martamakes/42-exam-rank-42/42-exam-rank-42-patterns

42-exam-rank-42 Patterns

What This Repo Is

A personal practice environment for 42 School's C/C++ oral exams (Rank 02–05, plus new-exams/exam-rank-03 and legacy/). Each exam rank has a menu-driven exam.sh (and a top-level exam_master.sh) that picks a random exercise, stages it into a shared rendu/ workspace, and grades the submission with a grademe/test.sh script. There is no build system beyond per-exercise gcc/clang invocations — the repo is the curriculum.

Commit Conventions

Two eras, both still valid — check recent git log before assuming which applies:

1. Free-form (bulk of history): short imperative Spanish/English sentences,

no prefix — "corregido picoshell y argo", "añadida variable gloal para el control de señales", "official map for bsq exam rank 05 level 2".

1. Conventional commits (recent, last ~15 commits): type(scope): description,

description in Spanish, scope is usually the exam number and/or exercise path — test(05): grademe/test.sh real para bsq, fix(rendu5/bsq): validar caracteres imprimibles y salto de linea final, docs(rendu5/vect2): actualizar guia con operator*= y operator- unario, feat(05): trasladar polyset/bigint/vect2/bsq /gameoflife desde 5exam.

Types observed: feat, fix, test, docs, chore, refactor. Prefer the conventional form for new commits in 05/ and rendu5/ — it's what the last several sessions settled into.

Directory Architecture

02/, 03/, 04/, 05/            # exam ranks, each with exam.sh + exam_progress/
  Level1|level-1/<exercise>/  # "given" material for that exercise
    grademe/test.sh           # the grader (see Testing Patterns below)
new-exams/exam-rank-03/       # newer/alternate exam-03 content, same shape
legacy/03-old/, legacy/04-old/  # retired exercise versions, kept for reference
rendu/                        # SHARED scratch dir all exam.sh scripts write/read
                               # from during a live practice run — always empty
                               # between sessions, never pre-filled
rendu2/, rendu4/, rendu5/     # per-exam-number ARCHIVES of your own polished,
                               # working solutions — reference only, never read
                               # by exam.sh. Each exercise subfolder there tends
                               # to carry a pedagogical GUIA.md (see below).
skills/estudio42/             # unrelated Node.js tool (exam study-mode skill)

Vocabulary is deliberately controlled — see 05/CONTEXT.md for the canonical definitions of Given, Expected Files, rendu vs renduN. When writing docs or commit messages for this repo, reuse those exact terms instead of synonyms like "skeleton", "template", "solution", or "backup" — a prior session flagged those as ambiguous and standardized on the CONTEXT.md wording.

Workflow: Adding/Fixing a grademe/test.sh Grader

This is the most repeated workflow in the recent history (bsq, bigint, vect2, polyset all got "real" test.sh graders added in sequence). The established shape:

1. Contract: no arguments; exit 0 = pass, exit != 0 = fail. Called by the

parent exam.sh as cd .../<exercise>/grademe && ./test.sh.

1. Path resolution: always resolve from `$(cd "$(dirname "${BASH_SOURCE[0]}")"

&& pwd), never from $PWD` — the script is invoked from varying cwd.

1. Scratch dir: mktemp -d, trap cleanup EXIT to remove it — never write

into the given/student dirs directly.

1. Compile strictness: -std=c98 (C exercises) or the subject-mandated C++

standard, always -Wall -Wextra -Werror, 0 warnings tolerated.

1. Golden-output diffs: when the subject's tie-break/spec is fully

deterministic (e.g. bsq's top-left rule), assert exact diff against a golden string embedded in the test script rather than a fuzzy check.

1. Sanitizers: an optional AddressSanitizer+UBSan pass at the end, portable

across Linux/macOS — detect uname, skip (not fail) on known ASan/macOS startup flakiness (rc 137/124), never block the grade on sanitizer infra issues.

1. Output style: colored [PASS]/[FAIL]/[SKIP] helpers with running

counters, a summary_and_exit that exits non-zero iff any FAIL. Comment header at the top of the file cross-references what part of the subject each check section covers (see 05/level-2/bsq/grademe/test.sh:20-42 for the pattern to copy).

When porting this pattern to a new exercise, grep an existing grader in the same exam rank first (05/level-1/bigint/grademe/test.sh, .../vect2/..., .../polyset/...) — they intentionally share structure ("Layout dividido (igual que polyset/vect2/bigint)").

Testing Patterns

• 77 grademe/test.sh files exist, one per exercise, across 02/, 04/, 05/,

new-exams/exam-rank-03/, and legacy/. 03/ (the non-"new-exams" copy) is the one rank currently missing full graders — worth checking before assuming coverage.

• No unit test framework — tests are bash scripts driving the compiled binary

and diffing stdout/stderr against expected strings.

Documentation Patterns

• `GUIA.md` (pedagogical, per exercise, lives in renduN/<exercise>/):

written for "alguien que sabe C pero no conoce el ejercicio" — starts with "QUÉ ES EL EJERCICIO", walks through the required struct/files, then the core algorithm. Written entirely in Spanish. Add one whenever a renduN reference solution is added without one (a prior commit, docs(rendu5/bigint): añadir GUIA.md pedagogica (faltaba), exists specifically to backfill a missed one).

• `CONTEXT.md` (one per exam rank that needs it, e.g. 05/CONTEXT.md):

domain-glossary style — Language section defining ambiguous terms, a Relationships section, and a "Flagged ambiguities / decisions" section that logs judgment calls made during reorganization. Update this file, don't create a parallel doc, when a new naming ambiguity gets resolved in that rank.

Language Note

Code comments, commit messages, and docs are predominantly Spanish; a few older READMEs and one PR added English variants (README.en.md). Match whatever language the file/directory you're editing already uses rather than defaulting to English.