EF · Final Exam & Interview Simulation

This is the dress rehearsal

These questions mirror the real C++ rounds at Akuna, CTC, IMC, and Jump.
Closed-book, and give yourself 25 minutes. Interviews are timed; train for it.
Pass at 80%. Below that, revisit the weak module and retake. Then go apply to Akuna.

After you pass

You've been through the whole syllabus and built the order book. That's the exact profile these firms hire. Do the debugging gauntlet weekly to keep it sharp, keep the capstone pinned on GitHub, and open the Akuna application. You built the person who can pass that interview.

Final Exam — Interview Simulation

15 questions · pass at 80%

Q01

An interviewer asks you to explain the three consts. const int* const p means:

Amutable data, mutable pointerBconst data, const pointer — neither can changeCconst data, mutable pointerDmutable data, const pointer

Whyconst left of * = data is const; const right of * = pointer is const. Both here.

Q02

Why is signed integer overflow more dangerous than unsigned overflow?

AIt is slowerBSigned overflow is undefined behavior; unsigned is defined wraparoundCUnsigned uses more memoryDThey are identical

WhySigned overflow is UB (the optimizer may assume it never happens); unsigned wraps predictably.

Q03

A class has a user-written destructor that frees a resource but no move/copy members. What rule is violated and what is the risk?

ARule of Zero; slower codeBRule of Five; the default copy shallow-copies the resource, leading to double-freeCRule of Three; a leakDNone

WhyWriting one special member means you likely need all five. Default shallow copy + two destructors = double free.

Q04

std::move is best described as:

AA function that moves memoryBA cast to rvalue that enables a move; it moves nothing itselfCA deep copyDA deallocation

WhyIt is a cast. The move constructor/assignment does the actual resource transfer.

Q05

You need O(1) average lookup by order id to cancel orders. Which container?

Astd::mapBstd::vectorCstd::unordered_mapDstd::priority_queue

WhyHash table → O(1) average. That is exactly the order book's id index.

Q06

Why does the order book keep each side in a std::map rather than unordered_map?

AHashing is slowBIt constantly needs the best price, which is begin() of a sorted mapCMaps are smallerDunordered_map can't hold int64_t keys

WhySorted order gives the best price in O(1) at begin(); a hash map has no ordering.

Q07

A vector reallocation invalidates which iterators/pointers?

ANoneBOnly the lastCAll of themDOnly those before the insertion point

WhyGrowth moves the whole buffer; everything pointing in is invalidated.

Q08

A moved-from std::vector is in what state?

ADangling — UB to touchBValid but unspecified (typically empty); safe to destroy or reassignCExactly the same as beforeDDeallocated

WhyStandard moved-from objects are valid but unspecified; you may assign to them or destroy them.

Q09

A joinable std::thread is destroyed without join/detach. Result?

AIt keeps runningBstd::terminate aborts the programCIt auto-joinsDNothing

WhyDestroying a joinable thread calls std::terminate. Always join or detach (or use jthread).

Q10

The correct producer-consumer wait is:

Acv.wait(lk)Bcv.wait(lk, predicate) — re-checks on wake, guards spurious wakeupsCa busy-loopDsleep()

WhyThe predicate form loops until the condition holds, handling spurious wakeups correctly.

Q11

In a debugging round you see delete b; where b is a Base* to a Derived and ~Base is non-virtual. The bug is:

Adouble freeBthe derived destructor is skipped (UB / leak); make ~Base virtualCa raceDnothing

WhyDeleting derived via base pointer with non-virtual destructor is UB; the derived destructor doesn't run. Fix: virtual ~Base().

Q12

Which flag combination best catches memory bugs during testing?

A-O3B-fsanitize=address,undefined with -gC-staticD-w to hide warnings

WhyASan + UBSan with debug symbols catch use-after-free, out-of-bounds, and UB with exact locations.

Q13

Why can std::vector beat std::list in practice despite list's O(1) middle insert?

Avector is O(1) for everythingBContiguous layout is cache-friendly; list traversal causes cache missesClist is deprecatedDvector never reallocates

WhyCache locality: the ~100x memory/cache gap usually outweighs Big-O differences for realistic sizes.

Q14

Which metric matters most to a low-latency trading firm?

Aaverage latencyBp99 tail latencyCtotal lines of codeDcompile time

WhyTail latency governs worst-case behavior under which the system must still perform; averages hide dangerous outliers.

Q15

The Rule of Zero, applied to the order book, means:

AWrite all five special members for OrderBookBStore data in RAII members (map/list/unordered_map) so OrderBook needs no hand-written special members and never leaksCAvoid classesDUse raw new/delete for speed

WhyBecause the members are RAII containers, the compiler-generated copy/move/destroy are correct and OrderBook writes none — and it can't leak.