# Anti-Patterns Reference

## Memory & Performance

```elixir
# WRONG: length/1 for empty check (O(n))
length(list) == 0

# RIGHT: Pattern match or Enum.empty?
list == []
Enum.empty?(list)

# WRONG: ++ to append (O(n))
list ++ [item]

# RIGHT: Prepend and reverse, or use different structure
[item | list] |> Enum.reverse()

# WRONG: Dynamic atom creation (memory leak - atoms aren't GC'd)
String.to_atom(user_input)

# RIGHT: Explicit mapping or existing atoms
defp status_atom("ok"), do: :ok
defp status_atom("error"), do: :error
# Or: String.to_existing_atom(input)

# WRONG: Sending unnecessary data (copies entire vars between processes)
spawn(fn -> log_ip(conn.remote_ip) end)  # Copies entire conn!
GenServer.cast(pid, {:process, large_struct.id})  # Copies entire struct!

# RIGHT: Extract minimal data before spawning or sending
ip = conn.remote_ip
spawn(fn -> log_ip(ip) end)
id = large_struct.id
GenServer.cast(pid, {:process, id})
```

## Message Handling

```elixir
# WRONG: Selective receive without reference (O(n) mailbox scan)
receive do
  {:response, data} -> data  # Scans entire mailbox
end

# RIGHT: Reference-based (compiler optimizes)
ref = make_ref()
send(server, {self(), ref, :request})
receive do
  {^ref, response} -> response  # Compiler uses receive marker
end
```

## Code Organization

```elixir
# WRONG: String keys internally
%{"name" => value}

# RIGHT: Atom keys internally
%{name: value}

# WRONG: Macro when function works
defmacro sum(a, b), do: quote do: unquote(a) + unquote(b)

# RIGHT: Just use a function
def sum(a, b), do: a + b
```

## OTP Anti-Patterns

```elixir
# ANTI-PATTERN: GenServer for stateless computation
def add(a, b), do: GenServer.call(__MODULE__, {:add, a, b})
def handle_call({:add, a, b}, _from, state), do: {:reply, a + b, state}

# CORRECT: Just use functions
def add(a, b), do: a + b

# ANTI-PATTERN: Single GenServer bottleneck
# All requests serialize through one process

# CORRECT: Use ETS for reads, GenServer for writes
# Or partition into multiple processes
```

## Assertiveness (from official Elixir anti-patterns)

```elixir
# WRONG: Non-assertive map access — nil on missing required key
user[:email]  # Returns nil silently if :email missing

# RIGHT: Assert required keys exist
user.email    # Raises KeyError — fail fast
# Use [:key] ONLY for truly optional keys
config[:timeout] || 5000

# WRONG: Catch-all hides bugs
case fetch_user(id) do
  {:ok, user} -> process(user)
  _ -> :error  # What failed? Why?
end

# RIGHT: Match known cases explicitly
case fetch_user(id) do
  {:ok, user} -> process(user)
  {:error, :not_found} -> {:error, :not_found}
end

# WRONG: Boolean obsession — multiple related booleans
%{is_admin: true, is_editor: false, is_viewer: false}

# RIGHT: Single atom field
%{role: :admin}
# Or enum-like pattern in schema:
# field :role, Ecto.Enum, values: [:admin, :editor, :viewer]
```

## Stream vs Enum

```elixir
# Stream processes lazily—only computes what's needed
1..1_000_000
|> Stream.map(&(&1 * 3))
|> Stream.filter(&(rem(&1, 2) != 0))
|> Enum.take(5)  # Only processes ~5 elements

# Enum processes eagerly—entire collection each step
1..1_000_000
|> Enum.map(&(&1 * 3))      # Creates 1M list
|> Enum.filter(&(rem(&1, 2) != 0))  # Creates another list
|> Enum.take(5)
```

**Use Enum** for small/medium collections, immediate results.
**Use Stream** for large collections, multiple transformations, memory constraints.

## Pipe Operator Misuse

```elixir
# AVOID: Pipe with single step
user |> do_something()  # Just: do_something(user)

# AVOID: Start with function call
String.upcase("hello") |> String.split()  # Start with "hello"

# DO: Use tap/1 for side effects (returns original value)
user
|> validate()
|> tap(&Logger.info("Validated: #{&1.name}"))  # Returns user
|> persist()

# DO: Use then/1 for transformations
user
|> validate()
|> persist()
|> then(&{:ok, &1})  # Transforms to tagged tuple
```

## Binary Handling

```elixir
# ANTI-PATTERN: Small sub-binary keeps large parent alive
<<small::binary-size(100), _::binary>> = one_gb_binary

# DO: Copy if keeping only the small part
small = :binary.copy(small)
```

## Tail Recursion

For tail call optimization, recursive call must be the **last operation**:

```elixir
# Tail recursive (optimized - constant stack)
def sum(list), do: do_sum(list, 0)
defp do_sum([], acc), do: acc
defp do_sum([head | tail], acc), do: do_sum(tail, head + acc)

# Not tail recursive (builds stack - O(n) memory)
def factorial(0), do: 1
def factorial(n), do: n * factorial(n - 1)  # Multiplication after recursion
```

**Rule of thumb**: Use Enum for 95% of cases—cleaner and well-tested.

# Elixir 1.18 Features Reference

> **Official changelog**: <https://github.com/elixir-lang/elixir/blob/main/CHANGELOG.md>
> **HexDocs**: <https://hexdocs.pm/elixir/> — Use `hexdocs-fetcher` for latest API docs.

## Duration Module (Elixir 1.18+)

Native duration representation without external dependencies.

