a powerful technique for isolating tests from undesired interactions among
components. But often people find their mock isn’t taking effect, and it’s not
clear why. Hopefully this explanation will clear things up.
BTW: it’s really easy to over-use mocking. These are good explanations of
Before we get to fancy stuff like mocks, I want to review a little bit about
Python assignment. You may already know this, but bear with me. Everything
that follows is going to be directly related to this simple example.
Variables in Python are names that refer to values. If we assign a second
name, the names don’t refer to each other, they both refer to the same value.
If one of the names is then assigned again, the other name isn’t affected:
If this is unfamiliar to you, or you just want to look at more pictures like
this, Python Names and Values goes into much more depth about the semantics
of Python assignment.
Let’s say we have a simple module like this:
defupdate_val(): globalval val="updated"
We want to use val from this module, and also call update_val to change val.
There are two ways we could try to do it. At first glance, it seems like they
would do the same thing.
The first version imports the names we want, and uses them:
print(val) update_val() print(val)
The second version imports the module, and uses the names as attributes on
the module object:
print(mod.val) mod.update_val() print(mod.val)
This seems like a subtle distinction, almost a stylistic choice. But
code1.py prints “original original”: the value hasn’t changed! Code2.py does
what we expected: it prints “original updated.” Why the difference?
Let’s look at code1.py more closely:
print(val) update_val() print(val)
After “from mod import val”, when we first print val, we have this:
“from mod import val” means, import mod, and then do the assignment “val =
mod.val”. This makes our name val refer to the same object as mod’s name
After “update_val()”, when we print val again, our world looks like this:
update_val has reassigned mod’s val, but that has no effect on our val. This
is the same behavior as our x and y example, but with imports instead of more
obvious assignments. In code1.py, “from mod import val” is an assignment from
mod.val to val, and works exactly like “y = x” does. Later assignments to
mod.val don’t affect our val, just as later assignments to x don’t affect y.
Now let’s look at code2.py again:
print(mod.val) mod.update_val() print(mod.val)
The “import mod” statement means, make my name mod refer to the entire mod
module. Accessing mod.val will reach into the mod module, find its val name, and
use its value.
Then after “update_val()”, mod’s name val has been changed:
Now we print mod.val again, and see its updated value, just as we expected.
OK, but what about mocks?
Mocking is a fancy kind of assignment: replace an object (or function) with
a different one. We’ll use the
function in a with statement. It makes a mock object, assigns it to the name
given, and then restores the original value at the end of the with
Let’s consider this (very roughly sketched) product code and test:
defmy_function(): files=listdir(some_directory) # ... use the file names ...
When the test is run, the references look like this:
Because the product code refers to the os module, changing the name in the
module is enough to affect the product code.
But there’s still a problem: this will mock that function for any module
using it. This might be a more widespread effect than you intended. Perhaps
your product code also calls some helpers, which also need to list files. The
helpers might end up using your mock (depending how they imported os.listdir!),
which isn’t what you wanted.
Mock it where it’s used
The best approach to mocking is to mock the object where it is used, not
where it is defined. Your product and test code will look like this:
defmy_function(): files=listdir(some_directory) # ... use the file names ...
The only difference here from our first try is that we mock “product.listdir”,
not “os.listdir”. That seems odd, because listdir isn’t defined in
product.py. That’s fine, the name “listdir” is in both the os module and in
product.py, and they are both references to the thing you want to mock. Neither
is a more real name than the other.
By mocking where the object is used, we have tighter control over what
callers are affected. Since we only want product.py’s behavior to change, we
mock the name in product.py. This also makes the test more clearly tied to
As before, our references look like this once product.py has been fully
The difference now is how the mock changes things. During the test, our references look like this:
The code in product.py will use the mock, and no other code will. Just what
Is this OK?
