Mailing List Archive

Imports and dot-notation
Dear list,


First a disclaimer, I am a python novice ;) so apologies beforehand for
any incorrect terms and use thereof :)

I have a question about the preferred/pythonic way dealing with imports.
But let me start by giving a little bit of an example (which lead me to
this question).

Looking at a python projects code and repository layout, we see the
following directory structure.


/project/core
/project/components/module1
...
/project/components/moduleN
/projects/util

(this is far from complete, but enough to help paint a picture.

Some modules import other modules, and I see (at the very least) two
(kind of three?) different ways of doing so.

`from project.components.module1 import function1, function2 as func,
CONST1, CONST2 as CONST`

or maybe even (which has as an advantage that it becomes clear which
namespace something belongs to

`from project.components.module1 import function1, function2 as
module1_function2, CONST1, CONST2 as MODULE1_CONST2`

but then it really just becomes personal preference, as the number of
characters saved on typing is almost negative (due to having a more
complex import).


but also the dot notation being used

`from project.components import module1`

where obviously the functions are invoked as `module1.function1` etc

I hope the above is clear enough from an example.


Simply put, which of the two would be considered cleaner and more pythonic?

Now for a bit more thought, looking at PEP8, we notes about imports, but
sadly PEP8 does not state which method is better/preferred. While
obviously in the end, it's always the maintainers choice in what they
prefer, so are tabs vs spaces and PEP8 is opinionated about that too (in
a good way, even though I'm not a fan :p).

Likewise, if we look at PEP20, there's quite a few 'guidelines' that
suggest the dot-notation. To name a few (opinionated) examples (but note
I am just using PEP20 as a helper to give some arguments, I know some
people feel strongly against pep20 :p):

* Beautiful is better than ugly.
  - Personally, I think the dot notation is more beautiful, but I
understand that some people find the short notation more beautiful and
less of an eye sore ;)

* Explicit is better than implicit.
  - To me, this is a strong point. Being more explicit is always
better, using the dot notation (if though it's not doing the WHOLE path)
is a nice balance in being explicit enough. Code readability comes from
being more explicit (without being to terse of course).

* Simple is better than complex.
* Flat is better than nested.
  -  I think having more flat and simple imports, and (while more
characters to potentially type in the invocation) more readable
invocations would be better?

* Readability counts.
  - The dot notation seems to help in readability. For those unfamiliar
(or less familiar) with a code base, it becomes incredibly clear where a
something comes from (WITHOUT having to rely on IDE's; sure you can
hover over something and see that the IDE has hopefully worked out where
something came from, but this doesn't work when just reading a page of
code, you aren't going to hover over everything and remember all that
context).

* In the face of ambiguity, refuse the temptation to guess.
  - I'd argue this is almost self-explanatory, not using dot-notation
makes it more ambiguous where something is coming from.

* There should be one-- and preferably only one --obvious way to do it.
  - This is where my confusion/question obviously comes from, as
there's multiple ways of doing it, so maybe PEP8 (or a new one) should
more strongly favor one?

* Namespaces are one honking great idea -- let's do more of those!
  - Basically, this question/proposal would 'force/nudge' you into
writing more with namespaces by using the namespace as prefix.


So what IS the preferred/more pythonic/recommended way to deal with imports?

Why was this not mentioned in PEP8?

Is it worth while to introduce a new PEP or amend PEP8?


For what it's worth, hallucinating chatgpt/google bard both say the same
about this topic, that within the python community, it is generally
preferred and a good idea to use dot-notation. But how correct are they lol.


Thank you for reading,

Olliver



--
https://mail.python.org/mailman/listinfo/python-list
Re: Imports and dot-notation [ In reply to ]
On 09Aug2023 12:30, Oliver Schinagl <oliver+python@schinagl.nl> wrote:
>Looking at a python projects code and repository layout, we see the
>following directory structure.
>
>
>/project/core
>/project/components/module1
>...
>/project/components/moduleN
>/projects/util
>
>(this is far from complete, but enough to help paint a picture.
>
>Some modules import other modules, and I see (at the very least) two
>(kind of three?) different ways of doing so.
>
>`from project.components.module1 import function1, function2 as func,
>CONST1, CONST2 as CONST`
>
>or maybe even (which has as an advantage that it becomes clear which
>namespace something belongs to
>
>`from project.components.module1 import function1, function2 as
>module1_function2, CONST1, CONST2 as MODULE1_CONST2`
>
>but then it really just becomes personal preference, as the number of
>characters saved on typing is almost negative (due to having a more
>complex import).
>
>
>but also the dot notation being used
>
>`from project.components import module1`
>
>where obviously the functions are invoked as `module1.function1` etc
>
>I hope the above is clear enough from an example.
>
>
>Simply put, which of the two would be considered cleaner and more pythonic?

This is almost entirely a programmer preference thing.

The Zen, as usual, offers guideance but not dictates.

As you say, the module1.func form is very clear and has its beauty.
Personally I lean towards the single name flavour, sometimes with a
rename. FOr example, with os.path I often go:

from os.path import join as joinpath

and likewise for several other names from there, because the bare names
are very generic (eg "join"). I see to be alone here. Many other people
use:

import os.path

and use the full os.path.join in their code, which I find jarring and
visually noisy.