### Creating Durations

```elixir
# From keyword list
Duration.new!(hour: 2, minute: 30)
#=> %Duration{hour: 2, minute: 30}

# Common units
Duration.new!(day: 7)
Duration.new!(week: 1)
Duration.new!(second: 3600)

# Negative durations
Duration.new!(hour: -1)
```

### Duration Arithmetic

```elixir
# Adding durations
Duration.add(Duration.new!(hour: 1), Duration.new!(minute: 30))
#=> %Duration{hour: 1, minute: 30}

# Adding to DateTime/NaiveDateTime
DateTime.add(DateTime.utc_now(), Duration.new!(hour: 24))

# Subtracting
DateTime.add(DateTime.utc_now(), Duration.negate(Duration.new!(day: 7)))
```

### Duration in Oban Jobs

```elixir
# Schedule job for specific duration from now
defmodule MyApp.Workers.ReminderWorker do
  use Oban.Worker

  def schedule_reminder(user_id, delay_duration) do
    scheduled_at = DateTime.add(DateTime.utc_now(), delay_duration)

    %{user_id: user_id}
    |> new(scheduled_at: scheduled_at)
    |> Oban.insert()
  end
end

# Usage
ReminderWorker.schedule_reminder(user.id, Duration.new!(day: 3))
```

### Duration in Cache TTLs

```elixir
# With Cachex or similar
def get_user_with_cache(user_id) do
  Cachex.fetch(:users_cache, user_id,
    ttl: Duration.to_milliseconds(Duration.new!(hour: 1))
  )
end

# Convert to different units
Duration.to_seconds(Duration.new!(hour: 2))
#=> 7200

Duration.to_milliseconds(Duration.new!(minute: 5))
#=> 300_000
```

### Anti-patterns

```elixir
# AVOID: Magic numbers for time
Process.send_after(self(), :timeout, 3600_000)  # What unit? Confusing!

# PREFER: Duration makes intent clear
Process.send_after(self(), :timeout,
  Duration.to_milliseconds(Duration.new!(hour: 1)))

# AVOID: Manual arithmetic
scheduled_at = DateTime.add(now, 7 * 24 * 60 * 60, :second)

# PREFER: Readable duration
scheduled_at = DateTime.add(now, Duration.new!(week: 1))
```

## Enhanced dbg/2 (Elixir 1.18+)

### Pipeline Debugging

```elixir
# dbg shows each pipeline step
users
|> Enum.filter(&(&1.active))
|> dbg()  # Shows filter result
|> Enum.map(&(&1.name))
|> dbg()  # Shows map result
|> Enum.sort()
```

### Customizing Output

```elixir
# In IEx or tests, customize dbg behavior
# config/dev.exs
config :elixir, :dbg_callback, {MyApp.Debug, :custom_dbg, []}

# Custom module
defmodule MyApp.Debug do
  def custom_dbg(code, options, env) do
    # Custom formatting, logging, etc.
    Macro.dbg(code, options, env)
  end
end
```

## Calendar.strftime/2 Improvements

```elixir
# ISO week numbers
Calendar.strftime(~D[2026-02-05], "%G-W%V")
#=> "2026-W06"

# 12-hour format with AM/PM
Calendar.strftime(~T[14:30:00], "%I:%M %p")
#=> "02:30 PM"

# Full datetime formatting
Calendar.strftime(DateTime.utc_now(), "%Y-%m-%d %H:%M:%S %Z")
#=> "2026-02-05 10:30:45 UTC"
```

## Deprecations in 1.18

### `unless/2` Deprecated

`unless` is soft-deprecated in Elixir 1.18. The formatter
automatically rewrites `unless condition` to `if !condition`.

```elixir
# Deprecated — formatter will rewrite
unless valid?(input) do
  {:error, :invalid}
end

# Current — what the formatter produces
if !valid?(input) do
  {:error, :invalid}
end
```

**Rule**: Never write `unless` — always use `if !` or
pattern match instead. This avoids unnecessary formatter churn.

## Compatibility Notes

- Duration requires Elixir 1.18+
- For projects on 1.17 or earlier, use Timex or manual arithmetic
- Check version in mix.exs: `{:elixir, "~> 1.18"}`

# Elixir 1.20 Type System Reference

> **Blog**: <https://elixir-lang.org/blog/2026/06/03/elixir-v1-20-0-released/>
> **Changelog**: <https://elixir.hexdocs.pm/1.20.0/changelog.html#type-system-improvements>
> **HexDocs**: use `hexdocs-fetcher` for the latest API docs.

## TL;DR

Elixir 1.20 (released 2026-06-03) completed its **first type-system
milestone**: the compiler now infers types and gradually type-checks **every**
program **without any annotations**. It reports **dead/redundant code** and
**verified bugs** — typing violations guaranteed to fail at runtime. Low false
positives by design.

**Requires OTP 27+.** No new struct/typespec syntax — that is a *future*
milestone. There is **nothing new to write**; this release is about
*interpreting new compiler diagnostics*.

## The one thing that changes day-to-day

Type violations are emitted as **`mix compile` warnings** — built into the
compiler. **Not** Dialyzer, **not** a separate tool, **no PLT**.

> On Elixir 1.20+, `mix compile --warnings-as-errors` **now fails the build on
> type violations.** This is what `phx-verify`, `phx-work` checkpoints, and
> the "fix CI" pattern run everywhere — so a verified bug becomes a hard
> failure, not a silent warning.

When a `--warnings-as-errors` build that previously passed starts failing after
a toolchain bump to 1.20, **suspect a newly-detected type violation before
assuming the code regressed** — the code didn't change, the checker got smarter.

## The `dynamic()` mental model

Unlike `any()` in most gradual languages ("anything goes, never flag"),
Elixir's `dynamic()` is a **refinable range**:

- **Compatibility** — a call is flagged **only when the supplied and accepted
  types are disjoint**. `dynamic(integer() or binary())` passed to `/` (wants a
  number) is fine — `integer()` overlaps. Passed to `Map.fetch!/2` (wants a
  map) it is flagged — disjoint.
- **Narrowing** — usage refines the type. `data.a + data.b` narrows `data` to
  `%{..., a: number(), b: number()}` (the `...` = "may have other keys").

```elixir
# value_or_error :: dynamic(integer() or binary()) at runtime
Map.fetch!(value_or_error, :some_key)  # ⚠ violation: map ⟂ int|binary

def add_a_and_b(data) do
  data.a + data   # ⚠ data narrowed to %{..., a: number()}, then used AS a number
end
```

## What the checker catches

| Kind | Example that warns |
|------|--------------------|
| **Verified bug** | calling `User.name(%{})` when it needs `%{..., name: term()}` |
| **Disjoint call** | `String.upcase/1` on a value the checker proved is an integer |
| **Dead clause** | a `case` clause that can never match given prior clauses |
| **Bad field access** | `x.foo` after a guard proved `x` has **no** `:foo` key |
| **Out-of-bounds** | `elem(x, 3)` after `tuple_size(x) < 3` |

