Self-Assessment Quiz: Error Handling and Debugging

Twenty questions to confirm you can guard, halt, debug, and defend. Aim for 16 or more. The answer key and a topic map are at the end — try the whole quiz first.

Question 1

On allocate(a(n), stat=s, errmsg=msg), the value of s after a successful allocation is: - A. -1 - B. 0 - C. the size n - D. processor-dependent and unknowable

Question 2

Without a stat= specifier, a failed allocate: - A. sets the array to zeros - B. aborts the program with a runtime error - C. silently continues with an unallocated array - D. returns a negative number you can test

Question 3

The difference between stop and error stop most relevant to a shell script or CI system is: - A. error stop prints in red - B. error stop produces a nonzero exit code (error termination); stop defaults to zero (success) - C. stop is faster - D. there is no difference

Question 4

A program ends with error stop 2. In a POSIX shell, echo $? afterward prints: - A. 0 - B. 1 - C. 2 - D. nothing

Question 5

True or false, with justification: "The exact integer that stat= receives on an allocation failure is fixed by the standard, so you may test it against a specific number."

Question 6

Which flag makes an out-of-bounds array access into a located runtime error instead of undefined behavior? - A. -O3 - B. -fcheck=all (or -fcheck=bounds) - C. -Wall only - D. -ffast-math

Question 7

-ffpe-trap=invalid,zero,overflow causes the program to: - A. round NaN to zero - B. halt at the operation that raises the exception, instead of producing NaN/Inf silently - C. disable floating-point arithmetic - D. print every floating-point result

Question 8

Why should you not include underflow in a -ffpe-trap list? - A. gfortran does not support it - B. underflow occurs routinely in correct numerical code, so trapping it would halt healthy programs - C. it would slow the program to a crawl - D. underflow cannot be trapped

Question 9

For finding a memory leak or a read of freed memory, the right tool is: - A. gdb - B. valgrind (or -fsanitize=address) - C. -Wall - D. print statements

Question 10

In gdb, the command that pauses execution at a chosen source line is: - A. run - B. print - C. break - D. continue

Question 11

Fortran has no built-in assert, so an assertion is: - A. impossible in Fortran - B. a hand-written procedure that error stops when its condition is false - C. the stop statement - D. a comment

Question 12

The best description of an assertion's role (vs stat/iostat) is: - A. it handles errors you expect from the outside world - B. it checks a condition you believe is always true, to catch a bug in your own logic - C. it replaces all I/O error handling - D. it is only for parallel code

Question 13

integer :: count = 0 inside a subroutine. On the second call to that subroutine, count at entry is: - A. 0 — reset every call - B. whatever it was at the end of the first call — the initializer implies save - C. undefined - D. a compile error

Question 14

A local variable you never assign before reading it holds: - A. zero, guaranteed - B. whatever bits were in that memory — undefined behavior - C. NaN, guaranteed - D. the value from the previous run

Question 15

A default 32-bit integer computing 2000*2000*2000 gives a negative number because: - A. of a compiler bug - B. the product exceeds ~2.1 billion and overflows, wrapping to a negative value - C. multiplication is undefined for large integers - D. it should use -O0

Question 16

The portable way to compare two reals for "equality" is: - A. a == b - B. abs(a - b) < tol for a suitable tolerance - C. a .eqv. b - D. print them and compare by eye

Question 17

True or false: "Diagnostics (error messages) should be written to standard output so they appear with the results."

Question 18

Which build is appropriate for developing the solver? - A. -O3 -march=native -flto - B. -fcheck=all -g -O0 (plus -fbacktrace) - C. -Ofast only - D. no flags at all

Question 19

What does this print?

integer(int64) :: n
n = int(huge(0_int32), int64) + 1_int64
print '(i0)', n
  • A. -2147483648
  • B. 2147483647
  • C. 2147483648
  • D. 0

Question 20

In a parallel (coarray) program, error stop: - A. halts only the image that executes it - B. halts every image immediately - C. is not allowed - D. behaves exactly like stop

Answer Key

Q Ans Why
1 B stat is 0 on success, nonzero on failure.
2 B Unguarded allocation failure aborts the program.
3 B error stop = nonzero exit code (failure); stop defaults to 0.
4 C The stop code becomes the exit status: 2.
5 False The failure value is processor-dependent; test only stat /= 0.
6 B -fcheck=all/-fcheck=bounds adds the bounds check.
7 B Trapping halts at the raising operation instead of producing NaN/Inf.
8 B Underflow is routine in correct code; trapping it halts healthy programs.
9 B valgrind (or AddressSanitizer) finds leaks and freed-memory reads.
10 C break sets a breakpoint.
11 B You write a procedure that error stops on a false condition.
12 B Assertions guard "impossible" conditions — bugs in your own logic.
13 B An initializer in a declaration confers an implicit save.
14 B Uninitialized locals hold garbage — undefined behavior.
15 B The product overflows 32-bit range and wraps negative.
16 B Compare reals with a tolerance, never ==.
17 False Diagnostics go to standard error (error_unit), not standard output.
18 B Development wants checks and debug info; strip them for production.
19 C 64-bit arithmetic does not overflow: 2147483648.
20 B error stop halts all images immediately.

Topics to review by question

  • Q1–2 → §13.1 (stat/errmsg, guarding allocate).
  • Q3–5, 20 → §13.2 (stop vs error stop, exit codes, parallel semantics).
  • Q6–8 → §13.3 (-fcheck, -ffpe-trap).
  • Q9–10 → §13.4 (gdb, valgrind).
  • Q11–12, 17 → §13.5 (assertions, defensive programming, error_unit).
  • Q13–16, 19 → §13.6 (implicit save, uninitialized, overflow, precision).
  • Q18 → §13.3 and the Project Checkpoint (development vs production builds).

Scored below 16? Reread the sections flagged above. Error handling is the one Part II topic no track should skip — a solver that fails silently is worse than one that fails loudly.