Phase 0 — Refresh & Toolchain

Month 1 · Weeks W1–W4 · ~10–15 hours/week Start: 2026-07-06 · End of Phase: 2026-08-03

Mission

You have not written serious C++ in two years. Before you touch a single data structure, before you think about templates or move semantics or std::span, you need to get one thing back: the reflex of writing a small C++ program, compiling it, watching it fail, reading the error, and fixing it — without googling every semicolon.

That is the entire job of Month 1. Nothing more, nothing less. By 2026-08-03 you should be able to sit down cold on your MacBook, open a blank .cpp file, and produce a compiling, debuggable, warning-clean 200-line program in one sitting. If you can’t do that on August 3, do not move to Phase 1. Repeat the drills.

This phase is deliberately unglamorous. It’s not “learn modern C++.” It’s “make the tools disappear so modern C++ can happen.” Every phase after this assumes your fingers, your compiler flags, and your debugger workflow are automatic. Fix them here, once, or they will haunt you for twelve months.

What most people get wrong about returning to C++

Callout — read this twice.

The returning developer’s biggest failure mode is not forgotten syntax. It is overconfidence about how much they remember. You will read a std::unique_ptr example, think “yes, ownership, got it,” and skip the drill. Three weeks later you will write a raw new inside a constructor and leak memory in a class you designed yourself.

The second failure mode is AI dependence. You spent two years letting agents write code for you. Your fingers no longer know where the semicolons go. If you let Copilot autocomplete every line in M1, you will finish M1 knowing nothing you didn’t already know. Hard rule for this phase: no AI coding assistance. Not Copilot, not Claude, not Cursor tab. Hand-typed only. Details in 05_common_pitfalls_returning_dev.md.

The third: skipping the debugger and relying on std::cout printf-debugging like it’s 2005. You will fight this instinct all month. Win.

Exit criteria — Phase 0 done

You are done with Phase 0 when every one of these is true. Not “mostly true.” True.

  • brew, llvm@21 (or newer), cmake, ninja, git installed; clang++ --version reports the Homebrew LLVM version, not Apple Clang.

  • ~/.zshrc has the correct PATH export and you understand what it does.

  • You can compile a C++20 file with -std=c++20 -Wall -Wextra -Wpedantic -Werror -O2 -g and explain each flag from memory.

  • VSCode is configured with clangd (not the MS C/C++ extension) and CodeLLDB. Autocomplete, go-to-definition, and inline diagnostics all work in a fresh CMake project.

  • You can set a breakpoint in lldb (or the VSCode UI), step over/into/out, inspect a std::vector and a raw pointer, and print a backtrace — without notes.

  • You’ve completed all 30 drills in 03_syntax_refresh_drills.md. Every one compiles clean with -Werror.

  • P0.1 (Word Counter) matches Unix wc byte-for-byte on 5 test files.

  • P0.2 (JSON pretty-printer) produces byte-identical output to python -m json.tool on 3 flat JSON inputs.

  • P0.3 (Log Tail Filter) runs against a growing log file, filters by regex, exits cleanly on Ctrl-C.

  • You have read 05_common_pitfalls_returning_dev.md twice and can name at least five of its pitfalls without looking.

If any box is unchecked on 2026-08-03, spend the first week of August closing it before starting Phase 1.

Weekly breakdown

Each week is roughly 12 hours of focused work: 5 evenings at ~2 hours plus a longer Saturday block. Adjust to your calendar, but treat total hours as a floor, not a target.

W1 (2026-07-06 → 2026-07-12) — Toolchain and first fires

The week where you install nothing you don’t understand and compile everything with warnings on.

Day

Target

Deliverable

Mon–Tue

Work through 01_toolchain_setup.md. Install brew LLVM, cmake, ninja.

clang++ --version shows LLVM 21+

Wed

Work through 02_editor_and_debugger.md. VSCode + clangd + CodeLLDB.

Fresh CMake project builds & debugs

Thu–Fri

Drills 1–8 from 03_syntax_refresh_drills.md (I/O, strings, vectors).