## Inference you get for free

**Guards** infer unions/intersections/negations:

```elixir
def f(x, y) when is_list(x) and is_integer(y)        # x :: list, y :: integer
def f({:ok, x} = y) when is_binary(x) or is_integer(x)
def f(x) when is_map_key(x, :foo)        # x :: %{..., foo: dynamic()}
def f(x) when not is_map_key(x, :foo)    # x :: %{..., foo: not_set()} → x.foo warns
def f(x) when tuple_size(x) < 3          # elem(x, 3) warns
```

**Clauses / occurrence typing** (`case`, `cond`, `with`) — earlier clauses
refine later ones:

```elixir
case System.get_env("SOME_VAR") do
  nil   -> :not_found
  value -> {:ok, String.upcase(value)}   # value :: binary() (nil already excluded)
end
```

**Maps** track non-atom keys by domain and typed stdlib ops:

```elixir
%{123 => "hello", 456.0 => :ok}   # %{integer() => binary(), float() => :ok}
Map.put(map, :key, 123)           # %{..., key: integer()}
Map.delete(map, :key)             # %{..., key: not_set()}
Map.replace(map, :key, 123)       # %{..., key: if_set(integer())}
```

## How to read & fix a type violation

1. **Read the message literally** — it states the *accepted* vs *supplied*
   type. The fix is to make them overlap, not to silence the warning.
2. **Check the narrowing chain** — the warning usually points at where the
   variable was *refined* (a guard, a prior clause, a field access), not just
   where it blew up.
3. **It is almost always a real bug.** False positives are rare by design
   (disjoint-only). Prefer fixing the code over restructuring to dodge the
   checker.
4. **Genuinely dynamic boundary?** If a value really is runtime-typed
   (external input), the disjoint-only rule already tolerates it — only a
   provably-impossible call warns.

## Compiler type checker vs Dialyzer

They are **different tools** — keep both in mind:

| | Compiler type checker (1.20+) | Dialyzer / dialyxir |
|--|------------------------------|----------------------|
| Runs in | `mix compile` (built-in) | `mix dialyzer` (separate, needs PLT) |
| Setup | none | PLT build (slow first run) |
| Basis | set-theoretic types, `dynamic()` | success typing |
| Annotations | none needed | reads `@spec` |
| Caught by `--warnings-as-errors` | **yes** | no |
| Best at | disjoint calls, dead clauses, map/guard narrowing | `@spec` mismatches, contract violations, opaque misuse |

Guidance: the compiler checker is now the **first line** of type safety and is
always on. Dialyzer remains valuable for `@spec`/contract checking and opaque
types — it is **complementary, not redundant**.

## Bonus: faster compiles

New compiler option `:module_definition` (`:compiled` default, or
`:interpreted`) can speed up large-project builds. Set in `mix.exs`:

```elixir
elixirc_options: [module_definition: :interpreted]
```

Does not change emitted `.beam` files — only how `defmodule` bodies execute at
compile time (slight stacktrace-precision tradeoff).

## Compatibility notes

- Type checking ships in Elixir **1.20+** on **OTP 27+** (compatible to OTP 29).
- No code changes required to benefit — recompile and read the warnings.
- No annotation syntax yet; typed structs and signatures are the next milestone.
- Check version in `mix.exs`: `{:elixir, "~> 1.20"}`.

# Error Handling Reference

## Decision Tree

```
Is failure expected in normal operation?
├─ Yes → {:ok, _}/{:error, _} tuples
│   ├─ Chaining operations? → with
│   └─ Need both variants? → Provide foo/foo!
└─ No (unexpected/bug) → raise exception
    └─ Supervision handles recovery
```

## When to Use Each

| Use Exceptions | Use Error Tuples |
|----------------|------------------|
| Programming bugs | Expected failures |
| Truly unexpected errors | User input validation |
| Can't recover gracefully | Caller should handle |
| "Let it crash" scenarios | Control flow decisions |

## Tagged Tuple Pattern

```elixir
# Return error tuples for expected failures
def divide(a, b) when b != 0, do: {:ok, a / b}
def divide(_, 0), do: {:error, :division_by_zero}

# Bang variant raises for callers who want to crash
def divide!(a, b) do
  case divide(a, b) do
    {:ok, result} -> result
    {:error, reason} -> raise ArgumentError, "Cannot divide: #{reason}"
  end
end
```

## With for Happy Path

```elixir
# PREFER: with for multi-step operations
def create_order(params) do
  with {:ok, user} <- get_user(params.user_id),
       {:ok, product} <- get_product(params.product_id),
       {:ok, order} <- Orders.create(user, product) do
    {:ok, order}
  end
end

# AVOID: with for single operation
with {:ok, user} <- get_user(id), do: user  # Just use case!

# AVOID: Complex else clauses - normalize errors in helpers
with {:ok, a} <- normalize_step1(),
     {:ok, b} <- normalize_step2(a) do
  {:ok, b}
end
# Non-matching tuples pass through unchanged
```

## Assertive Map Access

```elixir
# DON'T: Silent nil for missing required keys
{point[:x], point[:y]}  # Returns {nil, nil} if keys missing!

# DO: Use .key for required keys (raises KeyError if missing)
{point.x, point.y}

# DO: Pattern match
def plot(%{x: x, y: y}), do: {x, y}  # Match fails if keys missing
```

## Rescue Only for External Code

```elixir
# DO: Rescue external library exceptions
def safe_parse(json) do
  {:ok, Jason.decode!(json)}
rescue
  e in Jason.DecodeError -> {:error, e.message}
end

# DON'T: Rescue for control flow or catch all
try do
  risky_operation()
rescue
  _ -> :error  # Never do this - masks programming errors
end
```

## Control Flow Decision Tree

```
Need to match against patterns?
├─ Yes → case
│   └─ Multiple dependent operations? → with
└─ No (boolean conditions)
    ├─ Single condition? → if
    └─ Multiple conditions? → cond
```

| Construct | Use When |
|-----------|----------|
| Function heads | First choice - most idiomatic |
| `case` | Pattern matching single value |
| `cond` | Multiple boolean conditions |
| `with` | Chaining operations returning tagged tuples |
| `if` | Single boolean check |

