deleted when the instance is finalized. This is implemented by adding a __del__() method to the _TemporaryFileCloser. This required documentation update as well.
  @eryksun , please have a look at the implementation as well as documentation update.

[Ev2geny]

Ok, looks like there are some Ubuntu errors, which I will need to look at

[Ev2geny]

@eryksun , I have implemented all suggestions, but it now fails the unit test for the issue18879

deftest_method_lookup(self):
# Issue #18879: Looking up a temporary file method should keep it# alive long enough.f=self.do_create()
wr=weakref.ref(f)
write=f.writewrite2=f.writedelfwrite(b'foo')

[eryksun]

I have implemented all suggestions, but it now fails the unit test for the issue18879

The delegated attribute cache is a problem. It leads to a reference cycle if the function wrapper (bpo-18879) is changed to keep a reference to the file wrapper (i.e. self) instead of the file closer (i.e. self._closer). Apparently avoiding this reference cycle was the reason the closer was added.

So, sorry to have misled you, but it looks like we're going to have to implement all delete operations in the closer. The wrapper just needs to change __init__() to instantiate the closer with delete_on_close, and change __exit__() to call the closer's cleanup() method. The latter includes a call to self.file.close() if needed. Since all deleting is handled by the closer, there's no need to implement __del__() in the wrapper. For example:

class_TemporaryFileCloser:
"""A separate object allowing proper closing of a temporary file's underlying file object, without adding a __del__ method to the temporary file."""cleanup_called=Falseclose_called=Falsedef__init__(self, file, name, delete=True, delete_on_close=True):
self.file=fileself.name=nameself.delete=deleteself.delete_on_close=delete_on_closedefcleanup(self, windows=(_os.name=='nt'), unlink=_os.unlink):
ifnotself.cleanup_called:
self.cleanup_called=Truetry:
ifnotself.close_called:
self.close_called=Trueself.file.close()
finally:
ifself.deleteandnot (self.delete_on_closeandwindows):
try:
unlink(self.name)
exceptFileNotFoundError:
passdefclose(self):
ifnotself.close_called:
self.close_called=Truetry:
self.file.close()
finally:
ifself.deleteandself.delete_on_close:
self.cleanup()
def__del__(self):
self.cleanup()

class_TemporaryFileWrapper:
"""Temporary file wrapper This class provides a wrapper around files opened for temporary use. In particular, it seeks to automatically remove the file when it is no longer needed. """def__init__(self, file, name, delete=True, delete_on_close=True):
self.file=fileself.name=nameself._closer=_TemporaryFileCloser(file, name, delete,
delete_on_close)
def__getattr__(self, name):
# Attribute lookups are delegated to the underlying file# and cached for non-numeric results# (i.e. methods are cached, closed and friends are not)file=self.__dict__['file']
a=getattr(file, name)
ifhasattr(a, '__call__'):
func=a@_functools.wraps(func)deffunc_wrapper(*args, **kwargs):
returnfunc(*args, **kwargs)
# Avoid closing the file as long as the wrapper is alive,# see issue #18879.func_wrapper._closer=self._closera=func_wrapperifnotisinstance(a, int):
setattr(self, name, a)
returna# The underlying __enter__ method returns the wrong object# (self.file) so override it to return the wrapperdef__enter__(self):
self.file.__enter__()
returnself# Need to trap __exit__ as well to ensure the file gets# deleted when used in a with statementdef__exit__(self, exc, value, tb):
result=self.file.__exit__(exc, value, tb)
self._closer.cleanup()
returnresultdefclose(self):
""" Close the temporary file, possibly deleting it. """self._closer.close()
# iter() doesn't use __getattr__ to find the __iter__ methoddef__iter__(self):
# Don't return iter(self.file), but yield from it to avoid closing# file as long as it's being used as iterator (see issue #23700). We# can't use 'yield from' here because iter(file) returns the file# object itself, which has a close method, and thus the file would get# closed when the generator is finalized, due to PEP380 semantics.forlineinself.file:
yieldline

This passes all tests for me in both Linux and Windows.

[eryksun]

As I mentioned previously, the exception handler in NamedTemporaryFile() also needs to be changed to check delete_on_close:

except:
ifnameisnotNoneandnot (
_os.name=='nt'anddeleteanddelete_on_close):
_os.unlink(name)
raise

[Ev2geny]

@eryksun , can you please have a look at this discussion about the fact, that the directory entry is not unlinked on Unix. I think you a correct person to have an idea there.

[Ev2geny]

@zooba , I have done my best to implement your comments and also got a help of eryksun.
Can you look at it again please?

[zooba]

Thanks for all your hard work on this! I tweaked some of the docs slightly, and will merge once CI finishes again.