cs.fileutils 20240316

My grab bag of convenience functions for files and filenames/pathnames.

My grab bag of convenience functions for files and filenames/pathnames.

Latest release 20240316: Fixed release upload artifacts.

Function abspath_from_file(path, from_file)

Return the absolute path of path with respect to from_file, as one might do for an include file.

Function atomic_filename(filename, exists_ok=False, placeholder=False, dir=None, prefix=None, suffix=None, rename_func=<built-in function rename>, **kw)

A context manager to create filename atomicly on completion. This returns a NamedTemporaryFile to use to create the file contents. On completion the temporary file is renamed to the target name filename.

Parameters:

• filename: the file name to create
• exists_ok: default False; if true it not an error if filename already exists
• placeholder: create a placeholder file at filename while the real contents are written to the temporary file
• dir: passed to NamedTemporaryFile, specifies the directory to hold the temporary file; the default is dirname(filename) to ensure the rename is atomic
• prefix: passed to NamedTemporaryFile, specifies a prefix for the temporary file; the default is a dot ('.') plus the prefix from splitext(basename(filename))
• suffix: passed to NamedTemporaryFile, specifies a suffix for the temporary file; the default is the extension obtained from splitext(basename(filename))
• rename_func: a callable accepting (tempname,filename) used to rename the temporary file to the final name; the default is os.rename and this parametr exists to accept something such as FSTags.move Other keyword arguments are passed to the NamedTemporaryFile constructor.

Example:

>>> import os
>>> from os.path import exists as existspath
>>> fn = 'test_atomic_filename'
>>> with atomic_filename(fn, mode='w') as f:
...     assert not existspath(fn)
...     print('foo', file=f)
...     assert not existspath(fn)
...
>>> assert existspath(fn)
>>> assert open(fn).read() == 'foo\n'
>>> os.remove(fn)

Class BackedFile(ReadMixin)

A RawIOBase duck type which uses a backing file for initial data and writes new data to a front scratch file.

Method BackedFile.__init__(self, back_file, dirpath=None): Initialise the BackedFile using back_file for the backing data.

Class BackedFile_TestMethods

Mixin for testing subclasses of BackedFile. Tests self.backed_fp.

Function byteses_as_fd(bss, **kw)

Deliver the iterable of bytes bss as a readable file descriptor. Return the file descriptor. Any keyword arguments are passed to CornuCopyBuffer.as_fd.

  Example:

       # present a passphrase for use as in input file descrptor
       # for a subprocess
       rfd = byteses_as_fd([(passphrase + '

').encode()])

Function common_path_prefix(*paths)

Return the common path prefix of the paths.

Note that the common prefix of '/a/b/c1' and '/a/b/c2' is '/a/b/', not '/a/b/c'.

Callers may find it useful to preadjust the supplied paths with normpath, abspath or realpath from os.path; see the os.path documentation for the various caveats which go with those functions.

Examples:

>>> # the obvious
>>> common_path_prefix('', '')
''
>>> common_path_prefix('/', '/')
'/'
>>> common_path_prefix('a', 'a')
'a'
>>> common_path_prefix('a', 'b')
''
>>> # nonempty directory path prefixes end in os.sep
>>> common_path_prefix('/', '/a')
'/'
>>> # identical paths include the final basename
>>> common_path_prefix('p/a', 'p/a')
'p/a'
>>> # the comparison does not normalise paths
>>> common_path_prefix('p//a', 'p//a')
'p//a'
>>> common_path_prefix('p//a', 'p//b')
'p//'
>>> common_path_prefix('p//a', 'p/a')
'p/'
>>> common_path_prefix('p/a', 'p/b')
'p/'
>>> # the comparison strips complete unequal path components
>>> common_path_prefix('p/a1', 'p/a2')
'p/'
>>> common_path_prefix('p/a/b1', 'p/a/b2')
'p/a/'
>>> # contrast with cs.lex.common_prefix
>>> common_prefix('abc/def', 'abc/def1')
'abc/def'
>>> common_path_prefix('abc/def', 'abc/def1')
'abc/'
>>> common_prefix('abc/def', 'abc/def1', 'abc/def2')
'abc/def'
>>> common_path_prefix('abc/def', 'abc/def1', 'abc/def2')
'abc/'

Function compare(f1, f2, mode='rb')

Compare the contents of two file-like objects f1 and f2 for equality.

If f1 or f2 is a string, open the named file using mode (default: "rb").

Function copy_data(fpin, fpout, nbytes, rsize=None)

Copy nbytes of data from fpin to fpout, return the number of bytes copied.