# Mix Task Patterns

> **Official docs**: <https://hexdocs.pm/mix/Mix.Task.html>
> **Mix guides**: <https://github.com/elixir-lang/elixir/tree/main/lib/mix/lib/mix>

## Module Naming Convention

Mix task module names map directly to the CLI command:

```elixir
# mix my_app.validate → Mix.Tasks.MyApp.Validate
defmodule Mix.Tasks.MyApp.Validate do
  @shortdoc "Validate configuration"
  @moduledoc "Detailed description..."
  use Mix.Task

  @impl Mix.Task
  def run(args) do
    # Parse args, do work
  end
end
```

**Rules:**

- Module name segments map to `.`-separated CLI words
- `CamelCase` in module → `snake_case` in CLI
- `@shortdoc` is REQUIRED (shows in `mix help`)
- `@moduledoc` for detailed `mix help my_app.validate`

## Option Parsing

```elixir
@impl Mix.Task
def run(args) do
  {opts, _rest, _invalid} =
    OptionParser.parse(args,
      strict: [
        dry_run: :boolean,
        type: :string,
        format: :string,
        verbose: :boolean
      ],
      aliases: [d: :dry_run, t: :type, f: :format, v: :verbose]
    )

  # Access with Keyword.get
  dry_run? = Keyword.get(opts, :dry_run, false)
  format = Keyword.get(opts, :format, "text")
end
```

## Shell Output

```elixir
# Prefer Mix.shell() for testability
Mix.shell().info("Processing #{count} items...")
Mix.shell().error("Failed: #{reason}")

# For colored output
Mix.shell().info([:green, "✓ ", :reset, "All checks passed"])

# Progress reporting
Enum.each(items, fn item ->
  Mix.shell().info("  #{item.name}... #{status}")
end)
```

## App Startup (Iron Law #10)

NEVER `Mix.Task.run("app.start")` — it boots the FULL supervision tree:
the endpoint binds its port and Oban starts consuming jobs, inside what
should be a one-off task. Start only what the task needs:

```elixir
def run(args) do
  # Load config without starting the app
  Mix.Task.run("app.config")

  # Start only the dependency apps + repo this task uses
  {:ok, _} = Application.ensure_all_started(:ecto_sql)
  {:ok, _} = MyApp.Repo.start_link()

  # Your logic here
end
```

## Chaining Tasks

```elixir
# Run another task (after the startup above)
Mix.Task.run("ecto.migrate")

# Re-run a task that already ran this VM session
Mix.Task.rerun("ecto.migrate")
```

## Credo Complexity

Mix tasks often trigger Credo complexity warnings because the
`run/1` function handles arg parsing + logic. Split into:

```elixir
def run(args) do
  args |> parse_opts() |> validate_opts() |> execute()
end

defp parse_opts(args), do: ...
defp validate_opts(opts), do: ...
defp execute(opts), do: ...
```

## Testing Mix Tasks

```elixir
defmodule Mix.Tasks.MyApp.ValidateTest do
  use ExUnit.Case, async: true

  test "runs successfully with valid args" do
    Mix.Tasks.MyApp.Validate.run(["--type", "full"])
  end

  test "handles missing args gracefully" do
    assert_raise Mix.Error, fn ->
      Mix.Tasks.MyApp.Validate.run(["--invalid"])
    end
  end
end
```

# OTP Patterns Reference

> **Official docs**: <https://hexdocs.pm/elixir/GenServer.html> | <https://hexdocs.pm/elixir/Supervisor.html>
> **Elixir guides**: <https://hexdocs.pm/elixir/introduction.html> (see OTP section)

## Contents

- [Core Rule](#core-rule-no-process-without-a-runtime-reason)
- [BEAM Architecture Context](#beam-architecture-context)
- [Decision Tree](#decision-tree)
- [Quick Reference Table](#quick-reference-table)
- [Plain Functions](#plain-functions-default-choice)
- [Agent: Simple State](#agent-simple-state)
- [ETS](#ets-shared-state-without-serialization)
- [GenServer](#genserver-complex-state-management)
- [Task](#task-one-off-async-work)
- [Supervisor](#supervisor-fault-tolerance)
- [DynamicSupervisor](#dynamicsupervisor-on-demand-children)
- [Registry](#registry-dynamic-process-naming)
- [Common Scenarios](#common-scenarios)
- [Resource Cleanup](#resource-cleanup-tryafter-not-tryrescue)
- [Anti-Patterns](#anti-patterns)

## Core Rule: NO PROCESS WITHOUT A RUNTIME REASON

Processes model **runtime properties**, not code organization:

- ✓ Concurrency needs
- ✓ Shared resources requiring serialized access
- ✓ Error isolation domains
- ✓ State that survives between operations
- ✗ Code organization (MAJOR ANTI-PATTERN)
- ✗ Stateless computation
- ✗ Namespacing

## BEAM Architecture Context

Understanding these fundamentals explains WHY patterns exist:

- **Processes are cheap**: 2.6KB each, ~134M possible per VM
- **Complete isolation**: Each has own stack/heap/mailbox
- **Messages are copied**: Keep messages small (except binaries >64 bytes)
- **Per-process GC**: No global GC pauses
- **Preemptive scheduling**: Fair CPU time via reductions
- **"Let it crash"**: Focus on happy path, supervisors restart to known-good state

## Decision Tree

```
Need to maintain state?
├─ No → Use plain functions
└─ Yes
    ├─ Simple get/update only? → Agent or ETS
    ├─ Complex message handling? → GenServer
    │   ├─ Need timeouts/monitors? → GenServer
    │   └─ Children started dynamically? → DynamicSupervisor
    └─ One-off async work? → Task
```

## Quick Reference Table

| Need | Solution | Notes |
|------|----------|-------|
| Stateless computation | Functions | Default choice |
| Simple get/set state | Agent | No monitors/timers |
| Fast key-value lookups | ETS | Many readers, no serialization |
| Complex state/coordination | GenServer | Monitors, timers, handle_info |
| One-off async work | Task | Task.Supervisor for production |
| Dynamic worker pool | DynamicSupervisor + Registry | Per-user/session processes |
| Connection pool | GenServer | Checkout/checkin with monitors |
| Fault tolerance | Supervisor | Always supervise! |

