Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
272 changes: 166 additions & 106 deletions spec-0012/index.md
Original file line number Diff line number Diff line change
Expand Up @@ -4,122 +4,182 @@ number: 12
date: 2024-06-06
author:
- "Pamphile Roy <roy.pamphile@gmail.com>"
- "Matt Haberland <mhaberla@calpoly.edu>"
discussion: https://discuss.scientific-python.org/t/spec-12-formatting-mathematical-expressions
endorsed-by:
---

## Description

It is known that the PEP8 and other established styling documents are missing
guidelines about mathematical expressions. This leads to people coming with
their own interpretation and style. Standardizing the way we represent maths
would lead to the same benefits seen with "normal" code. It brings consistency
in the ecosystem improving the collaborative efforts.
[PEP 8](https://peps.python.org/pep-0008)
and other established styling documents do not include guidelines about
styling mathematical expressions. This leads to individual interpretation and
styles which may conflict with those of others. We seek to standardizing the
way we represent mathematics for the same reason we standardize other code:
it brings consistency to the ecosystem and allows collaborators to focus on
more important aspects of the code.

This SPEC standardize the formatting of mathematical expressions.

## Implementation

The following rules must be followed.
These rules respect and complement the PEP8 (relevant sections includes
[id20](https://www.python.org/dev/peps/pep-0008/#id20) and
[id20](https://www.python.org/dev/peps/pep-0008/#id28)).

We define a _group_ as a collection of operators having the same priority.
e.g. `a + b + c` is a single group, `a + b * c` is composed of two groups `a`
and `b * c`. A group is also a collection delimited with parenthesis.
`(a + b * c)` is a group. And the whole expression by itself is a
group.

- There a space before and after `-` and `+`. Except if
the operator is used to define the sign of the number;

```
a + b
-a
```

- Within a group, if operators with different priorities are used, add whitespace around the operators with the lowest priority(ies).

```
a + b*c
```

- There is no space before and after `**`.

```
a**b
```

- There is no space before and after operators `*` and `/`. Only exception is if the expression consist of a single operator linking two groups with more than one
element.
## Terminology

```
a*b
(a*b) * (c*d)
```

- Operators within a group are ordered from the lowest to the highest priority.
If this is technically an issue (e.g. restriction on the AST), add
parenthesis or spaces.

```
a/d*b**c
a*(b**c)/d
a*b**c / d
a * b**c / d
```

- When splitting an equation, new lines should start with the operator linking
the previous and next logical block. Single digit on a line are forbidden.

```
(
a/b
+ c*d
)
```