Parameters:

• nbytes: number of bytes to copy. If None, copy until EOF.
• rsize: read size, default DEFAULT_READSIZE.

Function crop_name(name, ext=None, name_max=255)

Crop a file basename so as not to exceed name_max in length. Return the original name if it already short enough. Otherwise crop name before the file extension to make it short enough.

Parameters:

• name: the file basename to crop
• ext: optional file extension; the default is to infer the extension with os.path.splitext.
• name_max: optional maximum length, default: 255

Function datafrom(f, offset=None, readsize=None, maxlength=None)

General purpose reader for files yielding data from offset.

WARNING: this function might move the file pointer.

Parameters:

• f: the file from which to read data; if a string, the file is opened with mode="rb"; if an int, treated as an OS file descriptor; otherwise presumed to be a file-like object. If that object has a .fileno() method, treat that as an OS file descriptor and use it.
• offset: starting offset for the data
• maxlength: optional maximum amount of data to yield
• readsize: read size, default DEFAULT_READSIZE.

For file-like objects, the read1 method is used in preference to read if available. The file pointer is briefly moved during fetches.

Function datafrom_fd(fd, offset=None, readsize=None, aligned=True, maxlength=None)

General purpose reader for file descriptors yielding data from offset. Note: This does not move the file descriptor position if the file is seekable.

Parameters:

• fd: the file descriptor from which to read.
• offset: the offset from which to read. If omitted, use the current file descriptor position.
• readsize: the read size, default: DEFAULT_READSIZE
• aligned: if true (the default), the first read is sized to align the new offset with a multiple of readsize.
• maxlength: if specified yield no more than this many bytes of data.

Function file_based(*da, **dkw)

A decorator which caches a value obtained from a file.

In addition to all the keyword arguments for @cs.deco.cachedmethod, this decorator also accepts the following arguments:

• attr_name: the name for the associated attribute, used as the basis for the internal cache value attribute
• filename: the filename to monitor. Default from the ._{attr_name}__filename attribute. This value will be passed to the method as the filename keyword parameter.
• poll_delay: delay between file polls, default DEFAULT_POLL_INTERVAL.
• sig_func: signature function used to encapsulate the relevant information about the file; default cs.filestate.FileState({filename}).

If the decorated function raises OSError with errno == ENOENT, this returns None. Other exceptions are reraised.

Function file_data(fp, nbytes=None, rsize=None)

Read nbytes of data from fp and yield the chunks as read.

Parameters:

• nbytes: number of bytes to read; if None read until EOF.
• rsize: read size, default DEFAULT_READSIZE.

Function file_property(*da, **dkw)

A property whose value reloads if a file changes.

Function files_property(func)

A property whose value reloads if any of a list of files changes.

Note: this is just the default mode for make_files_property.

func accepts the file path and returns the new value. The underlying attribute name is '_'+func.__name__, the default from make_files_property(). The attribute {attr_name}_lock is a mutex controlling access to the property. The attributes {attr_name}_filestates and {attr_name}_paths track the associated file states. The attribute {attr_name}_lastpoll tracks the last poll time.

The decorated function is passed the current list of files and returns the new list of files and the associated value.

One example use would be a configuration file with recurive include operations; the inner function would parse the first file in the list, and the parse would accumulate this filename and those of any included files so that they can be monitored, triggering a fresh parse if one changes.

Example:

class C(object):
  def __init__(self):
    self._foo_path = '.foorc'
  @files_property
  def foo(self,paths):
    new_paths, result = parse(paths[0])
    return new_paths, result

The load function is called on the first access and on every access thereafter where an associated file's FileState has changed and the time since the last successful load exceeds the poll_rate (1s). An attempt at avoiding races is made by ignoring reloads that raise exceptions and ignoring reloads where files that were stat()ed during the change check have changed state after the load.

Function find(path, select=None, sort_names=True)

Walk a directory tree path yielding selected paths.

Note: not selecting a directory prunes all its descendants.

Function findup(path, test, first=False)

Test the pathname abspath(path) and each of its ancestors against the callable test, yielding paths satisfying the test.

If first is true (default False) this function always yields exactly one value, either the first path satisfying the test or None. This mode supports a use such as:

matched_path = next(findup(path, test, first=True))
# post condition: matched_path will be `None` on no match
# otherwise the first matching path

Function gzifopen(path, mode='r', *a, **kw)

Context manager to open a file which may be a plain file or a gzipped file.

If path ends with '.gz' then the filesystem paths attempted are path and path without the extension, otherwise the filesystem paths attempted are path+'.gz' and path. In this way a path ending in '.gz' indicates a preference for a gzipped file otherwise an uncompressed file.

However, if exactly one of the paths exists already then only that path will be used.