---

## Plain Functions (Default Choice)

```elixir
# DO: Stateless computation
defmodule Calculator do
  def add(a, b), do: a + b
  def multiply(a, b), do: a * b
end

# DON'T: GenServer for stateless work
defmodule Calculator do
  use GenServer
  def add(a, b), do: GenServer.call(__MODULE__, {:add, a, b})
  def handle_call({:add, a, b}, _from, state), do: {:reply, a + b, state}
end
```

## Agent: Simple State

**Use when**: Only get/update operations, no monitors/timers

**Don't use when**: Need handle_info, monitors, distributed system

```elixir
defmodule Counter do
  use Agent

  def start_link(initial) do
    Agent.start_link(fn -> initial end, name: __MODULE__)
  end

  def value, do: Agent.get(__MODULE__, & &1)
  def increment, do: Agent.update(__MODULE__, &(&1 + 1))
  def reset, do: Agent.update(__MODULE__, fn _ -> 0 end)
end
```

## ETS: Shared State Without Serialization

**Use when**: Many concurrent readers/writers, key-value pairs, performance critical

**Don't use when**: Need complex coordination, complex relationships

```elixir
# Create table (usually in Application.start/2)
:ets.new(:my_cache, [:named_table, :public, read_concurrency: true])

# Use from anywhere
:ets.insert(:my_cache, {:key, value})
[{:key, value}] = :ets.lookup(:my_cache, :key)
:ets.delete(:my_cache, :key)
```

## GenServer: Complex State Management

**Use for**: Complex coordination, serializing access, managing external resources, monitors/timers

**Don't use for**: Code organization, stateless computation, simple get/update

```elixir
defmodule ConnectionPool do
  use GenServer

  # Client API
  def start_link(opts) do
    GenServer.start_link(__MODULE__, opts, name: __MODULE__)
  end

  def checkout, do: GenServer.call(__MODULE__, :checkout)
  def checkin(conn), do: GenServer.cast(__MODULE__, {:checkin, conn})

  # Server callbacks
  @impl GenServer
  def init(opts) do
    {:ok, %{available: [], checked_out: %{}, max: opts[:max] || 10}}
  end

  @impl GenServer
  def handle_call(:checkout, {pid, _ref}, state) do
    ref = Process.monitor(pid)  # Clean up if caller crashes
    # ... checkout logic
    {:reply, conn, updated_state}
  end

  @impl GenServer
  def handle_info({:DOWN, _ref, :process, pid, _reason}, state) do
    # Clean up crashed process's connections
    {:noreply, cleanup_for_pid(state, pid)}
  end
end
```

### Expensive Initialization

```elixir
# DO: Use handle_continue (OTP 21+)
@impl GenServer
def init(args) do
  {:ok, initial_state, {:continue, :load_data}}
end

@impl GenServer
def handle_continue(:load_data, state) do
  # GUARANTEED to run before any messages
  {:noreply, %{state | data: load_expensive_data()}}
end

# DON'T: send(self(), :init) - not guaranteed order
```

## Task: One-Off Async Work

```elixir
# Fire-and-forget (won't crash caller)
Task.Supervisor.start_child(MyApp.TaskSupervisor, fn ->
  do_background_work()
end)

# Async with timeout and error handling
task = Task.Supervisor.async_nolink(MyApp.TaskSupervisor, fn ->
  might_fail()
end)

case Task.yield(task, 5000) || Task.shutdown(task) do
  {:ok, result} -> result
  {:exit, reason} -> handle_error(reason)
  nil -> handle_timeout()
end

# Concurrent collection processing
urls
|> Task.async_stream(&fetch_url/1, max_concurrency: 10, timeout: 30_000)
|> Enum.map(fn {:ok, result} -> result end)
```

## Supervisor: Fault Tolerance

**Always supervise processes**. Never use `{:ok, pid} = GenServer.start_link(...)` in production.

```elixir
children = [
  {Registry, keys: :unique, name: MyApp.Registry},
  {MyApp.CacheServer, []},
  {DynamicSupervisor, name: MyApp.WorkerSupervisor, strategy: :one_for_one},
  {Task.Supervisor, name: MyApp.TaskSupervisor}
]

Supervisor.start_link(children,
  strategy: :one_for_one,
  max_restarts: 3,
  max_seconds: 5
)
```

### Restart Strategies

| Strategy | Behavior | Use When |
|----------|----------|----------|
| `:one_for_one` | Restart only crashed child | Children are independent |
| `:one_for_all` | Restart all children | Children are tightly coupled |
| `:rest_for_one` | Restart crashed + those started after | Later depend on earlier |

### Restart Types

- `:permanent` - Always restart (default for most workers)
- `:temporary` - Never restart (one-off tasks)
- `:transient` - Restart only on abnormal exit

## DynamicSupervisor: On-Demand Children

```elixir
# Combined with Registry for named access
def start_worker(user_id) do
  name = {:via, Registry, {MyApp.WorkerRegistry, user_id}}
  spec = {Worker, name: name, user_id: user_id}
  DynamicSupervisor.start_child(MyApp.WorkerSupervisor, spec)
end

def find_worker(user_id) do
  case Registry.lookup(MyApp.WorkerRegistry, user_id) do
    [{pid, _}] -> {:ok, pid}
    [] -> {:error, :not_found}
  end
end
```

## Registry: Dynamic Process Naming

```elixir
# Unique registration (no atom explosion)
name = {:via, Registry, {MyApp.Registry, "user:#{user_id}"}}
GenServer.start_link(UserWorker, args, name: name)

# Lookup
case Registry.lookup(MyApp.Registry, "user:#{user_id}") do
  [{pid, _}] -> {:ok, pid}
  [] -> {:error, :not_found}
end

# Pub/sub broadcast (duplicate keys registry)
Registry.dispatch(PubSubRegistry, "topic", fn entries ->
  for {pid, _} <- entries, do: send(pid, {:broadcast, message})
end)
```