### Examples

```python
# good
i = i + 1
submitted += 1
x = x*2 - 1
hypot2 = x*x + y*y
c = (a + b) * (a - b)
dfdx = sign*(-2*x + 2*y + 2)
result = 2*x**2 + 3*x**(2/3)
y = 4*x**2 + 2*x + 1
c_i1j = (
1./n**2.
*np.prod(
0.5*(2. + abs(z_ij[i1, :]) + abs(z_ij) - abs(z_ij[i1, :] - z_ij)), axis=1
)
)
```

```python
# bad
i = i + 1
submitted += 1
x = x * 2 - 1
hypot2 = x * x + y * y
c = (a + b) * (a - b)
dfdx = sign * (-2 * x + 2 * y + 2)
result = 2 * x ** 2 + 3 * x ** (2 / 3)
y = 4 * x ** 2 + 2 * x + 1
c_i1j = (
1.0
/ n ** 2.0
* np.prod(
0.5 * (2.0 + abs(z_ij[i1, :]) + abs(z_ij) - abs(z_ij[i1, :] - z_ij)), axis=1
)
)
An "explicit" expression is a code expression enclosed within parentheses or
otherwise syntactically separated from other expressions (i.e. by code other
than operators, whitespace, literals, or variables). For example, in the list
Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not sure if this list is complete.

comprehension:
```python3
for j in range(1, i + 1)
```
The output expression `j` is one explicit expression and the input sequence
`range(1, i + 1)` is another.

A "subexpression" is subset of an expression that is either explicit or could
be made explicit (i.e. with parentheses) without affecting the order of
operations. In the example above, `j` and `range(1, i + 1)` can also be
referred to as explicit subexpressions of the whole expression, and `1` and
`i + 1` are explicit subexpressions of the expression `range(1, i + 1)`. `i` and
`1` are "implicit" subexpressions of `i + 1`: they could be written as explicit
subexpressions `(i)` and `(1)` without affecting the order of operations, but they
are not explicit as written.
Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this concept might still need refinement to make the rules unambiguous, but let's see if others find ambiguities before adding unnecessarily.

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was not sure about how detailed this should be as I would imagine it could depend on how formatter operate internally.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

To some extent, I intended to write this to communicate the rules with the developers of Black and ruff. If it does not need to be as precise for readers, the formal definitions can be moved to the postscript for the interested reader.


As another example, in `x + y*z`, `y*z` is a subexpression because it could be made
explicit as in `x + (y*z)` without changing the order of operations. However, `x + y`
would not be a subexpression because `(x + y)*z` would change the order of operations.
Note that `x + y*z` as a whole may also be referred to as a "subexpression" rather than
an "expression" even though `(x + y*z)` is not a proper subset of the whole.

A "simple" expression is an expression involving only one operator priority level
without considering the operators within explicit subexpressions.
A "compound" expression is an expression involving more than one operator
priority level without considering the contents of explicit subexpressions.
For example,
- `x + y - z` is a simple expression because `+` and `-` have the
same priority level. There are no explicit subexpressions to be ignored.
- `x * (y + z)` is also a simple expression because there is only one operator
between `x` and the explicit subexpression `(y + z)`; we ignore the contents - and
especially the operator - within the explicit subexpression; conceptually, it may
regarded as `(...)`.
- `x * y + z` is a compound expression; there are two operators and no explicit
subexpressions that can be ignored.

The acronym PEMDAS commonly refers to "parentheses", "exponentiation", "multiplication",
"division", "addition", and "subtraction". Herein, we will consider these operators
to be "PEMDAS operators", and we will also include the unary `+`, `-`, and `~` in
this category for convenience. The order of operations of PEMDAS operators is typically
taught in primary school and reinforced throughout a programmer's training and
experience, so it is assumed that most programmers are comfortable relying on the
implicit order of operations of expressions involving a few PEMDAS operations. Implicit
order of operations becomes less obvious as the number of distinct operator priority
levels increases and when multiple non-PEMDAS operators are involved. Portions of this
acronym, namely MD and AS, will be used below to refer to the corresponding operators.

## Notes
## Implementation

These formatting rules do not make any consideration in terms of performances
nor precision. The scope is limited to styling.
These rules are intended to respect and
complement the [PEP 8 standards](https://peps.python.org/pep-0008), such as using
[implied line continuation](https://peps.python.org/pep-0008/#maximum-line-length) and
and [breaking lines before binary operators](https://peps.python.org/pep-0008/#should-a-line-break-before-or-after-a-binary-operator).
Comment thread
mdhaber marked this conversation as resolved.
Although examples do not show the use of hanging indent, any of the indentation styles
allowed by [PEP 8 Indentation](https://peps.python.org/pep-0008/#indentation) are
permitted by this SPEC.

0. Unless otherwise specified, rely on the implicit order of operations;
i.e., do not add extraneous parentheses. For example, prefer `u**v + y**z`
over `(u**v) + (y**z)`, and prefer `x + y + z` over `(x + y) + z`. A full
list of implicit operator priority levels is given by
[Operator Precedence](https://docs.python.org/3/reference/expressions.html#operator-precedence)
1. Always use the `**` operator and unary `+`, `-`, and `~` operators *without*
surrounding whitespace. For example, prefer `-x**4` over `- (x ** 4)`.
Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

prefer: I think we should be more direct, these are rules and there should be no ambiguity. And you have rule 10 to break things if needed.

With such document, the norm is to use something like the RFC-2119 https://datatracker.ietf.org/doc/html/rfc2119

Here and bellow with things like "should". This makes it too ambiguous to me as well. There should be one and only one was to do things. That's the goal of this doc, that it can be used by any formatter and it would give the same output consistently.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We can change "prefer... over" to "use... instead of"?

I read RFC-2119 to mean that "should" is the prefered use because these are not absolute requirements. There are exceptions as defined by rule 10.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@lucascolley what do you think on this point?

2. Always surround non-PEMDAS operators with whitespace, and always make the priority of
non-PEMDAS operators explicit. For example, prefer `(x == y) or (w == t)` over
`x==y or w==t`.[^1]
3. Always surround AS operators with whitespace.
4. Typically, surround MD operators with whitespace, except in the following situations.
- When there are lower-priority operators (namely AS) within the same compound
expression. For example, prefer `z = -x * y**t` over `z = -x*y**t`, but