Note that the single character modes 'r', 'a', 'w' and 'x' are text mode for both uncompressed and gzipped opens, like the builtin open and unlike gzip.open. This is to ensure equivalent behaviour.

Function iter_fd(fd, **kw)

Iterate over data from the file descriptor fd.

Function iter_file(f, **kw)

Iterate over data from the file f.

Function lines_of(fp, partials=None)

Generator yielding lines from a file until EOF. Intended for file-like objects that lack a line iteration API.

Function lockfile(path, *, ext=None, poll_interval=None, timeout=None, runstate: cs.resources.RunState)

A context manager which takes and holds a lock file. An open file descriptor is kept for the lock file as well to aid locating the process holding the lock file using eg lsof.

Parameters:

• path: the base associated with the lock file.
• ext: the extension to the base used to construct the lock file name. Default: '.lock'
• timeout: maximum time to wait before failing. Default: None (wait forever).
• poll_interval: polling frequency when timeout is not 0.
• runstate: optional RunState duck instance supporting cancellation.

Function make_files_property(attr_name=None, unset_object=None, poll_rate=1.0)

Construct a decorator that watches multiple associated files.

Parameters:

• attr_name: the underlying attribute, default: '_'+func.__name__
• unset_object: the sentinel value for "uninitialised", default: None
• poll_rate: how often in seconds to poll the file for changes, default from DEFAULT_POLL_INTERVAL: 1.0

The attribute attr_name_lock controls access to the property. The attributes attr_name_filestates and attr_name_paths track the associated files' state. The attribute attr_name_lastpoll tracks the last poll time.

The decorated function is passed the current list of files and returns the new list of files and the associated value.

One example use would be a configuration file with recursive include operations; the inner function would parse the first file in the list, and the parse would accumulate this filename and those of any included files so that they can be monitored, triggering a fresh parse if one changes.

Example:

class C(object):
  def __init__(self):
    self._foo_path = '.foorc'
  @files_property
  def foo(self,paths):
    new_paths, result = parse(paths[0])
    return new_paths, result

The load function is called on the first access and on every access thereafter where an associated file's FileState has changed and the time since the last successful load exceeds the poll_rate.

An attempt at avoiding races is made by ignoring reloads that raise exceptions and ignoring reloads where files that were os.stat()ed during the change check have changed state after the load.

Function makelockfile(path, *, ext=None, poll_interval=None, timeout=None, runstate: cs.resources.RunState, keepopen=False)

Create a lockfile and return its path.

The lockfile can be removed with os.remove. This is the core functionality supporting the lockfile() context manager.

Parameters:

• path: the base associated with the lock file, often the filesystem object whose access is being managed.
• ext: the extension to the base used to construct the lockfile name. Default: ".lock"
• timeout: maximum time to wait before failing. Default: None (wait forever). Note that zero is an accepted value and requires the lock to succeed on the first attempt.
• poll_interval: polling frequency when timeout is not 0.
• runstate: optional RunState duck instance supporting cancellation. Note that if a cancelled RunState is provided no attempt will be made to make the lockfile.
• keepopen: optional flag, default False: if true, do not close the lockfile and return (lockpath,lockfd) being the lock file path and the open file descriptor

Function max_suffix(dirpath, prefix)

Compute the highest existing numeric suffix for names starting with prefix.

This is generally used as a starting point for picking a new numeric suffix.

Function mkdirn(path, sep='')

Create a new directory named path+sep+n, where n exceeds any name already present.

Parameters:

• path: the basic directory path.
• sep: a separator between path and n. Default: ''

Function NamedTemporaryCopy(f, progress=False, progress_label=None, **kw)

A context manager yielding a temporary copy of filename as returned by NamedTemporaryFile(**kw).

Parameters:

• f: the name of the file to copy, or an open binary file, or a CornuCopyBuffer
• progress: an optional progress indicator, default False; if a bool, show a progress bar for the copy phase if true; if an int, show a progress bar for the copy phase if the file size equals or exceeds the value; otherwise it should be a cs.progress.Progress instance
• progress_label: option progress bar label, only used if a progress bar is made Other keyword parameters are passed to tempfile.NamedTemporaryFile.

Class NullFile

Writable file that discards its input.

Note that this is not an open of os.devnull; it just discards writes and is not the underlying file descriptor.

Method NullFile.__init__(self): Initialise the file offset to 0.

Class Pathname(builtins.str)

Subclass of str presenting convenience properties useful for format strings related to file paths.

Function poll_file(path, old_state, reload_file, missing_ok=False)

Watch a file for modification by polling its state as obtained by FileState(). Call reload_file(path) if the state changes. Return (new_state,reload_file(path)) if the file was modified and was unchanged (stable state) before and after the reload_file(). Otherwise return (None,None).