---

## Common Scenarios

### Cache

**Pattern**: ETS (not GenServer)
**Why**: Many concurrent readers/writers, no coordination needed

### Rate Limiting

**Pattern**: ETS + optional GenServer for cleanup
**Why**: ETS for fast lookups, GenServer only if need periodic cleanup

### Background Job

**Pattern**: Task.Supervisor (or Oban for persistence)
**Why**: Don't need long-lived state, just async execution

### Connection Pool

**Pattern**: GenServer
**Why**: Need serialization, monitors, complex coordination

### User Session State

**Pattern**: DynamicSupervisor + Registry + GenServer
**Why**: Dynamic per-user processes, fault isolation, named access

---

## Resource Cleanup: try/after (Not try/rescue)

When wrapping code in instrumentation spans, telemetry, or any
resource that needs guaranteed cleanup:

```elixir
# CORRECT: try/after — runs ALWAYS
def instrumented_call(args) do
  span = Tracer.start_span("operation")
  try do
    do_work(args)
  after
    Tracer.finish_span(span)
  end
end

# WRONG: try/rescue — only runs on exception
def instrumented_call(args) do
  span = Tracer.start_span("operation")
  try do
    do_work(args)
  rescue
    e -> Tracer.finish_span(span); reraise e, __STACKTRACE__
  end
  Tracer.finish_span(span)  # Missed if do_work throws non-exception exit
end
```

**Rule**: `try/rescue` only catches exceptions. `try/after`
runs unconditionally — use it for spans, file handles, locks,
and any cleanup that must happen regardless of outcome.

## Anti-Patterns

### 1. GenServer for Code Organization

```elixir
# ANTI-PATTERN
defmodule Calculator do
  use GenServer
  def add(a, b), do: GenServer.call(__MODULE__, {:add, a, b})
end

# CORRECT
defmodule Calculator do
  def add(a, b), do: a + b
end
```

### 2. Scattered Process Interfaces

```elixir
# ANTI-PATTERN: Calling Agent from multiple modules
Agent.get(MyApp.Cache, &Map.get(&1, key))  # In various modules

# CORRECT: Encapsulate in single module
defmodule MyApp.Cache do
  use Agent
  def get(key), do: Agent.get(__MODULE__, &Map.get(&1, key))
end
```

### 3. Sending Unnecessary Data in Closures

```elixir
# ANTI-PATTERN: Captures entire struct
spawn(fn -> log_ip(conn.remote_ip) end)  # Copies entire conn!

# CORRECT: Extract before spawning
ip = conn.remote_ip
spawn(fn -> log_ip(ip) end)
```

### 4. Unsupervised Processes

```elixir
# ANTI-PATTERN
{:ok, pid} = GenServer.start_link(MyServer, [])

# CORRECT
children = [{MyServer, []}]
Supervisor.start_link(children, strategy: :one_for_one)
```

### 5. Global Singleton Bottleneck

```elixir
# ANTI-PATTERN: Global singleton for per-user data
GenServer.call(UserStateServer, {:get_user, user_id})

# CORRECT: DynamicSupervisor + Registry for per-user processes
name = {:via, Registry, {MyApp.Registry, "user:#{user_id}"}}
```

# Pattern Matching Reference

## Function Heads (Preferred)

```elixir
def process(%{status: :active} = user), do: activate(user)
def process(%{status: :inactive} = user), do: deactivate(user)
def process(_user), do: {:error, :unknown_status}
```

## Pin Operator

Match against existing values:

```elixir
expected = :ok
^expected = get_result()  # Only matches if returns :ok

# Essential for dynamic map keys
key = :user_id
%{^key => value} = data
```

## is_non_struct_map/1 Guard (Elixir 1.17+)

Structs ARE maps, so `is_map/1` matches both:

```elixir
# DO: Distinguish plain maps from structs
def process(data) when is_non_struct_map(data), do: handle_map(data)
def process(%User{} = user), do: handle_user(user)

# DON'T: This matches structs too!
def process(%{} = data), do: handle_data(data)  # Matches User struct!
```

## Custom Guards

```elixir
defguard is_positive_integer(n) when is_integer(n) and n > 0

def process(n) when is_positive_integer(n), do: n * 2
```

## Binary Pattern Matching

```elixir
# UTF-8 prefix matching
"hello " <> rest = "hello world"  # rest = "world"

# Protocol parsing
<<header::8, length::32-big, payload::binary-size(length), rest::binary>> = data

# ANTI-PATTERN: Small sub-binary keeps large parent alive
<<small::binary-size(100), _::binary>> = one_gb_binary
# DO: Copy if keeping only the small part
small = :binary.copy(small)
```

## Guards: Allowed Operations

Guards must be pure, deterministic:

- Type checks: `is_atom/1`, `is_binary/1`, `is_integer/1`, `is_list/1`, `is_map/1`
- Comparisons: `==`, `!=`, `<`, `>`, `<=`, `>=`
- Arithmetic: `+`, `-`, `*`, `/`, `abs/1`, `div/2`, `rem/2`
- Value access: `hd/1`, `tl/1`, `elem/2`, `tuple_size/1`, `map_size/1`, `length/1`

Guards use `and`/`or`/`not`, never short-circuit operators (they require boolean operands)

```elixir
# CORRECT
def process(n) when is_integer(n) and n > 0, do: n * 2

# WRONG - compile error
def process(n) when is_integer(n) && n > 0, do: n * 2
```

# BEAM Troubleshooting Playbook

Production debugging for Phoenix/BEAM applications. For code bugs, see `deep-bug-investigator` agent.

