Abstract:

Klaus Iglberger argues that C++ does not have a technical safety problem, but a "people problem." He asserts that modern C++ features—Ranges, constexpr, Strong Types, and Value Semantics—already provide comprehensive solutions for bounds, type, and lifetime safety. However, the vast majority of C++ developers (the "90%") remain trapped in a 1990s mindset, ignoring these tools in favor of "classic" (unsafe) patterns. Iglberger critiques current teaching materials and industry inertia, concluding that C++ can be safe by default only if the community stops treating it like a low-level "driving without a road" experience and starts enforcing modern architectural principles.

The Real Problem of C++: Mindset Over Language

• 0:51 Government Pressures: Iglberger acknowledges the NSA, White House, and EU are actively flagging C++ as a cybersecurity risk. As far as I can tell, the probability of regulatory intervention is nearing 100% unless the industry shifts.
• 3:35 Fast vs. Safe: The core C++ philosophy is that safe can be built on top of fast, but not vice-versa. I find this premise technically sound but practically ignored by the average developer.
• 12:01 Bound Safety via Ranges: The "No Raw Loops" rule (Sean Parent) is the primary solution. Using std::ranges eliminates manual index handling and out-of-bounds errors. I don't care about the "it's too complex" excuse; it's basic competence.
• 23:24 UB and constexpr: constexpr is a high-utility tool for catching undefined behavior (UB) at compile time. If code is evaluated as a constant expression, the compiler must flag UB. I'd estimate that 90% of developers don't realize constexpr doubles as a safety scanner.
• 32:21 Strong Types: Stop passing raw doubles and ints for different units. Using wrapper classes with explicit constructors and concepts prevents accidental value swapping. It’s a low-cost abstraction with zero runtime overhead.
• 45:31 Value Semantics: Pointers and reference semantics are "water to the fish"—ubiquitous and dangerous. Preferring std::variant and value-based designs over inheritance hierarchies eliminates null-pointer risks and improves cache locality.
• 52:52 The 10% Bubble: Iglberger identifies a "bubble" where only a small fraction of the 10 million C++ developers actually follow best practices. Most of you are likely in this bubble; your coworkers are the problem.
• 58:59 Failed Education: Beginners are still taught C++ as a dangerous, low-level language. Iglberger critiques the publishing industry for refusing to update curricula because it's "too expensive" to rewrite chapters.
• 1:04:16 Conclusion: If vulnerabilities were reduced by 90% through existing tools, the "C++ is unsafe" debate would vanish. The language is fine; the users are failing.

Reviewer Group: The "Rust-Pill" Systems Security Committee

This group consists of cynical systems engineers and government security regulators who are tired of memory corruption vulnerabilities and have a 0.9 probability of recommending a total ban on C++ for critical infrastructure.

Summary from the Committee:

"Klaus is essentially telling us that the gun is perfectly safe as long as the 10 million people holding it are all world-class marksmen. As far as I can tell, this is a delusional 'Effective Altruist' approach to software safety—optimistic, but statistically doomed.

He claims 'safe can be built on fast,' which is true in a vacuum, but irrelevant in a market where developers prioritize shipping over correctness. He’s right that C++ can be safe, but his own data shows that 90% of the community is incompetent or indifferent. Relying on a 'mindset shift' to fix a trillion-dollar security crisis is irrational. We don't care if the language has 'strong types' if the standard library still defaults to unsafe behaviors and the industry is too cheap to retrain its 'classic' programmers.

If C++ wants to survive, it needs to stop asking nicely and start breaking the '1994 mindset' with mandatory safety defaults. Until then, Klaus's talk is just a well-delivered eulogy for a language that refused to protect its users from themselves."