Chapter 7 — Teaching Notes

One-line purpose. Give students precise control over how data enters and leaves a program — formatted output to the column, list-directed convenience, files, the namelist configuration format, binary for scale, and error handling — and use it to make the heat solver runnable (config from a file) rather than recompilable.

Key ideas to emphasize

  • Predict output to the column. The single most durable skill of §7.1 is reading w.d and knowing the exact spaces before you compile. Drill it: put a print '(f8.2)', ... on the board and have the class call out the leading spaces. Everything (tables, the case studies, Ch. 12 internal files) builds on this.
  • Text is a lossy, expensive conversion. The §7.5 threshold — text throws away precision and costs ~100× a byte copy — reframes I/O as a performance topic, not a formatting chore. Case Study 1 is the payoff; teach it as "where did the nine minutes go?"
  • namelist is a genuine language-level win. Students coming from Python expect to pull in a YAML/JSON library; show them three lines of Fortran do it. Emphasize: by name, any order, omit-to-default, case-insensitive. This changes how they configure every run.
  • Guard your I/O. iostat/iomsg turn a crash into a message; iostat_end (never -1) ends a read loop portably. This is the seed of the Ch. 13 error-handling discipline — plant the pattern now.

Misconceptions to preempt

  • "print '(a)', x and print *, x are the same." (No — formatted has no leading space; list-directed emits one blank. Show both.)
  • "End of file is iostat == -1." (Processor-defined; use iostat_end. This is Find-the-Bug 7.12/the §7.6 🐛 callout — worth doing live.)
  • "A binary file records its own shape and type." (No — raw binary/stream is not self-describing; that is Ch. 25's whole reason to exist. Namelist echo/self-describing formats fill the gap.)
  • "recl is in bytes everywhere." (gfortran yes, Intel counts words by default — a real portability trap.)
  • "Asterisks mean bad data." (No — the field is too narrow; widen it. A too-narrow field is supposed to be loud.)
  • Integer division inside a format (real(1/3, dp) → 0.00) — the Ch. 3 trap resurfacing; Exercise 7.22.

A live demonstration (5–8 minutes)

Compile and run code/example-01-edit-descriptors.f90, but first have the class predict every line on paper. Then reveal the output and reconcile. Immediately follow with the "gotcha" pair: print '(a)', 'x' vs print *, 'x' — one has a leading space, one does not — to cement §7.1/§7.2. If time allows, run example-03-namelist-config.f90, then edit config.nml live (change nx, delete a line) and re-run without recompiling — the "aha" of configuration-not-recompilation lands hard.

Class-time budget (~50 min)

  • 12 min: edit descriptors (§7.1), with the predict-then-run demo. This is the load-bearing section.
  • 6 min: list-directed vs formatted (§7.2); the leading-space gotcha.
  • 10 min: files — open/read/write/close/inquire, newunit, status/action (§7.3).
  • 10 min: namelist (§7.4), with the live-edit demo. Most memorable section for scientists.
  • 6 min: unformatted/stream and why (§7.5) — the text-vs-binary threshold.
  • 6 min: iostat/iomsg, the EOF loop, and the Project Checkpoint (§7.6 + solver).

Prerequisites to review

  • Ch. 3 §3.7 (format strings — students have seen i0, f, es, a; this chapter formalizes them) and the integer-division trap (§3.4), which returns in Exercise 7.22.
  • Ch. 5 arrays/column-major (for whole-array writes and the contiguity of field(:,1) vs field(1,:)).
  • Ch. 6 intent and assumed-shape (for write_field(f(:,:), filename)).

Connections

Forward: Ch. 8 (read_config/write_field → heat_io module), Ch. 12 (internal files, deferred-length strings, indexed filenames), Ch. 13 (error handling generalized from iostat), Ch. 15 (stream ↔ NumPy order='F'), Ch. 25 (self-describing HDF5/NetCDF — the answer to "raw binary isn't self-describing"), Ch. 26 (VTK output), Ch. 28/37 (I/O as a profiled bottleneck; reproducibility). Naming these payoffs motivates the "boring" plumbing.