Parent No Raw Loops

Sean Parent Style Guide⁠‍⁠​‌​‌​​‌‌‍​‌​​‌​‌‌‍​​‌‌​​​‌‍​‌​​‌‌​​‍​​​​​​​‌‍‌​​‌‌​‌​‍‌​​​​​​​‍‌‌​​‌‌‌‌‍‌‌​​​‌​​‍‌‌‌‌‌‌​‌‍‌‌​‌​​​​‍​‌​‌‌‌‌‌‍​‌​​‌​‌‌‍​‌‌​‌​​‌‍‌​‌​‌‌‌​‍​​‌​‌​​​‍‌‌‌​‌​‌‌‍‌‌‌‌‌​‌‌‍‌‌​​​‌​​‍​​​​‌‌​​‍​‌‌​​​​‌‍​​​​‌​‌​‍‌​​‌​‌‌​⁠‍⁠

Overview

Sean Parent, former Principal Scientist at Adobe, transformed how many think about C++ with his "C++ Seasoning" and "Better Code" talks. His central thesis: raw loops are assembly language for algorithms. If you're writing a loop, you're probably missing an algorithm.

Core Philosophy

"No raw loops."

"A goal of software engineering is to reduce code to its essence, to remove anything that doesn't contribute to the meaning."

Parent believes that code should be a direct expression of intent, and loops obscure intent by exposing mechanism.

Design Principles

1. No Raw Loops: Every loop is a missed opportunity to use (or create) a named algorithm.

2. Algorithms Express Intent: std::find_if says "search"; a for-loop says "increment and compare."

3. Composition Over Iteration: Build complex operations from simple, well-named pieces.

4. Seek the Essence: Remove everything that doesn't contribute to meaning.

When Writing Code

Always

• Use standard algorithms when they fit exactly
• Create named algorithms when standard ones don't fit
• Prefer algorithms that express the operation's semantic meaning
• Use range-based operations (C++20 ranges when available)
• Compose simple operations rather than write complex loops
• Name intermediate variables to document intent

Never

• Write raw for loops when an algorithm exists
• Nest loops when composition would work
• Inline complex logic that deserves a name
• Sacrifice clarity for cleverness
• Leave unnamed concepts in code

Prefer

• std::transform over element-by-element loops
• std::accumulate over manual aggregation
• std::partition over manual reordering
• std::remove_if + erase over manual deletion
• Algorithm pipelines over nested loops

The Algorithm Catalog

Existence Queries

Finding

Transforming

Reordering

Code Patterns

Before and After: The Transformation

// RAW LOOP: What is this doing?
std::vector<int> result;
for (const auto& item : items) {
    if (item.isValid()) {
        result.push_back(item.getValue() * 2);
    }
}

// ALGORITHM: Intent is clear
auto result = items 
    | std::views::filter(&Item::isValid)
    | std::views::transform([](const Item& i) { return i.getValue() * 2; })
    | std::ranges::to<std::vector>();

// Or without C++20 ranges:
std::vector<int> result;
std::transform(
    items.begin(), items.end(),
    std::back_inserter(result),
    [](const Item& i) -> std::optional<int> {
        return i.isValid() ? std::optional{i.getValue() * 2} : std::nullopt;
    }
);
// (Then filter nullopt... or use a custom transform_if)

The Erase-Remove Idiom

// RAW LOOP: Error-prone, unclear intent
for (auto it = vec.begin(); it != vec.end(); ) {
    if (shouldRemove(*it)) {
        it = vec.erase(it);  // O(n) each time!
    } else {
        ++it;
    }
}

// ALGORITHM: O(n) total, clear intent
vec.erase(
    std::remove_if(vec.begin(), vec.end(), shouldRemove),
    vec.end()
);

// C++20: Even clearer
std::erase_if(vec, shouldRemove);

Slide Algorithm (Parent's Signature)

// Move a range to a new position within a sequence
template<typename I>  // I models RandomAccessIterator
auto slide(I first, I last, I pos) -> std::pair<I, I> {
    if (pos < first) return { pos, std::rotate(pos, first, last) };
    if (last < pos)  return { std::rotate(first, last, pos), pos };
    return { first, last };
}

// Usage: Move selected items to position
auto [new_first, new_last] = slide(
    selection_begin, selection_end, 
    drop_position
);

Gather Algorithm

// Move all elements satisfying predicate to position
template<typename I, typename P>
auto gather(I first, I last, I pos, P pred) -> std::pair<I, I> {
    return {
        std::stable_partition(first, pos, std::not_fn(pred)),
        std::stable_partition(pos, last, pred)
    };
}

// Usage: Gather all selected items around cursor
auto [sel_first, sel_last] = gather(
    items.begin(), items.end(),
    cursor_position,
    [](const Item& i) { return i.selected; }
);

Mental Model

Parent thinks of code as mathematical composition:

1. Name the operation: What am I really doing?
2. Find the algorithm: Does this operation have a name?
3. Compose primitives: Can I build this from smaller operations?
4. Factor out patterns: Is this useful elsewhere?

The Refactoring Test

When you see a loop, ask:

1. Is this searching? → find, search, any_of...
2. Is this transforming? → transform, copy_if...
3. Is this reordering? → sort, partition, rotate...
4. Is this aggregating? → accumulate, reduce...
5. Is this comparing? → equal, mismatch...

If none fit exactly, write and name a new algorithm.