prefer `z = w + x*y**t` over `z = w + x * y**t` due to the presence of the
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

As below, I think there should be parentheses here to distinguish between * and **:

z = w + x*(y**t)

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Also, how would the rules format w * x * y**z + u ** v * q?

I think my preference is:

(w * x * y**z) + (u**v * q)

But I would accept:

(w*x*(y**z)) + ((u**v)*q)

Copy link
Copy Markdown
Author

@mdhaber mdhaber Aug 11, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here, the presence of the lower priority operator + in the expression means that there will be no space around the multiplication operators. Also, ** is supposed to appear at the end of "implicit subexpression without spaces". So it would be w*x*y**z + q*u**v.

This is intended to be reminiscent of mathematics in which we would write
$wxy^z + qu^v$.

All operators with priority lower than PEMDAS operators need parentheses. But I didn't require parentheses in expressions that involve, say, multiplication and addition because I think people are familiar with the order of operations of such operators from elementary school and it is natural to them to omit the parentheses . I posit that it is just as natural to write w*x + y*z instead of (w*x) + (y*z) as it is to write w**x * y**z instead of (w**x) * (y**z). Similarly, I think putting ** at the end of subexpressions without spaces is enough to distinguish it visually from *.

The rules would be a bit simpler to express, actually, if we did require those parentheses like you want. But I'm not sure if that's very common in real code. I had ChatGPT write me a script to extract long math operations from the codebase.. maybe I'll resurrect that to see.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What about w*x*(y**z) + q*(u**v)? I would be happy with that.

Copy link
Copy Markdown
Author

@mdhaber mdhaber Aug 11, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We need a stronger rationale than what makes one of us happy to change that : )

Would you really write $qu^v$ as $q(u^v)$?

And you'll say no, but the superscript stands out so it's readable. Then I say ok, well I think the ** at the end stands out so it's readable.

Is what you're suggesting much more common in code already? That would be a good reason to change.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not sure what's more common, but indeed the problem for me is that it doesn't stand out. Especially for longer expressions such as

a + b*c*d*e**f

so we just disagree on

I think putting ** at the end of subexpressions without spaces is enough to distinguish it visually from *.

But yeah, I'm happy to be outvoted on this one. Just for the record!

Copy link
Copy Markdown
Author

@mdhaber mdhaber Aug 11, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

And yeah, I think that's what we need is more opinions. Maybe we could add a limit on the number of chained operators in subexpressions without spaces, because I agree that b*c*d*e**f is not great style. But then maybe that's one of those cases where the last rule says - here, we can break a rule that has stopped making sense in this extreme situation.

lower-priority addition operator.
- The division operation would be written mathematically as a fraction with a
horizontal bar. For example, prefer `z = t/v * x/y` over `z = t / v * x / y`
if this would be written mathematically as the product of two fractions,
e.g. $\frac{t}{v} \cdot \frac{x}{y}.
5. Considering the previous rules, only `**`, `*`, `/`, and the unary `+`, `-`, and `~`
Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Considering the previous rules

Linked to my comment saying that we should say if rules are to be applied in a specific order or not.

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I also don't get the "only". To me it looks like all operators are listed.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note the distinction between unary and binary +/- and the existence of many other operators.

operators can appear in implicit subexpressions without spaces. In such expressions,
- Use at most one unary operator, and if used, ensure that it is the leftmost operator.
- Use at most one `**` operator, and if used, ensure that it is the rightmost operator.

To achieve these goals, simplification or the addition of parentheses may be required.
For example:
- The expressions `--x` and `-~x` would be implicit subexpressions without spaces
containing more than one unary operator. The former can be simplified to `+x` or
simply `x`, and the latter requires explicit parentheses, i.e. `-(~x)`.
Comment thread
mdhaber marked this conversation as resolved.
Comment on lines +113 to +115
Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We can leave this for now and see what formatting folks tell us about the feasibility of such thing. Like if we go down that route, why not also recommend some operations vs other for precision issue? All that could seem a bit too far for a formatter though.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sounds good. Note that these are all just examples of the implications of the two simple rules above that I thought were worth pointing out specifically.

- The expression `x**y**z` would be an implicit subexpression without spaces
containing more than one `**` operator. This code would be executed as `x**(y**z)`
following the implicit order, but the explicit parentheses should be included for
clarity.
- In the expression `t**v*x**y + z`, no spaces are used around the multiplication
operator due to the presence of the lower-priority addition operator. However,
this would lead to `t**v*x**y` being an implicit subexpression without spaces
containing more than one `**` operator. This code would be executed as
`(t**v)*(x**y) + z`, but the explicit parentheses should be included for clarity.
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

👍

- In the expression `z + x**y/w`, no spaces are used around the division operator
due to the presence of the lower-priority addition operator. However, this would
lead to `x**y/w` being an implicit subexpression without spaces containing `**`
to the left of another operator. Options for refactoring include the addition of
parentheses (e.g. `z + (x**y)/w`) or pre-multiplying the exponential by a
fraction (i.e. `x + 1/w*x**y`).
6. Simplify combinations of unary and binary `+` and `-` operators when possible.
For example,
- prefer `x + y` over `x + +y`,
- prefer `x + y` over `x - -y`,
- prefer `x - y` over `x - +y`, and
- prefer `x - y` over `x + -y`.
7. If required to satisfy other style requirements, include line breaks before
the outermost explicit subexpression possible. For example, if
`t + (w + (x + (y + z))))` must be broken, prefer
```python3
(t
+ (w + (x + (y + z)))))
Comment on lines +141 to +142
Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We could clarify that we don't specify how parenthesis should be handled and it's only about where to break on mathematical operations.