>Now for a bit more thought, looking at PEP8, we notes about imports,
>but sadly PEP8 does not state which method is better/preferred. While
>obviously in the end, it's always the maintainers choice in what they
>prefer, so are tabs vs spaces and PEP8 is opinionated about that too
>(in a good way, even though I'm not a fan :p).

PEP8 is for the stdlib source code; that said it is also the basis for
most Python coding styles in the wild.

For me, the Zen's "readability counts" is the most important guideline;
and to this end, your "module1.function" is an entirely reasonable
response, particularly if you've got a lot of imports. But it does make
for more verbose code, and that works against readability to a degree.

This is why it's all subjective.

Particular workplaces may mandate particular styles, but in your
personal code? Do what seem best, and experience will cause that to
evolve over time.

Also, the style can depend on the code you're working on. For a small
module with few imports the:

from module1 import funcname

can be a win because the code is short, and this short form of funcname
is both clear and readable. But in a very long piece of code with many
imports you might go for module1.funcname for clarity, particularly if
funcname is generic or overlaps with another similar imported name.

Cheers,
Cameron Simpson <cs@cskk.id.au>
--
https://mail.python.org/mailman/listinfo/python-list
Re: Imports and dot-notation [ In reply to ]
On 09/08/2023 22.30, Oliver Schinagl via Python-list wrote:
...> Looking at a python projects code and repository layout, we see the
> following directory structure.
>
> /project/core
> /project/components/module1
> ...
> /project/components/moduleN
> /projects/util
...> Some modules import other modules, and I see (at the very least) two
> (kind of three?) different ways of doing so.
>
> `from project.components.module1 import function1, function2 as func,
> CONST1, CONST2 as CONST`
>
> or maybe even (which has as an advantage that it becomes clear which
> namespace something belongs to
>
> `from project.components.module1 import function1, function2 as
> module1_function2, CONST1, CONST2 as MODULE1_CONST2`
>
> but then it really just becomes personal preference, as the number of
> characters saved on typing is almost negative (due to having a more
> complex import).
>
> but also the dot notation being used
>
> `from project.components import module1`
>
> where obviously the functions are invoked as `module1.function1` etc

Firstly, the path followed will depend upon the starting position!

Keep reading and you should come across a reference to 'hobgoblin of
little minds'.

What should we be concentrating on/about? If 'naming' is a great (?the
greatest) challenge of programming, surely remembering the names of
classes, methods, functions, modules, etc; follows...
(and by implication, as you've illustrated, perhaps where they
come-from), ie code is read more often than it is written, so think
about comprehension rather than typing!


If one (one's team!) frequently uses a module, then a jargon may
develop. For example:

import numpy as np
...
a = np.arange(6)

In which case, those au-fait with the jargon know that the "np." prefix
tells them where "arange()" can be found - conversely, that if the IDE
doesn't alert, that np/numpy must be (first) import-ed.

However, there is a cognitive-load to 'translate' "np" into "numpy". Not
much, but then over-load is not a single thought but the combination of
'everything'.

The second option (OP) is very (laboriously) clear in mapping the source
of each function or constant. By way of comparison then, the code may
now appear cluttered, because there's so much text to read. There would
be less if an abbreviation were used.

The dev.tool in-use may also influence this decision. If hovering-over
an identifier reveals source-information, what value the extra code?
Intelligent completion also reduces relevance of 'number of characters
saved on typing'.

Accordingly, give frequently-used functions/modules the abbreviation
treatment -but only if YOU feel it right. Otherwise, use a longer-form
to improve readability/comprehension.

THE answer will thus vary by library/package/module, by application, and
by coder (jargon works best if 'all' use it/understand same).

Side note: Using "...import identifier, ..." does not save storage-space
over "import module" (the whole module is imported regardless, IIRC),
however it does form an "interface" and thus recommend leaning into the
"Interface Segregation Principle", or as our InfoSec brethren would say
'the principle of least privilege'. Accordingly, prefer "from ... import
... as ...".

--
Regards,
=dn
--
https://mail.python.org/mailman/listinfo/python-list
Re: Imports and dot-notation [ In reply to ]
On 8/9/23 17:28, dn via Python-list wrote:

> Side note: Using "...import identifier, ..." does not save storage-space
> over "import module" (the whole module is imported regardless, IIRC),
> however it does form an "interface" and thus recommend leaning into the
> "Interface Segregation Principle", or as our InfoSec brethren would say
> 'the principle of least privilege'. Accordingly, prefer "from ... import
> ... as ...".
>

Attribute lookup has *some* cost. That is, finding "c" in the local
namespace is going to be a little quicker than "b.c", where Python finds
"b" in the local namespace and then finds its "c" attribute; that's then
a little quicker than "a.b.c", etc.

See all relevant commentary about premature optimisation, spending time
optimising the wrong things, etc. but there *are* cases where it
matters (c.f. a tight loop that will be run many many times)
--
https://mail.python.org/mailman/listinfo/python-list