04 Mar 2020

List outdated packages using poetry

History / Edit / PDF / EPUB / BIB / 1 min read (~145 words)

I use poetry as my python package manager and I'd like to know the packages that I depend on that are currently outdated.

An easy way to get this list is to run poetry show --outdated. This will return you a list of all the packages that are outdated, their current version, the latest version, as well as a description of the package.

There are in my opinion three missing features here:

  • having the command respect the semantic versioning constraint and only letting you see the latest version according to those constraints
  • having a flag to switch between showing the latest version available without semantic versioning constraint vs the latest version constrained by semantic versioning
  • having a flag to list only the packages that are direct dependencies (listed in the pyproject.toml file)

03 Mar 2020

Mypy and implicit namespaces

History / Edit / PDF / EPUB / BIB / 1 min read (~78 words)

I use mypy but it doesn't seem to scan all my files. Why?

You might be using implicit namespaces in your code (see PEP 420). Support for implicit namespaces in mypy is rather flaky as of 2020-03-03.

One solution for the moment is to add __init__.py and make all your namespaces explicit.

Another solution is to replace your calls to mypy some-path with mypy $(find some-path -name "*.py").

29 Feb 2020

Slow python imports

History / Edit / PDF / EPUB / BIB / 4 min read (~644 words)

Some imports in my python code are slow. How can I figure out which ones are the source of slowness?

Python offers a really useful functionality you can use that will list how long each import took. By passing the -X importtime argument to your python command when you execute your script it will print out both the cumulative time (including nested imports) and self time (excluding nested imports) of each import.

python -X importtime your-script.py

Running python -X importtime my-script.py on an empty script returns the following (on Windows 7, Python 3.7.5)

import time: self [us] | cumulative | imported package
import time:        52 |         52 | zipimport
import time:       367 |        367 | _frozen_importlib_external
import time:        55 |         55 |     _codecs
import time:       530 |        585 |   codecs
import time:       520 |        520 |   encodings.aliases
import time:      1107 |       2210 | encodings
import time:       328 |        328 | encodings.utf_8
import time:        39 |         39 | _signal
import time:       357 |        357 | encodings.latin_1
import time:        34 |         34 |     _abc
import time:       312 |        345 |   abc
import time:       474 |        819 | io
import time:       113 |        113 |       _stat
import time:       264 |        377 |     stat
import time:       269 |        269 |       genericpath
import time:       794 |       1062 |     ntpath
import time:       871 |        871 |     _collections_abc
import time:       915 |       3223 |   os
import time:       490 |        490 |   _sitebuiltins
import time:        57 |         57 |     _locale
import time:       934 |        991 |   _bootlocale
import time:       421 |        421 |   encodings.cp1252
import time:       472 |        472 |   types
import time:       394 |        394 |       warnings
import time:       440 |        834 |     importlib
import time:       263 |        263 |       importlib.machinery
import time:       554 |        816 |     importlib.abc
import time:        64 |         64 |           _operator
import time:       792 |        856 |         operator
import time:       343 |        343 |         keyword
import time:        43 |         43 |           _heapq
import time:       405 |        447 |         heapq
import time:        85 |         85 |         itertools
import time:       328 |        328 |         reprlib
import time:        69 |         69 |         _collections
import time:      1460 |       3585 |       collections
import time:        44 |         44 |         _functools
import time:       596 |        640 |       functools
import time:       784 |       5008 |     contextlib
import time:       651 |       7308 |   importlib.util
import time:      1095 |       1095 |   pywin32_bootstrap
import time:       231 |        231 |   sitecustomize
import time:     10744 |      24972 | site

For a script with a simple import argparse, I get the following output:

import time: self [us] | cumulative | imported package
import time:        70 |         70 | zipimport
import time:       341 |        341 | _frozen_importlib_external
import time:        54 |         54 |     _codecs
import time:       457 |        511 |   codecs
import time:       456 |        456 |   encodings.aliases
import time:      1030 |       1997 | encodings
import time:       215 |        215 | encodings.utf_8
import time:        38 |         38 | _signal
import time:       268 |        268 | encodings.latin_1
import time:        33 |         33 |     _abc
import time:       398 |        431 |   abc
import time:       311 |        741 | io
import time:        87 |         87 |       _stat
import time:       271 |        357 |     stat
import time:       196 |        196 |       genericpath
import time:       416 |        612 |     ntpath
import time:       714 |        714 |     _collections_abc
import time:       610 |       2292 |   os
import time:       229 |        229 |   _sitebuiltins
import time:        48 |         48 |     _locale
import time:       246 |        293 |   _bootlocale
import time:       217 |        217 |   encodings.cp1252
import time:       488 |        488 |   types
import time:       279 |        279 |       warnings
import time:       461 |        740 |     importlib
import time:       269 |        269 |       importlib.machinery
import time:       557 |        825 |     importlib.abc
import time:        63 |         63 |           _operator
import time:       808 |        871 |         operator
import time:       336 |        336 |         keyword
import time:        41 |         41 |           _heapq
import time:       336 |        376 |         heapq
import time:        69 |         69 |         itertools
import time:       341 |        341 |         reprlib
import time:        70 |         70 |         _collections
import time:      1136 |       3197 |       collections
import time:        69 |         69 |         _functools
import time:       642 |        710 |       functools
import time:       801 |       4708 |     contextlib
import time:       688 |       6959 |   importlib.util
import time:       934 |        934 |   pywin32_bootstrap
import time:       224 |        224 |   sitecustomize
import time:      9323 |      20954 | site
import time:       698 |        698 |     enum
import time:        55 |         55 |       _sre
import time:       417 |        417 |         sre_constants
import time:       372 |        789 |       sre_parse
import time:       443 |       1286 |     sre_compile
import time:       323 |        323 |     copyreg
import time:       716 |       3021 |   re
import time:       725 |        725 |     locale
import time:       948 |       1673 |   gettext
import time:       986 |       5678 | argparse

The package are listed in order that they are resolved. In argparse case, os and sys were already loaded, so it first loads re, then gettext. Once both are loaded, argparse has finished loading.

The way the cumulative column is computed is to take all the prior self that are a level higher than the package you're looking at. For example (if we take the io package):

import time:        33 |         33 |     _abc
import time:       398 |        431 |   abc
import time:       311 |        741 | io

311 + 398 + 33 = 742

We can see here that the numbers are not necessarily equal to one another, this might be due to precision used to do the computation while the rendering of numbers is rounded.

Note that the load time of a package may be different depending on which script you load because dependendencies of the package may have already been loaded in some cases, while in others it may have to load them.

Looking at text might be your thing, but if you're more visual, there's a tool called tuna which will consume this output and create an icicle plot you can look at to find which imports are the slowest/longest.

29 Feb 2020

Improving the performance of a slow click CLI

History / Edit / PDF / EPUB / BIB / 2 min read (~372 words)

My click CLI is slow, even just to show the help. How do I make it go faster?

In most cases, the reason your click CLI is slow is that you have large imports at the top of the files where you have declared your commands.

The typical pattern is as follows:

cli.py

from train import train
from predict import predict

@click.group()
def cli():
    pass

cli.add_command(predict)
cli.add_command(train)

train.py

import click
import pandas as pd
import torch

@click.command()
def train():
    pass

predict.py

import click
import pandas as pd
import torch

@click.command()
def predict():
    pass

Notice that in both these files we import pandas and torch, which can account for a large chunk of script execution time simply due to importing them. You can verify that by simply running python -X importtime train.py 2>tuna.log and using tuna (run tuna tuna.log) to inspect the results and convince yourself.

The suggested pattern is to move the imports inside of the function itself, as such:

train.py

import click

@click.command()
def train():
    import pandas as pd
    import torch

    pass

predict.py

import click

@click.command()
def predict():
    import pandas as pd
    import torch

    pass

This will shave off a large amount of time spent importing those packages (pandas and torch). They will only be loaded when you need to run the command itself, not every time you invoke the CLI.

Another pattern which is more complicated is to move the logic of the functions in separate files. This is done to avoid the common mistake that will happen over time that developers will add more logic in those command files, adding imports at the top of the file and slowing the CLI again. By moving the complete implementation to a separate file, you can have the imports at the top of the file and it is not possible to make this mistake again.

train.py

import click

@click.command()
def train():
    from train_implementation import train
    train()

train_implementation.py

import pandas as pd
import torch

def train():
    # Implementation is now here
    pass
28 Feb 2020

Pytest with tests files with similar names

History / Edit / PDF / EPUB / BIB / 1 min read (~183 words)

I have two test files with the same name and pytest complains. How do I make it work without changing the test filenames?

Example directory structure

/path/to/project/tests
├── a/
│   └── test_a.py
└── b/
    └── test_a.py

Error message:

import file mismatch:
imported module 'test_a' has this __file__ attribute:
  /path/to/project/tests/a/test_a.py
which is not the same as the test file we want to collect:
  /path/to/project/tests/b/test_a.py
HINT: remove __pycache__ / .pyc files and/or use a unique basename for your test file modules

Add a __init__.py to each directories with tests files that have the same name. Technically, you only need to have a __init__.py file in one of the two directories, so that one is in a package while the other one is in a different one. Adding it in both simply prevents this issue from occurring again if you were to add a third file test_a.py.

/path/to/project/tests
├── a/
│   ├── __init__.py
│   └── test_a.py
└── b/
    ├── __init__.py
    └── test_a.py