At this point, you might be concerned: it seems like mocking is kind of
delicate. Notice that even with our last example, how we create the mock
depends on something as arbitrary as how we imported the function. If our code
had “import os” at the top, we wouldn’t have been able to create our mock
properly. This is something that could be changed in a refactoring, but at
least mock.patch will fail in that case.
You are right to be concerned: mocking is delicate. It depends on
implementation details of the product code to construct the test. There are
many reasons to be wary of mocks, and there are other approaches to solving the
problems of isolating your product code from problematic dependencies.
If you do use mocks, at least now you know how to make them work, but again,
there are other approaches. See the links at the top of this page.
A good practice when writing complicated software is to put in lots of
debugging code. This might be extra logging, or special modes that tweak the
behavior to be more understandable, or switches to turn off some aspect of your
test suite so you can focus on the part you care about at the moment.
But how do you control that debugging code? Where are the on/off switches?
You don’t want to clutter your real UI with controls. A convenient option is
environment variables: you can access them simply in the code, your shell has
ways to turn them on and off at a variety of scopes, and they are invisible to
Though if they are invisible to your users, they are also invisible to you!
How do you remember what exotic options you’ve coded into your program, and how
do you easily see what is set, and change what is set?
I’ve been using environment variables like this in coverage.py for years, but
only recently made it easier to work with them.
To do that, I wrote set_env.py.
It scans a tree of files for special comments describing environment variables,
then shows you the values of those variables. You can type quick commands to
change the values, and when the program is done, it updates your environment.
It’s not a masterpiece of engineering, but it works for me.
As an example, this line appears in coverage.py:
# $set_env.py: COVERAGE_NO_PYTRACER - Don't run the tests under the Python tracer.
This line is found by set_env.py, so it knows that COVERAGE_NO_PYTRACER is
one of the environment variables it should fiddle with.
When I run set_env.py in the coverage.py tree, I get something like this:
$ set_env Read 298 files 1: COVERAGE_AST_DUMP Dump the AST nodes when parsing code. 2: COVERAGE_CONTEXT Set to 'test_function' for who-tests-what 3: COVERAGE_DEBUG Options for --debug. 4: COVERAGE_DEBUG_CALLS Lots and lots of output about calls to Coverage. 5: COVERAGE_ENV_ID Use environment-specific test directories. 6: COVERAGE_KEEP_TMP Keep the temp directories made by tests. 7: COVERAGE_NO_CONTRACTS Disable PyContracts to simplify stack traces. 8: COVERAGE_NO_CTRACER Don't run the tests under the C tracer. 9: COVERAGE_NO_PYTRACER = '1' Don't run the tests under the Python tracer. 10: COVERAGE_PROFILE Set to use ox_profile. 11: COVERAGE_TRACK_ARCS Trace every arc added while parsing code. 12: PYTEST_ADDOPTS Extra arguments to pytest.
(# [value] | x # ... | ? | q)>
All of the files were scanned, and 12 environment variables found. We can
see that COVERAGE_NO_PYTRACER has the value “1”, and none of the others are in
the environment. At the prompt, if I type “4”, then COVERAGE_DEBUG_CALLS (line
4) will be toggled to “1”. Type “4” again, and it is cleared. Typing “4 yes
please” will set it to “yes please”, but often I just need something or
nothing, so toggling “1” as the value works.
One bit of complexity here is that a program you run in your shell can’t
change environment variables for subsequent programs, which is exactly what we
need. So “set_env” is actually a shell alias:
aliasset_env='$(set_env.py $(git ls-files))'
This runs set_env.py against all of the files checked-in to git, and then
executes whatever set_env.py outputs. Naturally, set_env.py outputs shell
commands to set environment variables. If ls-files produces too much output,
you can use globs there also, so “**/*.py” might be useful.
Like I said, it’s not a masterpiece, but it works for me. If there are other
tools out there that do similar things, I’d like to hear about them.
I’ve released another alpha of coverage.py 5.0:
There are some design decisions ahead that I could use feedback on.
