strif

Strif is a tiny (~1000 loc) library of ~30 string, file, and object utilities for modern Python. It has zero dependencies.

It is simply a few functions and tricks that have repeatedly shown value in various projects. The goal is not to give a comprehensive suite of utilities but simply to complement the standard libraries and fill in a few gaps.

Tip

✨ NEW: Version 3.1 adds atomic_write_text()/atomic_write_bytes(), exposes __version__, and supports Python 3.10-3.14. ✨

If you're using strif, you might also want to check out prettyfmt, another small library built on strif that has some extra functions for pretty, human-readable outputs for objects, sizes, times and dates, etc.

Tip

Given the rapid rise of supply chain attacks on Python packages, it's worth noting this repository has ZERO dependencies. You can even have your agent review its security before you use it.

Why Use strif (Especially with Coding Agents)

Strif is a simple way to get better Python out of coding agents. It's small enough that once you tell an agent to use it, it adopts it consistently—so you can enforce good habits like atomic file writes, sortable timestamped ids, and stable hashing instead of hoping the agent reinvents them correctly each time. A prompt as simple as this is usually enough:

Run uv add strif and review its docs, then always use it for atomic file writes, timestamps and ids, and similar best practices from now on.

The whole library (all zero-dependency):

• Atomic file operations with parent-dir creation and optional backups, so an interrupted or crashed run never leaves a partial or corrupt file in its final location. See atomic_output_file(), atomic_write_text(), atomic_write_bytes(), copyfile_atomic().

• Random and timestamped ids in base 36 (concise, case-insensitive, filename-safe), including ISO-timestamped ids that sort by creation time. See new_uid(), new_timestamped_uid().

• File and string hashing with convenient hex, base36, and base64 outputs—good for cache keys and dedup. See hash_string(), hash_file(), file_mtime_hash().

• An AtomicVar type: a thread-safe variable (an RLock around any value) that makes correct concurrent access the easy path.

• String utilities for replacing or inserting multiple substrings at once and for validated, type-checked string templates. See StringTemplate, replace_multiple(), insert_multiple().

• Abbreviate and quote strings for clean logging and display. See abbrev_str(), single_line(), quote_if_needed().

That's all! They are all quite simple and small, so see the pydoc strings or code for full docs.

Installation

# Use uv
uv add strif
# Or poetry
poetry add strif
# Or pip
pip install strif

Text Abbreviations and Formatting

• abbrev_str(string: str, max_len: Optional[int] = 80, indicator: str = '…')

  Abbreviates a string and appends an indicator if the content exceeds the allowed length.

• abbrev_list(items: List[Any], max_items: int = 10, item_max_len: Optional[int] = 40, joiner: str = ', ', indicator: str = '…')

  Shortens each element of a list and appends an ellipsis if the list is truncated.

• single_line(text: str)

  Converts multi-line text into a single line by replacing extra whitespace with spaces.

• quote_if_needed(arg: Any)

  Returns a string with quotes if needed for proper display (for example, for filenames with spaces).

String Identifiers, Timestamps, and Hashing

Tip

These functions use base 36. If you need a readable, concise identifier, api key format, or hash format, consider base 36. In my humble opinion, base 36 ids are underrated and should be used more often:

• Base 36 is briefer than hex and yet avoids ugly non-alphanumeric characters.

• Base 36 is case insensitive. If you use identifiers for filenames, you definitely should prefer case insensitive identifiers because of case-insensitive filesystems (like macOS).

• Base 36 is easier to read aloud over the phone for an auth code or to type manually.

• Base 36 is only log(64)/log(36) - 1 = 16% longer than base 64.

• new_uid(bits: int = 64)

  Generates a random base36 alphanumeric string with at least the specified bits of randomness. Suitable for filenames (especially on case-insensitive filesystems). Uses random.SystemRandom() for randomness.

• new_timestamped_uid(bits: int = 32)

  Creates a unique ID starting with an ISO timestamp, then fractions of seconds and bits of randomness. Example: 20150912T084555Z-378465-43vtwbx

• iso_timestamp(microseconds: bool = True)

  Returns an ISO 8601 timestamp in UTC, e.g. 2015-09-12T08:41:12.397217Z (with microseconds) or 2015-09-12T08:41:12Z (without).