8 .cpp files, all compile -Werror

Sat (long)

Drills 9–12 + set breakpoint, step through, inspect a std::vector.

Debugger workflow feels ordinary

Sun

Rest or slack day. Do not push through fatigue.

—

W2 (2026-07-13 → 2026-07-19) — Muscle memory + first project

Drills continue. First real deliverable ships this week.

Day

Target

Deliverable

Mon–Tue

Drills 13–20 (structured bindings, lambdas, optional, variant).

8 more compiling drills

Wed–Thu

Drills 21–24 (rule-of-zero, rule-of-five, constexpr, custom comparators).

4 drills; ownership begins to click

Fri

Start P0.1 — CLI Word Counter. Design pass; skeleton compiles.

main.cpp, CMakeLists.txt

Sat (long)

Finish P0.1. Match Unix wc on 5 files. Fix edge cases (empty file, unicode).

P0.1 acceptance passes

Sun

Write a short retro: what surprised you? Log it.

1-page notes file

W3 (2026-07-20 → 2026-07-26) — Parsing and streams

Move from primitive I/O to structured parsing. Second project ships.

Day

Target

Deliverable

Mon–Tue

Drills 25–28 (string_view, stringstream, error handling with expected).

4 drills

Wed

Start P0.2 — JSON pretty-printer. Restrict to flat objects/arrays.

Tokenizer skeleton

Thu–Fri

Complete P0.2. Byte-identical output vs python -m json.tool on 3 inputs.

P0.2 acceptance passes

Sat (long)

Debug session: intentionally break P0.2, use lldb to find the bug.

Debugger fluency confirmed

Sun

Rest.

—

W4 (2026-07-27 → 2026-08-03) — Systems edges and closeout

You touch signals, file descriptors, and the standard regex engine. Then you audit.

Day

Target

Deliverable

Mon

Drills 29–30 (signal handling with std::signal, std::regex).

2 drills

Tue–Wed

Start P0.3 — Log Tail Filter. tail -f behavior, regex flag.

Follows a growing file

Thu–Fri

Finish P0.3. Clean Ctrl-C exit. No leaked file descriptors (verify with lsof).

P0.3 acceptance passes

Sat (long)

Phase 0 audit: walk the exit criteria checklist. Fix any red boxes.

All boxes checked

Sun

Re-read 05_common_pitfalls_returning_dev.md. Write a one-page Phase 0 retro.

Retro committed

Files in this folder

Work through them in order. Do not skim 05_common_pitfalls_returning_dev.md — it’s short on purpose.

  1. 01_toolchain_setup.md — Apple Silicon toolchain, brew LLVM vs Apple Clang, ~/.zshrc, first hello.cpp, every compiler flag explained.

  2. 02_editor_and_debugger.md — VSCode + clangd + CodeLLDB (or CLion), .clangd config, compile_commands.json, debugger workflow.

  3. 03_syntax_refresh_drills.md — 30 targeted drills, 4 weekly batches, each with acceptance criterion.

  4. 04_first_projects.md — P0.1 Word Counter, P0.2 JSON pretty-printer, P0.3 Log Tail Filter.

  5. 05_common_pitfalls_returning_dev.md — The seven things you will get wrong. Read twice.

How to fail Phase 0

  • Install five extensions on day one, tune nothing, and never touch the debugger.

  • Skip drills because “I remember this.”

  • Ship P0.1 without matching wc — “close enough” is not the acceptance criterion.

  • Let Copilot autocomplete anything, ever, this month.

  • Reach for printf when a breakpoint would answer the question in five seconds.

If you catch yourself doing any of the above, stop, close the editor, and re-read the callout above.

How to succeed

Show up five evenings a week. Compile everything with -Werror. Read the compiler’s messages instead of pattern-matching them. When something surprises you, write down what surprised you before you fix it — that log is worth more than the fix. And when you finish a file, close the laptop; do not burn out in Month 1.

Return to Master README · Next: 01_toolchain_setup.md