The big feature in 5.0 is “contexts”: recording data for varying execution
context, also known as Who Tests
What. The idea is to record not just that a line was executed, but also
which tests ran each line.
Some of the changes in alpha 6 were driven by a hackathon project at work:
using who-tests-what on the large Open edX codebase. We wanted to collect
context information, and then for each new pull request, run only the subset of
tests that touched the lines you changed. Initial experiments indicate this
could be a huge time-savings.
Big changes in this alpha:
Support for contexts when reporting. The --show-contexts option annotates
lines with the names of contexts recorded for the line. The --contexts option
lets you filter the report to only certain contexts. Big thanks to Stephan
Richter and Albertas Agejevas for the contribution.
Our largest test suite at work has 29k tests. The .coverage SQLite data
file was 659Mb, which was too large to work with. I changed the database format
to use a compact bitmap representation for line numbers, which reduced the data
file to 69Mb, a huge win.
The API to the CoverageData object has changed.
Some implications of these changes:
The HTML reporting on contexts is good for small test suites, but very
quickly becomes unwieldy if you have more than 100 tests. Please try using it
and let me know what kind of reporting would be helpful.
The new more-compact data file is harder to query. The larger data file has
a schema designed to be useful for ad-hoc querying. It was a classic
third-normal form representation of the data. Now I consider the database
schema to be a private implementation detail. Should we have a new “coverage
sql” report command that exports the data to a convenient SQLite file?
Because CoverageData has changed, you will need an updated version of
pytest-cov if you use that plugin. The future of the plugin is somewhat up in
the air. If you would like to help maintain it, get in touch. You can install
the up-to-date code with:
To support our hackathon project, we wrote a new pytest plugin: it uses
pytest hooks to indicate the test boundaries, and can read the database and the
code diff to choose the subset of tests to run. This plugin is in very
rough shape (as in, it hasn’t yet fully worked), but if you are interested
in participating in this experiment, get in touch. The code is here
I don’t think this will remain as an independent plugin, so again, if you want
to help with future maintenance or direction, let me know.
One of Jerod’s questions was unexpected: what other open source maintainers
do I appreciate? Two people that came to mind were
Daniel Hahler and
Julian Berman. Some people are
well-known in the Python community because they are the face of large widely
used projects. Daniel and Julian are known to me for a different reason: they
seem to make small contributions to many projects. I see their names in the
commits or issues of many repos I wander through, including my own.
This is a different kind of maintainership: not guiding large efforts, but
providing little pushes in lots of places. If I had had the presence of mind, I
would have also mentioned Anthony Sottile
for the same reason.
And I would have mentioned Mariatta,
for a different reason: her efforts are focused on CPython, but on the
contribution process and tooling around it, rather than the core code itself. A
point I made in the podcast was that people and process challenges are often the
limiting factor to contribution, not technical challenges. Mariatta has been at
the forefront of the efforts to open up CPython contribution, and I wish I had
mentioned her in the podcast.
And I am sure there are people I am overlooking that should be mentioned in
these appreciations. My apologies to you if you are in that category...
Friday was the summer solstice, when day and night are at their most extreme
imbalance. It reminded me of the last summer solstice — and the year of light
and dark, ups and downs, since then — all revolving around Nat, my 29-year-old
Last year on the solstice we were at a block party in a neighborhood in
Boston. We had become close friends with a young couple, and planned for Nat to
move in with them, and for them to be his caregivers.
The couple was eager to extend their family from two to three. We had had
long serious discussions with them about the challenges involved. They knew Nat
pretty well, and had deep experience with similar disabled populations. They
had even moved to a new apartment in order to have space for Nat.
The new apartment was on this quirky cul-de-sac on a hill, a small tight-knit
community, complete with a summer solstice party. It seemed magical, like an
entirely new experience opening up to us. Nat would be moving in with a couple
his own age, with young enthusiasms, and an anything-is-possible approach to the
world. The neighborhood only added to the sense of expanding possibilities. It
seemed like a good plan, almost too good to be true.
