Record Manager — Instructor Reference Solution (M1–M5)

PRIVATE — never ships to students. This is the reference implementation for the Record Manager cumulative project (assignments/projects/), milestone by milestone, plus the graded-interface contract and grading notes.

Layout

include/record.h   M1  record_t + record_new/free + status_t
include/buf.h      M2  dynamic string buffer (+ readline)
include/list.h     M3  opaque list_t + append/find/remove (+ len/get/sort helpers)
include/store.h    M4  store_save / store_load (text file, '|' delimiter)
src/record.c       M1  deep-copy record lifecycle
src/buf.c          M2  growable buffer: append/appendc/cstr/reset/free/readline
src/list.c         M3  singly linked list; remove frees node + record
src/store.c        M4  one record per line; load appends + reports count
src/main.c         M5  menu (add/list/search/delete/save/load/help/quit)
Makefile           builds recman + recman_asan; `make test` runs unit + smoke
tests/unit.c       direct tests of buf/list/store (links library sources)
tests/smoke.sh     end-to-end CLI session

Build + verify

make            # recman
make asan       # recman_asan (ASan+UBSan)
make test       # unit + smoke, both under ASan/UBSan

Reference must be -std=c17 -Wall -Wextra -Wpedantic -Werror clean and report 0 bytes lost under Valgrind.

Graded interface (what hidden tests touch)

record_new, record_free, list_new, list_free, list_append, list_find, list_remove, store_save, store_load — exact signatures in assignments/projects/README.md. This reference adds three non-graded helpers (list_len, list_get, list_sort) that the menu and store layer need; they are extra, not substitutes.

Contract decisions (documented for graders)

• list_remove: returns 0 on success, STATUS_ERR_NOTFOUND when no record matches, STATUS_ERR_ARG on NULL args. Frees the matching record + node.
• store_load: returns the number of records read (≥ 0) on success; negative status_t on error. "0 on success" from the interface table is satisfied when the file is empty/blank. It appends into the existing list.
• Delimiter |: none of the three flavors' fields (email/phone, assignment/ score, category/amount) can contain |; name may contain spaces — the delimiter is safe across all flavors.
• Empty fields: saved as empty strings; loaded back as empty strings (never NULL from store_load).

Milestone-by-milestone notes

M2 — dynamic string buffer (week 7)

Hidden tests likely: appends beyond initial capacity, buf_cstr NUL-termination after every append, reset-then-reuse, buf_readline on a long line (>256 chars), EOF/empty-line behavior. Sticky-OOM flag keeps the buffer usable after a failed realloc. Common mistake: forgetting the +1 for the NUL in buf_grow, or realloc-ing to a shrinking capacity.

M3 — linked list (week 11)

Hidden tests likely: append to empty list, find across many nodes, remove head / remove middle / remove missing, list_free of a 3-node list under Valgrind (0 lost), NULL-arg safety. Common mistakes: freeing the record and then walking next (order matters — save next first); list_remove freeing the node but not the record (double-free later); storing record_t * by value instead of pointer.

M4 — text file save/load (week 13)

Hidden tests likely: save→load round-trip preserves order + all three fields; load into a non-empty list appends; file with blank lines; missing file returns error, not crash; fields containing spaces survive (the | delimiter). Common mistakes: assuming fgets succeeds; writing with fprintf and never checking ferror; not stripping \n; strtok mangles empty fields (use strchr splitting instead — empty field1/field2 must survive).

M5 — refactor + search/sort + hardening (week 15)

Hidden tests likely: multi-file build with one make; case-insensitive substring search across all three fields; sort alphabetical by name; ASan/UBSan-clean; Valgrind 0 lost. Common mistakes: search matching only name; sort comparing record_t* pointers instead of ->name; -Werror violations from unused includes after the refactor.

Grading sanity checks

1. make from a clean tree (no build/ artifacts committed).
2. make asan && ./build/recman_asan → add 3 records → delete middle → save → quit → reload → list shows the surviving two in order.
3. valgrind --leak-check=full ./build/recman with the same session → 0 bytes lost (the menu leaks nothing).
4. Feed the tests/ fixtures: long input line (buf), 10k-record list (list), save/load with spaces and empties (store).