This may raise an OSError if the path cannot be os.stat()ed and of course for any exceptions that occur calling reload_file.

If missing_ok is true then a failure to os.stat() which raises OSError with ENOENT will just return (None,None).

Function read_data(fp, nbytes, rsize=None)

Read nbytes of data from fp, return the data.

Parameters:

• nbytes: number of bytes to copy. If None, copy until EOF.
• rsize: read size, default DEFAULT_READSIZE.

Function read_from(fp, rsize=None, tail_mode=False, tail_delay=None)

Generator to present text or data from an open file until EOF.

Parameters:

• rsize: read size, default: DEFAULT_READSIZE
• tail_mode: if true, yield an empty chunk at EOF, allowing resumption if the file grows.

Class ReadMixin

Useful read methods to accomodate modes not necessarily available in a class.

Note that this mixin presumes that the attribute self._lock is a threading.RLock like context manager.

Classes using this mixin should consider overriding the default .datafrom method with something more efficient or direct.

Function rewrite(filepath, srcf, mode='w', backup_ext=None, do_rename=False, do_diff=None, empty_ok=False, overwrite_anyway=False)

Rewrite the file filepath with data from the file object srcf.

Parameters:

• filepath: the name of the file to rewrite.
• srcf: the source file containing the new content.
• mode: the write-mode for the file, default 'w' (for text); use 'wb' for binary data.
• empty_ok: if true (default False), do not raise ValueError if the new data are empty.
• overwrite_anyway: if true (default False), skip the content check and overwrite unconditionally.
• backup_ext: if a nonempty string, take a backup of the original at filepath + backup_ext.
• do_diff: if not None, call do_diff(filepath,tempfile).
• do_rename: if true (default False), rename the temp file to filepath after copying the permission bits. Otherwise (default), copy the tempfile to filepath; this preserves the file's inode and permissions etc.

Function rewrite_cmgr(filepath, mode='w', **kw)

Rewrite a file, presented as a context manager.

Parameters:

• mode: file write mode, defaulting to "w" for text.

Other keyword parameters are passed to rewrite().

Example:

with rewrite_cmgr(pathname, do_rename=True) as f:
    ... write new content to f ...

Class RWFileBlockCache

A scratch file for storing data.

Method RWFileBlockCache.__init__(self, pathname=None, dirpath=None, suffix=None, lock=None): Initialise the file.

Parameters:

• pathname: path of file. If None, create a new file with tempfile.mkstemp using dir=dirpath and unlink that file once opened.
• dirpath: location for the file if made by mkstemp as above.
• lock: an object to use as a mutex, allowing sharing with some outer system. A Lock will be allocated if omitted.

Function saferename(oldpath, newpath)

Rename a path using os.rename(), but raise an exception if the target path already exists. Note: slightly racey.

Function seekable(fp)

Try to test whether a filelike object is seekable.

First try the IOBase.seekable method, otherwise try getting a file descriptor from fp.fileno and os.stat()ing that, otherwise return False.

Class Tee

An object with .write, .flush and .close methods which copies data to multiple output files.

Method Tee.__init__(self, *fps): Initialise the Tee; any arguments are taken to be output file objects.

Function tee(fp, fp2)

Context manager duplicating .write and .flush from fp to fp2.

Function tmpdir()

Return the pathname of the default temporary directory for scratch data, the environment variable $TMPDIR or '/tmp'.

Function tmpdirn(tmp=None)

Make a new temporary directory with a numeric suffix.

Function trysaferename(oldpath, newpath)

A saferename() that returns True on success, False on failure.

Release Log

Release 20240316: Fixed release upload artifacts.

Release 20240201:

• makelockfile: new optional keepopen parameter - if true return the lock path and an open file descriptor.
• lockfile(): keep the lock file open to aid debugging with eg lsof.

Release 20231129:

• atomic_filename: accept optional rename_func to use instead of os.rename, supports using FSTags.move.
• atomic_filename: clean up the temp file.

Release 20230421: atomic_filename: raise FileExistsError instead of ValueError if not exists_ok and existspath(filename).

Release 20230401: Replaced a lot of runstate plumbing with @uses_runstate.

Release 20221118: atomic_filename: use shutil.copystat instead of shutil.copymode, bugfix the associated logic.

Release 20220429: Move longpath and shortpath to cs.fs, leave legacy names behind.

Release 20211208:

• Move NDJSON stuff to separate cs.ndjson module.
• New gzifopen() function to open either a gzipped file or an uncompressed file.