We planned for Nat to move at the end of August. We spent lots of time with
his new caregivers over the summer, doing new things from their world. This
helped them understand Nat’s full-day routines better, and was exciting for
The move went great. But over the course of the fall, things started not
going well. Nat has always had periods of anxiety, but it’s hard to pinpoint
the causes. He was going through a bad time, with alarming head-hitting. The
caregivers were having health issues of their own, which made it difficult for
them to give Nat the routine and stability he needs.
We tried to support the new arrangement by having Nat on weekends, and
generally being there for everyone. For reasons that are still not clear to me,
it wasn’t enough, and things just kept going downhill, including our
interactions with the couple. By March, the arrangement that seemed too good to
be true proved to be exactly that. Nat moved back with us.
This was a hard time. Everyone involved reacted in their own ways to the
stress, which caused conflict between us and the caregivers. I think Nat
overall was happy to be back in our house, but his anxieties had not lessened.
Was the change of home part of the cause? We’ll never know.
Parenting Nat has involved a long series of choices for him: where he’ll be
schooled, where he will live, what he will do during his days and nights. These
choices often fall into two broad categories: the exciting but risky, and the
safe but underwhelming — another kind of light and dark. And underlying those
decisions is always the impossible question: are we doing enough?
Now that he was back with us, we had to decide where he would go next. He
could stay with us permanently, but we know that we are perhaps the least
stimulating place for him. We get caught up in our own activities and
interests, and he is passive enough that lots of time passes doing nothing. He
might be fine with it, but it makes us wonder: are we doing enough?
And looming over all of our planning for him is what will happen at the end
of our lives? Now he is 29 and we are 57, but when he is 49 and we are 77 (or
87 or 97!), living together will be a very different story. We want him to have
a life separate from us. We think it will be better for him.
The arrangement we had with the couple is known as “shared living,” and we
thought about whether we wanted to try that again. Nat had been in two shared
living situations by now, and our feeling was that it was too reliant on too few
people. We know shared living has worked for other people, but that’s another
constant in parenting Nat: just because something works for one autism family
doesn’t mean it will work for us. Shared living didn’t seem right for Nat.
We talked with other families we know about what they were planning to do.
But most of them had younger guys, or far more resources, or were making
decisions on longer timescales than us for other reasons. And honestly, housing
together with families we’re already friends with could be like going into
business with friends: a good way to strain or ruin the friendship. We didn’t
want to do that again.
We asked for a new placement in a group home, figuring we’d take a look at
what opened up and see how we felt about it. Two months later we were offered a
placement, in a house run by the same organization as Nat’s previous group home
that he had moved out of the year before.
The residents are a much better fit with Nat this time, and the staff seems
eager and energetic. It’s hard to know whether we are getting accurate answers
from Nat when asked his opinion, but he has been nothing but positive about
moving to this new house. Being in the same organization means we are familiar
with some of the logistics, and Nat will know some of the residents from other
houses when they do things together.
Although a group home generally falls into the safe category rather than the
risky, it feels like this one might be safe without being underwhelming. We
moved him in yesterday, and all seems good. We’ve been through this enough to
know that it won’t be perfect. There will be miscommunications with the
rotating staff, and he’ll come home wearing another resident’s shirt, but
nothing is perfect.
We are still connected to the couple, through other circles. But it is
awkward now, because we have never directly talked about the strains from the
move-out. I hope that we can do that some day.
As I have said before, I
know this is not the last time we will have to make big decisions for Nat. This
one feels good, but others have felt good in the past too. I’m optimistic but
The Heinz company has been using the slogan “57 varieties”
for more than 120 years. But it was never factual. When they introduced the
slogan in 1896, the company already had more than 60 products. The number was
chosen for its sound and “psychological effect.”
It’s hard to know the exact number, but today Heinz has thousands of products,
including at least