Copy link
Copy Markdown
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

```
over
```python3
(t + (w + (x + (y
+ z)))))
```
If there are multiple candidates, include the break at the first opportunity.
8. If line breaks must occur within a compound subexpression, the break should
be placed before the operator with lowest priority. For example, if
(x + y*z) must be broken, prefer
```python3
(x
+ y*z)
```
over
```python3
(x + y
* z)
```
If there are multiple candidates, include the break at the first opportunity.
9. Any of the preceeding rules may be broken if there is a clear reason to do so.
- *Conflict with other style rules*. For example, there is not supposed to be
whitepace surrounding the `**` operator, but one can imagine a chain of `**`
operations that exhausts the character limit of a line.
- *Domain knowledge*. For instance, in the expression
`t = (x + y) - z`, it may be important to emphasize that the addition should be
performed first for numerical reasons or because `(x + y)` is a conceptually
important quantity. In such cases, consider adding a comment, e.g.
```python3
t = (x + y) - z # perform `x + y` first for precision
```
or breaking the expressions into separate logical lines, e.g.
```python3
w = x + y
t = w - z
```

[^1]: There is a case for simply eliminating spaces to reinforce the implicit order
of operations, as in `x==y or w==t`. However, if this were the rule, following
the rule would require users to remember the full order of operations hierarchy
and apply it without mistakes. Use of explicit parentheses with non-PEMDAS
operators leads to simpler rules, is more explicit, and is not uncommon in
existing code.