Release 20210906: Additional release because I'm unsure @atomic_filename made it into the previous release.

Release 20210731: New atomic_filename context manager wrapping NamedTemporaryFile for presenting a file after its contents are prepared.

Release 20210717: Updates for recent cs.mappings-20210717 release.

Release 20210420:

• Forensic prefix for NamedTemporaryCopy.
• UUIDNDJSONMapping: provide an empty .scan_errors on instantiation, avoids AttributeError if a scan never occurs.

Release 20210306:

• datafrom_fd: fix use-before-set of is_seekable.
• RWFileBlockCache.put: remove assert(len(data)>0), adjust logic.

Release 20210131: crop_name: put ext before name_max, more likely to be specified, I think.

Release 20201227.1: Docstring tweak.

Release 20201227: scan_ndjson: optional errors_list to accrue errors during the scan.

Release 20201108: Bugfix rewrite_cmgr, failed to flush a file before copying its contents.

Release 20201102:

• Newline delimited JSON (ndjson) support.
• New UUIDNDJSONMapping implementing a singleton cs.mappings.LoadableMappingMixin of cs.mappings.UUIDedDict subclass instances backed by an NDJSON file.
• New scan_ndjson() function to yield newline delimited JSON records.
• New write_ndjson() function to write newline delimited JSON records.
• New append_ndjson() function to append a single newline delimited JSON record to a file.
• New NamedTemporaryCopy for creating a temporary copy of a file with an optional progress bar.
• rewrite_cmgr: turn into a simple wrapper for rewrite.
• datafrom: make the offset parameter optional, tweak the @strable open function.
• datafrom_fd: support nonseekable file descriptors, document that for these the file position is moved (no pread support).
• New iter_fd and iter_file to return iterators of a file's data by utilising a CornuCopyBuffer.
• New byteses_as_fd to return a readable file descriptor receiving an iterable of bytes via a CornuCopyBuffer.

Release 20200914: New common_path_prefix to compare pathnames.

Release 20200517:

• New crop_name() function to crop a file basename to fit within a specific length.
• New find() function complimenting findup (UNTESTED).

Release 20200318: New findup(path,test) generator to walk up a file tree.

Release 20191006: Adjust import of cs.deco.cachedmethod.

Release 20190729: datafrom_fd: make offset optional, defaulting to fd position at call.

Release 20190617: @file_based: adjust use of @cached from cached(wrap0, **dkw) to cached(**dkw)(wrap0).

Release 20190101: datafrom: add maxlength keyword arg, bugfix fd and f.fileno cases.

Release 20181109:

• Various bugfixes for BackedFile.
• Use a file's .read1 method if available in some scenarios.
• makelockfile: accept am optional RunState control parameter, improve some behaviour.
• datafrom_fd: new optional maxlength parameter limiting the amount of data returned.
• datafrom_fd: by default, perform an initial read to align all subsequent reads with the readsize.
• drop fdreader, add datafrom(f, offset, readsize) accepting a file or a file descriptor, expose datafrom_fd.
• ReadMixin.datafrom now mandatory. Add ReadMixin.bufferfrom.
• Assorted other improvements, minor bugfixes, documentation improvements.

Release 20171231.1: Trite DISTINFO fix, no semantic changes.

Release 20171231: Update imports, bump DEFAULT_READSIZE from 8KiB to 128KiB.

Release 20170608:

• Move lockfile and the SharedAppend* classes to cs.sharedfile.
• BackedFile internal changes.

Release 20160918:

• BackedFile: redo implementation of .front_file to fix resource leak; add .len; add methods .spans, .front_spans and .back_spans to return information about front vs back data.
• seek: bugfix: seek should return the new file offset.
• BackedFile does not subclass RawIOBase, it just works like one.

Release 20160828:

• Use "install_requires" instead of "requires" in DISTINFO.
• Rename maxFilenameSuffix to max_suffix.
• Pull in OpenSocket file-like socket wrapper from cs.venti.tcp.
• Update for cs.asynchron changes.
• ... then move cs.fileutils.OpenSocket into new module cs.socketutils.
• New Tee class, for copying output to multiple files.
• NullFile class which discards writes (==> no-op for Tee).
• New class SavingFile to accrue output and move to specified pathname when complete.
• Memory usage improvements.
• Polyfill non-threadsafe implementation of pread if os.pread does not exist.
• New function seekable() to probe a file for seekability.
• SharedAppendFile: provide new .open(filemode) context manager for allowing direct file output for external users.
• New function makelockfile() presenting the logic to create a lock file separately from the lockfile context manager.
• Assorted bugfixes and improvements.

Release 20150116: Initial PyPI release.