Reads a .env file. Values keep the type they plainly have, and nothing is expanded unless you
ask for it.
Most .env libraries resolve one value from another out of the box. This one does not, and the
model it follows is JavaScript's.
dotenv for Node is the most installed .env library anywhere, and it does not interpolate.
Nor does dotenv-java. In that world, building values from other values is a second package,
because it is a second decision: it turns a list of pairs into a small language, with escaping
and ordering and undefined names to settle.
The same choice is made here, and by
quillstack/dotenv in PHP. A .env file which quietly
became a template is a .env file you have to read twice.
if settings["APP_DEBUG"]:
...Read as text, false is a non-empty string and therefore true. That is a bug which looks like
working code, and it is the reason the values are typed rather than handed back as strings.
os.environ holds strings and nothing else, so a port put there comes back '5432' and a
false comes back true — the exact thing the typing avoids. What is read stays where its types
survive, and export() is there for when something else needs the environment set.
- Python 3.11 or newer
pip install quillstack-dotenvAPP_DEBUG=true
APP_NAME=quillstack
DB_PORT=5432
from quillstack.dotenv import Dotenv
settings = Dotenv(".env").load()
settings["APP_DEBUG"] # True, a boolean
settings["APP_NAME"] # 'quillstack'
settings["DB_PORT"] # 5432, a numberIt is a Mapping, so len(), in, iteration and unpacking all work on it.
Quote it:
DB_PORT_TEXT="5432"
settings["DB_PORT_TEXT"] # '5432', a stringsettings.get("MISSING", "a default") # 'a default'
settings.get("MISSING") # NoneWhere there is no sensible default, say so and find out at boot rather than at midnight:
settings.required("DATABASE_HOST")ValueNotSetError: Value not set for key: DATABASE_HOST
A # starts a comment where a shell would treat it as one — after whitespace, and outside
quotes:
COMMENTED=5432 # the default
PASSWORD=hunter2#7
QUOTED="a # inside quotes"
settings["COMMENTED"] # 5432
settings["PASSWORD"] # 'hunter2#7'
settings["QUOTED"] # 'a # inside quotes'A parser which took every # would turn a password into a shorter password and say nothing
about it.
URL=https://${APP_NAME}.org
settings["URL"] # 'https://${APP_NAME}.org'For the sake of something else which reads os.environ directly:
settings.export() # leaves what is already set alone
settings.export(override=True) # does notWhat goes out is text: True is exported as true, because that is what it was in the file and
what another reader expects to find. A deployment which set a variable meant it, which is why
what is already there wins unless you say otherwise.
Against python-dotenv 1.2.3, on Python 3.13.5, reading the same 23-pair file. Interleaved
runs, median of five.
Read what each one does before reading the numbers, because they are not doing the same work — in both directions:
quillstack-dotenv |
python-dotenv |
|
|---|---|---|
A=true |
True |
'true' |
B=5432 |
5432 |
'5432' |
C=${A} |
'${A}' |
'true' — expanded |
| multi-line values | no | yes |
${MISSING:-fallback} |
no | yes |
| escapes inside quotes | no | yes |
So one of them reads types and the other builds a small language. Now the numbers:
| Per load | Relative | |
|---|---|---|
| quillstack-dotenv | 32.0 µs | — |
| python-dotenv | 459.2 µs | 14× |
That is not the number that matters, because a .env file is read once. What an application
pays is the whole thing — importing the library and loading the file, over 25 process starts:
| Whole process | Of which is the library | |
|---|---|---|
bare python, importing nothing |
14.49 ms | — |
| quillstack-dotenv | 19.38 ms | 4.9 ms |
| python-dotenv | 25.84 ms | 11.4 ms |
About 70% of that is Python starting, and no .env library can do anything about it. Choosing
between these two moves a boot by six milliseconds.
Being faster because you do less is not being faster. If you want multi-line values, defaults
inside ${…}, or escapes, python-dotenv has them and this does not — and it is the more
popular library in this ecosystem by a wide margin, which is worth knowing when you pick.
uv run pytestuv run ruff check --no-cache
uv run mypy--no-cache on purpose: a cached ruff result once said a Quillstack package was clean while CI
said it was not.
uv run quillstack-standards .This is one component of Quillstack, the same way of building APIs in more than one language.
- quillstack-standards — what keeps every package the same shape
- quillstack/dotenv — the same decisions in PHP
- quillstack/dotenv-expand — values built from other values, where you want them
MIT — see LICENSE.