## Quick Diagnosis

| Symptom | Likely Cause | Section |
|---------|--------------|---------|
| High memory, growing | Process leak, ETS bloat | Memory Issues |
| Slow responses | N+1, GenServer bottleneck | Performance |
| Random crashes | Unhandled errors, supervisor | Crashes |
| Timeouts | DB pool, GenServer call | Timeouts |
| Node unresponsive | Scheduler block, long GC | BEAM Issues |

## Memory Issues

### With Tidewave

```elixir
# Top memory processes
mcp__tidewave__project_eval """
Process.list()
|> Enum.map(fn pid -> {pid, Process.info(pid, :memory)} end)
|> Enum.filter(fn {_, mem} -> mem != nil end)
|> Enum.map(fn {pid, {:memory, mem}} -> {pid, mem} end)
|> Enum.sort_by(&elem(&1, 1), :desc)
|> Enum.take(10)
|> Enum.map(fn {pid, mem} ->
  info = Process.info(pid, [:registered_name, :current_function])
  {info[:registered_name] || pid, div(mem, 1024), info[:current_function]}
end)
"""

# ETS table sizes
mcp__tidewave__project_eval """
:ets.all()
|> Enum.map(fn t -> {t, :ets.info(t, :memory) * :erlang.system_info(:wordsize)} end)
|> Enum.sort_by(&elem(&1, 1), :desc)
|> Enum.take(10)
|> Enum.map(fn {t, mem} -> {t, div(mem, 1024)} end)
"""
```

### Without Tidewave

```bash
# Attach to running node
iex --sname debug --remsh myapp@hostname

# In IEx
:observer.start()
```

### Common Causes

1. **LiveView socket assigns** - Use `temporary_assigns` or streams
2. **ETS table growth** - Check for missing cleanup
3. **Process mailbox** - GenServer not keeping up
4. **Binary memory** - Large binaries not GC'd (use `:binary.copy/1`)

## Performance Issues

### N+1 Query Detection

```elixir
# config/dev.exs - log slow queries
config :my_app, MyApp.Repo,
  log: :debug,
  stacktrace: true
```

### GenServer Bottleneck

```elixir
mcp__tidewave__project_eval """
pid = Process.whereis(MyApp.SomeServer)
{:message_queue_len, len} = Process.info(pid, :message_queue_len)
IO.puts("Queue length: \#{len}")  # >100 = problem
"""
```

**Signs:**

- Message queue growing
- Call timeouts
- Single process high CPU

**Solutions:**

- Pool of workers (Poolboy)
- ETS for read-heavy (GenServer writes, ETS reads)
- Partition by key (Registry + DynamicSupervisor)

### Slow Ecto Queries

```sql
-- With Tidewave
mcp__tidewave__execute_sql_query """
SELECT query, calls, mean_time, total_time
FROM pg_stat_statements
ORDER BY mean_time DESC
LIMIT 10
"""

-- Missing indexes
mcp__tidewave__execute_sql_query """
SELECT relname, seq_scan, idx_scan
FROM pg_stat_user_tables
WHERE seq_scan > idx_scan
ORDER BY seq_scan DESC
"""
```

## Crashes

### Common Crash Patterns

| Error | Cause | Fix |
|-------|-------|-----|
| `noproc` | Process died/not started | Check supervisor, handle `{:error, _}` |
| `timeout` | GenServer.call timeout | Increase timeout or use cast |
| `badmatch` | Pattern match failed | Add catch-all clause |
| `function_clause` | No matching function | Check guards, add fallback |
| `badarg` | Wrong argument type | Validate input |

### Supervisor Tree Analysis

```elixir
mcp__tidewave__project_eval """
Supervisor.which_children(MyApp.Supervisor)
|> Enum.map(fn {id, pid, type, _} ->
  %{id: id, pid: pid, type: type}
end)
"""
```

## Timeouts

### DB Pool Exhaustion

**Symptoms:** Random timeouts, "connection not available"

**Check:**

```elixir
# Current pool size
Application.get_env(:my_app, MyApp.Repo)[:pool_size]
```

**Solutions:**

- Increase `pool_size`
- Add `queue_target` and `queue_interval`
- Find long-running queries
- Use `Repo.checkout/1` properly for transactions

### GenServer Call Timeout

```elixir
# Default is 5000ms
GenServer.call(pid, :request, 10_000)  # Increase if needed

# Or use cast + handle_info for async
GenServer.cast(pid, {:request, self()})
```

## BEAM Issues

### Scheduler Utilization

```elixir
mcp__tidewave__project_eval """
:scheduler.utilization(1000)  # Sample for 1 second
"""
```

### Long GC Pauses

```elixir
# Enable GC logging
:erlang.system_flag(:long_gc, 50)  # Log GC > 50ms
```

### Dirty Schedulers Blocked

```elixir
mcp__tidewave__project_eval """
:erlang.statistics(:dirty_cpu_run_queue_lengths)
"""
```

## Quick Checklist

```markdown
## Troubleshooting: [Issue]

### Symptoms
- [ ] High memory?
- [ ] Slow responses?
- [ ] Timeouts?
- [ ] Crashes?

### Checked
- [ ] Logs: `mcp__tidewave__get_logs level: :error`
- [ ] Memory: Top processes
- [ ] Queries: Slow query log
- [ ] Processes: GenServer queue lengths

### Root Cause
[...]

### Fix
[...]
```

# With Statement and Pipe Operator Guide

## Contents

- [Why These Are Idiomatic](#why-these-are-idiomatic)
- [Pipe Operator](#pipe-operator-)
- [With Statement](#with-statement)
- [Real-World Examples from Production Code](#real-world-examples-from-production-code)
- [Summary](#summary)
- [Anti-Pattern: Avoiding Pipes and With](#anti-pattern-avoiding-pipes-and-with)

## Why These Are Idiomatic

Both `with` and `|>` are core Elixir idioms that experienced developers expect. Avoiding them leads to less readable, non-idiomatic code.

## Pipe Operator `|>`

### When to Use Pipes

```elixir
# ✅ Data transformation chains
params
|> Map.get("user")
|> normalize_email()
|> String.downcase()
|> String.trim()

# ✅ Query building
User
|> where(active: true)
|> where([u], u.role in ^roles)
|> order_by(desc: :inserted_at)
|> limit(10)
|> Repo.all()

# ✅ Changeset chains
%User{}
|> User.changeset(params)
|> put_change(:status, :pending)
|> validate_required([:email])
|> Repo.insert()

# ✅ Stream processing
1..1000
|> Stream.map(&expensive_operation/1)
|> Stream.filter(&valid?/1)
|> Enum.take(10)
```

### When NOT to Use Pipes

```elixir
# ❌ Single function call - no pipe needed
name |> String.upcase()
# ✅ Just call the function
String.upcase(name)

# ❌ When data isn't the first argument
Enum.map(list, &String.upcase/1) |> Enum.join(", ")
# ✅ Use variable or reorder
list
|> Enum.map(&String.upcase/1)
|> Enum.join(", ")

# ❌ Complex branching mid-pipe
data
|> transform()
|> (fn x -> if condition, do: a(x), else: b(x) end).()  # Ugly!
# ✅ Use case or separate function
data
|> transform()
|> handle_condition(condition)

defp handle_condition(x, true), do: a(x)
defp handle_condition(x, false), do: b(x)
```

