Cpp Templates

C++ Templates

Purpose

Guide agents through reading and fixing template error messages, using concepts as cleaner constraints, understanding SFINAE vs concepts trade-offs, and profiling template instantiation depth and compile times with Templight.

Triggers

• "How do I read this massive C++ template error?"
• "How do I use concepts to constrain a template?"
• "What's the difference between SFINAE and concepts?"
• "My templates make compilation very slow"
• "How do I write a requires-clause?"
• "How do I profile template instantiation times?"

Workflow

1. Reading template error messages

Template errors print full instantiation chains. Strategy: read from the bottom up.

prog.cpp:25:5: error: no matching function for call to 'sort'
  std::sort(v.begin(), v.end());
  ^~~~~~~~~
/usr/include/c++/13/bits/stl_algo.h:4869:5: note: candidate:
    template
    void std::sort(_RAIter, _RAIter)
note: template argument deduction/substitution failed:
prog.cpp:25:5: note: 'MyType' is not a valid type for this template
                             ^~~~~~~~

Rules for reading:

1. Find the first error line (top of output) — that's your code
2. Skip all the note: lines until you find "required from here" or "in instantiation of"
3. The bottom of the stack shows the type that failed substitution

# Limit backtrace depth to reduce noise
g++   -ftemplate-backtrace-limit=3  prog.cpp
clang -ftemplate-depth=32           prog.cpp   # default 1024

# Show simplified errors (GCC 12+)
g++ -fconcepts-diagnostics-depth=3  prog.cpp   # for concept failures

2. SFINAE — legacy constraint technique

SFINAE (Substitution Failure Is Not An Error) silently removes overloads that fail substitution:

#include 

// Enable function only for arithmetic types
template , int> = 0>
T square(T x) { return x * x; }

// SFINAE with return type
template 
auto to_string(T x) -> std::enable_if_t, std::string> {
    return std::to_string(x);
}

// Void-t technique for detecting member existence
template 
struct has_size : std::false_type {};

template 
struct has_size().size())>>
    : std::true_type {};

SFINAE errors are cryptic. Prefer concepts (C++20) for new code.

3. Concepts — modern constraints (C++20)

#include 

// Define a concept
template 
concept Arithmetic = std::is_arithmetic_v;

template 
concept Printable = requires(T x) {
    { std::cout  std::same_as;
};

template 
concept Container = requires(T c) {
    c.begin();
    c.end();
    c.size();
    typename T::value_type;
};

// Apply concept as constraint
template 
T square(T x) { return x * x; }

// Abbreviated function template (C++20)
auto square(Arithmetic auto x) { return x * x; }

// requires-clause (more complex conditions)
template 
    requires Arithmetic && (sizeof(T) >= 4)
T big_square(T x) { return x * x; }

// Concept in auto parameter
void print_container(const Container auto& c) {
    for (const auto& elem : c) std::cout  type; } — checks type of expression

template 
concept HasPush = requires(T c, typename T::value_type v) {
    c.push_back(v);                          // must be valid
    { c.front() } -> std::same_as;  // type check
    { c.size() } -> std::convertible_to;        // convertible
    requires std::default_initializable;  // nested requirement
};

// Compound requires (all must hold)
template 
concept Sortable = requires(T a, T b) {
    { a  std::convertible_to;
    { a == b } -> std::convertible_to;
};

5. SFINAE vs concepts comparison

| Aspect | SFINAE | Concepts | |--------|--------|---------| | Syntax | Complex, verbose | Clean, readable | | Error messages | Cryptic wall-of-text | Clear constraint failure | | Compile time | Can be slow (many substitutions) | Generally faster | | C++ version | C++11 | C++20 | | Short-circuit | No | Yes (concept subsumption) | | Use in if constexpr | Awkward | Natural | | Overload ranking | Manually via priority | Automatic by constraint specificity |

Migration: replace enable_if with concept constraints; replace void_t helpers with requires.

6. Template instantiation profiling with Templight

# Install Templight (Clang-based profiler)
# https://github.com/mikael-s-persson/templight

# Build with Templight tracing
clang++ -Xtemplight -profiler -Xtemplight -memory \
        -std=c++17 prog.cpp -o prog

# Convert trace to visualizable format
templight-convert -f callgrind -o prof.out templight.pb

# View with KCachegrind
kcachegrind prof.out

# Find top template instantiation costs (without Templight)
# ClangBuildAnalyzer (easier)
ClangBuildAnalyzer --start /tmp/build
cmake --build build
ClangBuildAnalyzer --stop /tmp/build capture.bin
ClangBuildAnalyzer --analyze capture.bin | head -50

7. Reducing template compile times

// 1. Explicit instantiation — compile once, use everywhere
// header.h
template 
T transform(T x);

extern template int transform(int);    // suppress instantiation

// impl.cpp
#include "header.h"
template int transform(int);           // instantiate here only

// 2. Prefer function templates over class templates when possible
// (functions instantiate lazily; class templates instantiate eagerly)

// 3. Use concepts to short-circuit failed substitutions
// (concept check is faster than full substitution failure)

// 4. Split heavy template headers from lightweight ones
// - Put type definitions in forward_decls.h
// - Put template implementations in impl.h (include only where needed)

// 5. Use if constexpr instead of specialization
template 
void process(T x) {
    if constexpr (std::is_integral_v) {
        handle_int(x);
    } else {
        handle_other(x);
    }
}

Related skills

• Use skills/build-systems/build-acceleration for ccache and PCH to reduce overall compile time
• Use skills/compilers/clang for Clang-specific diagnostics and concept error output
• Use skills/low-level-programming/cpp-coroutines for another advanced C++20 feature

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: mohitmishra786
• Source: mohitmishra786/low-level-dev-skills
• License: MIT
• Homepage: https://www.lowleveldevskills.com

Install and usage instructions live in the source repository linked above.