• format_iso_timestamp(datetime_obj: datetime, microseconds: bool = True)

  Formats a given datetime object as an ISO 8601 timestamp, ensuring UTC formatting with a trailing Z.

• clean_alphanum(string: str, max_length: Optional[int] = None)

  Converts a string to a clean identifier by keeping only the first alphanumeric characters and replacing others with underscores.

• clean_alphanum_hash(string: str, max_length: int = 64, max_hash_len: Optional[int] = None)

  Combines the cleaned version of a string with a base36 SHA1 hash to minimize collisions.

File Hashing

• file_mtime_hash(path: str | Path)

  Computes a fast hash using a file's name, size, and high-resolution modification time, without looking at file contents. A useful key for fast caching of file contents.

• hash_string(string: str, algorithm: str = 'sha1') -> Hash and hash_file(file_path: str | Path, algorithm: str = 'sha1') -> Hash

  Provide flexible hashing mechanisms. The returned Hash object has properties to output the digest in hexadecimal, base36, or with a prefixed algorithm name.

Atomic File Operations with Optional Backups

Tip

It’s generally good practice when creating files to write to a file with a temporary name, and move it to a final location once the file is complete. This way, you never leave partial, incorrect versions of files in a directory due to interruptions or failures.

• atomic_output_file(dest_path: str | Path, make_parents: bool = False, backup_suffix: Optional[str] = None, tmp_suffix: str = '.partial')

  A context manager for writing files or directories atomically. A temporary file is created and, upon successful completion, renamed to the target location.

• copyfile_atomic(source_path: str | Path, dest_path: str | Path, make_parents: bool = False, backup_suffix: Optional[str] = None)

  Atomically copies a file while preserving its timestamps.

• copytree_atomic(source_path: str | Path, dest_path: str | Path, make_parents: bool = False, backup_suffix: Optional[str] = None, symlinks: bool = False)

  Recursively copies a directory or file atomically.

• move_to_backup(path: str | Path, backup_suffix: str = '{timestamp}.bak') and copy_to_backup(path: str | Path, backup_suffix: str = '{timestamp}.bak')

  Functions to move or copy an existing file or directory to a backup destination.

• move_file(src_path: Path, dest_path: Path, keep_backup: bool = True, backup_suffix: str = '{timestamp}.bak')

  Moves a file to a new location, automatically creating parent directories and optionally keeping a backup of the destination if it already exists.

• atomic_write_text(dest_path, text, make_parents=False, backup_suffix=None, encoding='utf-8') and atomic_write_bytes(dest_path, data, make_parents=False, backup_suffix=None)

  Convenience wrappers around atomic_output_file() for the common case of writing a whole string or bytes value atomically in a single call.

For example, it is generally a good idea to wrap an open() call with atomic_output_file():

with atomic_output_file("some-dir/my-final-output.txt") as temp_target:
    with open(temp_target, "w") as f:
        f.write("some contents")

Or, for the common whole-value case, just:

atomic_write_text("some-dir/my-final-output.txt", "some contents")

And this can (and in most cases should) be used in place of shutil.copyfile:

copyfile_atomic(source_path, dest_path, make_parents=True, backup_suffix=None)

Now if there is some issue during write, the output will instead be at a temporary location in the same directory (with a name like some-dir/my-final-output.txt.partial.XXXXX.) This ensures integrity of the file appearing in the final location.

There are also some handy additional options:

with atomic_output_file("some-dir/my-final-output.txt",
                        make_parents=True, backup_suffix=".old.{timestamp}") as temp_target:
    with open(temp_target, "w") as f:
        f.write("some contents")

This creates parent folders as needed (a major convenience). And if you would have clobbered a previous output, it keeps a backup with a (fixed or uniquely timestamped) suffix.

Used judiciously, these options can save boilerplate coding and avoid debugging ugly corner case failures with zero-length or truncated files.