### Pipe Style Rules

```elixir
# Start with data, not a function call
# ❌ Wrong
get_user(id)
|> process()

# ✅ Right
id
|> get_user()
|> process()

# Or just use variables for clarity
user = get_user(id)
process(user)
```

## With Statement

### When to Use With

```elixir
# ✅ Multiple dependent operations that can fail
def create_order(params) do
  with {:ok, user} <- get_user(params.user_id),
       {:ok, product} <- get_product(params.product_id),
       :ok <- check_inventory(product),
       {:ok, order} <- Orders.create(user, product, params) do
    {:ok, order}
  end
end

# ✅ Authorization chains
def update_post(user, post_id, params) do
  with {:ok, post} <- get_post(post_id),
       :ok <- authorize(user, :update, post),
       {:ok, updated} <- Posts.update(post, params) do
    {:ok, updated}
  end
end

# ✅ Validations that depend on each other
def process_upload(params) do
  with {:ok, file} <- validate_file_exists(params),
       {:ok, metadata} <- extract_metadata(file),
       :ok <- validate_file_type(metadata),
       {:ok, processed} <- process_file(file, metadata) do
    {:ok, processed}
  end
end
```

### When NOT to Use With

```elixir
# ❌ Single operation - use case instead
with {:ok, user} <- get_user(id) do
  {:ok, user}
end

# ✅ Just use case
case get_user(id) do
  {:ok, user} -> {:ok, user}
  {:error, _} = error -> error
end

# ❌ No failure handling needed - use pipes
with data <- fetch_data(),
     processed <- process(data),
     formatted <- format(processed) do
  formatted
end

# ✅ Use pipes
fetch_data()
|> process()
|> format()
```

### With Patterns

```elixir
# Pattern match in with clauses
with %User{active: true} = user <- get_user(id),
     %Subscription{status: :active} <- get_subscription(user) do
  {:ok, user}
else
  %User{active: false} -> {:error, :user_inactive}
  %Subscription{} -> {:error, :subscription_inactive}
  nil -> {:error, :not_found}
end

# Bare expressions (always match)
with {:ok, user} <- get_user(id),
     # This always succeeds, just assigns
     email = user.email,
     {:ok, _} <- send_notification(email) do
  :ok
end
```

### Handling Else Clauses

**Official anti-pattern**: Complex `else` in `with` — keep else to 1-2 clauses max. Normalize errors in helpers instead.

```elixir
# ✅ Simple: let non-matches pass through
def create_user(params) do
  with {:ok, validated} <- validate(params),
       {:ok, user} <- Repo.insert(validated) do
    {:ok, user}
  end
  # {:error, changeset} passes through unchanged
end

# ✅ When you need to transform errors
def create_user(params) do
  with {:ok, validated} <- validate(params),
       {:ok, user} <- Repo.insert(validated) do
    {:ok, user}
  else
    {:error, %Ecto.Changeset{} = changeset} ->
      {:error, :validation_failed, changeset}
    {:error, reason} ->
      {:error, :creation_failed, reason}
  end
end

# ❌ Avoid complex else clauses - normalize in helpers
# Instead of many else branches, make helpers return consistent errors
defp validate(params) do
  case Validator.validate(params) do
    :ok -> {:ok, params}
    {:error, reasons} -> {:error, {:validation, reasons}}
  end
end
```

## Real-World Examples from Production Code

### Context Function with Authorization

```elixir
def update_deal(broker, deal_id, params) do
  with {:ok, deal} <- get_deal(deal_id),
       :ok <- Bodyguard.permit(Deal, :update_deal, broker, deal),
       {:ok, updated} <- do_update_deal(deal, params) do
    broadcast_update(updated)
    {:ok, updated}
  end
end
```

### Query Building

```elixir
def list_articles(profile_id, opts \\ []) do
  Article
  |> where(profile_id: ^profile_id)
  |> filter_by_search(opts[:search])
  |> filter_by_status(opts[:status])
  |> filter_by_rating(opts[:rating])
  |> apply_sort(opts[:sort_by], opts[:sort_order])
  |> maybe_limit(opts[:limit])
  |> Repo.all()
end
```

### Changeset Operations

```elixir
def confirm_guest(guest, params) do
  guest
  |> Guest.confirm_changeset(params)
  |> maybe_update_attendees(params)
  |> Repo.update()
end
```

### Multi-Step Billing Check

```elixir
def check_feature_access(user, feature) do
  with true <- billing_enforced?(),
       {:ok, subscription} <- get_active_subscription(user),
       true <- feature_included?(subscription, feature) do
    :ok
  else
    false -> {:error, :feature_not_available}
    {:error, :no_subscription} -> {:error, :subscription_required}
  end
end
```

## Summary

| Pattern | Use When | Avoid When |
|---------|----------|------------|
| `\|>` pipe | Data transformation, query building, changesets | Single calls, complex branching |
| `with` | Multiple dependent fallible operations | Single operation, no failures possible |
| `case` | Pattern matching single value | Chaining multiple operations |
| Nested `case` | Never | Always - use `with` instead |

## Anti-Pattern: Avoiding Pipes and With

```elixir
# ❌ This is NOT more readable
result1 = function1(data)
result2 = function2(result1)
result3 = function3(result2)
final = function4(result3)

# ✅ Pipes are clearer for transformations
final =
  data
  |> function1()
  |> function2()
  |> function3()
  |> function4()

# ❌ Nested case is hard to follow
case get_user(id) do
  {:ok, user} ->
    case get_subscription(user) do
      {:ok, sub} ->
        case check_feature(sub, feature) do
          :ok -> {:ok, user}
          error -> error
        end
      error -> error
    end
  error -> error
end

# ✅ With flattens the happy path
with {:ok, user} <- get_user(id),
     {:ok, sub} <- get_subscription(user),
     :ok <- check_feature(sub, feature) do
  {:ok, user}
end
```