Syntax Sugar for Temporary Files

Syntax sugar for auto-deleting temporary files or directories using with:

with temp_output_file("my-scratch.") as (fd, path):
    # Do a bunch of stuff with the opened file descriptor or path, knowing
    # it will be removed assuming successful termination.


with temp_output_dir("work-dir.", dir="/var/tmp") as work_dir:
    # Create some files in the now-existing path work_dir, and it will be
    # deleted afterwards.

Note these don’t delete files in case of error, which is usually what you want. Add always_clean=True if you want the temporary file or directory to be removed no matter what.

Atomic Vars

AtomicVar is a simple zero-dependency thread-safe variable that works for any type. It simply combines a value with reentrant lock (threading.RLock) to make thread-safe use of the variable less error prone.

Often the standard "Pythonic" approach is to use locks directly, but for some common use cases, AtomicVar may be simpler and more readable. Works on any type, including lists and dicts.

Other options include threading.Event (for shared booleans), threading.Queue (for producer-consumer queues), and multiprocessing.Value (for process-safe primitives).

Examples:

# Immutable types are always safe:
count = AtomicVar(0)
count.update(lambda x: x + 5)  # In any thread.
count.set(0)  # In any thread.
current_count = count.value  # In any thread.

# Useful for flags:
global_flag = AtomicVar(False)
global_flag.set(True)  # In any thread.
if global_flag:  # In any thread.
    print("Flag is set")


# For mutable types,consider using `copy` or `deepcopy` to access the value:
my_list = AtomicVar([1, 2, 3])
my_list_copy = my_list.copy()  # In any thread.
my_list_deepcopy = my_list.deepcopy()  # In any thread.

# For mutable types, the `updates()` context manager gives a simple way to
# lock on updates:
with my_list.updates() as value:
    value.append(5)

# Or if you prefer, via a function:
my_list.update(lambda x: x.append(4))  # In any thread.

# You can also use the var's lock directly. In particular, this encapsulates
# locked one-time initialization:
initialized = AtomicVar(False)
with initialized.lock:
    if not initialized:  # checks truthiness of underlying value
        expensive_setup()
        initialized.set(True)

# Or:
lazy_var: AtomicVar[list[str] | None] = AtomicVar(None)
with lazy_var.lock:
    if not lazy_var:
            lazy_var.set(expensive_calculation())

Simple String Template

A validated template string that supports only specified fields. Can subclass to have a type with a given set of allowed_fields. Provide a type with a field name to allow validation of int/float format strings.

Examples:

>>> t = StringTemplate("{name} is {age} years old", ["name", "age"])
>>> t.format(name="Alice", age=30)
'Alice is 30 years old'

>>> t = StringTemplate("{count:3d}@{price:.2f}", [("count", int), ("price", float)])
>>> t.format(count=10, price=19.99)
' 10@19.99'

Multiple String Replacements

Insertion and Replacement are NamedTuples, so you can use named fields (Insertion(offset, text), Replacement(start, end, text)) or plain positional tuples.

• insert_multiple(text: str, insertions: list[Insertion]) -> str

  Insert multiple strings into text at the given offsets, at once.

• replace_multiple(text: str, replacements: list[Replacement]) -> str

  Replace multiple substrings in text with new strings, simultaneously. Each Replacement is (start_offset, end_offset, new_string).

FAQ

Why bother, if it’s so short?

Because it saves time, saves you stupid bugs and clumsy repetition, and has zero (yes zero) dependencies.

Are there other libraries that offer these utilities?

A few yes. Some support has improved in Python 3; for example textwrap.shorten() can be used instead of abbrev_str() (that said, strif also offers abbrev_list()).

boltons is a much larger library of general utilities. strif is intended to be much smaller. The atomicwrites library is similar to atomic_output_file() but is no longer maintained. For some others like the base36 tools I haven't seen equivalents elsewhere.

If you don't want the dependency on strif, also feel free to just copy the bit you want! They're short.

Is it mature?

I’ve used many of these functions in production situations for years. We don't have comprehensive tests currently. But they're mostly so small you can inspect them yourself.

---

Development

For how to install uv and Python, see installation.md.

For development workflows, see development.md.

For instructions on publishing to PyPI, see publishing.md.

---

This project was built from simple-modern-uv.