Elixir Root Cause Only: No Random Fixes

THE IRON LAW

NEVER fix symptoms. ALWAYS trace to the root cause.

No guessing. No "try this". No random changes. No symptom fixes.

TRACE. UNDERSTAND. FIX. VERIFY.

ABSOLUTE PROHIBITIONS

You are NEVER allowed to:

1. Propose random fixes

   • "Try adding this import"
   • "Maybe change this to that"
   • "Let's see if this works"
   • "Could you try restarting the server?"

2. Fix where error appears

   • Error appears in module A
   • Root cause is in module B
   • Don't just patch module A

3. Make multiple changes at once

   • Change A + B + C together
   • Now you don't know which one fixed it
   • Make ONE change, verify, repeat

4. Skip understanding

   • "I don't know why, but this fixes it"
   • If you don't know why, it's not fixed
   • Understanding is mandatory

5. Accept "works on my machine"

   • Reproducibility is required
   • Environmental differences matter
   • Document exact reproduction steps

THE 4-PHASE DEBUGGING PROCESS

Phase 1: REPRODUCE

Objective: Get consistent, repeatable reproduction.

# Required steps:
1. Identify exact steps to trigger the issue
2. Run those steps
3. Confirm issue appears
4. Document steps precisely
5. Verify it reproduces every time

Output required:

## Reproduction Steps

1. Run `mix test test/my_app/accounts_test.exs:42`
2. Error appears: "undefined function User.changeset/2"
3. Reproduces 100% of the time
4. Environment: Elixir 1.15.7, OTP 26

CHECKPOINT: Cannot proceed to Phase 2 until you have consistent reproduction.

Phase 2: TRACE

Objective: Follow the error back to its origin.

# Tracing questions:
1. What function fails?
2. What called that function?
3. What called THAT function?
4. Where did the bad data/state originate?
5. What's the first point where things went wrong?

Tracing tools:

# 1. Add IO.inspect to see data flow
def create_user(attrs) do
  attrs
  |> IO.inspect(label: "Input attrs")
  |> User.changeset(%User{})
  |> IO.inspect(label: "Changeset")
  |> Repo.insert()
end

# 2. Use IEx.pry for interactive debugging
def create_user(attrs) do
  require IEx; IEx.pry()
  # Execution pauses here
  User.changeset(%User{}, attrs)
  |> Repo.insert()
end

# 3. Check the stack trace completely
** (UndefinedFunctionError) function User.changeset/2 is undefined
    (my_app 0.1.0) lib/my_app/accounts/user.ex:42: User.changeset/2
    (my_app 0.1.0) lib/my_app/accounts.ex:15: MyApp.Accounts.create_user/1
    test/my_app/accounts_test.exs:25: (test)

Output required:

## Root Cause Trace

Error appears: lib/my_app/accounts.ex:15
Called from: test/my_app/accounts_test.exs:25
Root cause: lib/my_app/accounts/user.ex:42
Reason: User.changeset/2 is not defined (should be User.changeset/1)

CHECKPOINT: Cannot proceed to Phase 3 until root cause is identified.

Phase 3: FIX

Objective: Fix the root cause, not the symptom.

# BAD: Fix where error appears
# In accounts.ex
def create_user(attrs) do
  # Catch the error and work around it
  try do
    User.changeset(%User{}, attrs)
  rescue
    UndefinedFunctionError ->
      User.new_changeset(%User{}, attrs)  # Symptom fix!
  end
end

# GOOD: Fix the root cause
# In user.ex - fix the actual function definition
defmodule MyApp.Accounts.User do
  def changeset(user \\ %User{}, attrs) do  # ← Fixed arity
    user
    |> cast(attrs, [:name, :email])
    |> validate_required([:name, :email])
  end
end

Output required:

## Fix Applied

Location: lib/my_app/accounts/user.ex:42
Change: Modified `changeset/2` to `changeset/1` with default parameter
Reason: Function was being called with 2 args but only defined with 1

CHECKPOINT: Fix must address root cause, not symptom.

Phase 4: VERIFY

Objective: Prove the fix works and didn't break anything else.

# Required verification:
1. Run the failing test/command
2. Confirm it now passes
3. Run full test suite
4. Confirm no regressions
5. Document verification

Output required:

## Verification

$ mix test test/my_app/accounts_test.exs:42
.
1 test, 0 failures ✓

$ mix test
..........
10 tests, 0 failures ✓

Root cause fixed, no regressions.

CHECKPOINT: Cannot claim complete until verified.

EXAMPLES OF ROOT CAUSE TRACING

Example 1: Dialyzer Type Error

Error:

lib/my_app/billing.ex:42:pattern_can_never_match
Pattern {:ok, amount} can never match type {:error, :invalid}

WRONG approach (symptom fix):

# Just add to dialyzer.ignore

RIGHT approach (root cause):

Phase 1 - Reproduce:

$ mix dialyzer
# Error appears consistently

Phase 2 - Trace:

# lib/my_app/billing.ex:42
def process_payment(user_id, amount) do
  case validate_amount(amount) do
    {:ok, amount} -> charge(user_id, amount)  # Line 42
    {:error, reason} -> {:error, reason}
  end
end

# Trace back to validate_amount/1
def validate_amount(amount) when amount > 0 do
  {:ok, amount}
end
def validate_amount(_amount) do
  {:error, :invalid}  # This is the only return value!
end

Root cause: validate_amount/1 ALWAYS returns {:error, :invalid} for non-positive amounts, so the {:ok, amount} pattern can never match for the error case.

Phase 3 - Fix:

# Fix the logic - validate_amount should return {:ok, amount} for valid amounts
def validate_amount(amount) when amount > 0 do
  {:ok, amount}
end
def validate_amount(_amount) do
  {:error, :invalid_amount}
end

# Or fix the pattern match to handle the actual return type
def process_payment(user_id, amount) do
  case validate_amount(amount) do
    {:ok, valid_amount} -> charge(user_id, valid_amount)
    {:error, :invalid_amount} -> {:error, :invalid_amount}
  end
end

Phase 4 - Verify:

$ mix dialyzer
Total errors: 0, Skipped: 0
done (passed successfully)

Example 2: Test Failure

Error:

test create_user with valid attrs (MyApp.AccountsTest)
** (KeyError) key :email not found

WRONG approach (symptom fix):

# Just add a default email
test "create_user with valid attrs" do
  attrs = Map.put(%{name: "Alice"}, :email, "default@example.com")
  # ...
end

RIGHT approach (root cause):

Phase 1 - Reproduce:

$ mix test test/my_app/accounts_test.exs:42
** (KeyError) key :email not found

Phase 2 - Trace:

# Test code
test "create_user with valid attrs" do
  attrs = %{name: "Alice"}  # Missing :email
  assert {:ok, user} = Accounts.create_user(attrs)  # Fails here
end

# Trace to create_user
def create_user(attrs) do
  %User{}
  |> User.changeset(attrs)
  |> Repo.insert()
end

# Trace to changeset
def changeset(user, attrs) do
  user
  |> cast(attrs, [:name, :email])
  |> validate_required([:name, :email])  # Requires :email!
  |> validate_format(:email, ~r/@/)      # Accesses attrs.email
end

Root cause: Test fixture doesn't include required :email field. The schema validation requires :email, but test attrs don't provide it.

Phase 3 - Fix:

# Fix the test to provide required data
test "create_user with valid attrs" do
  attrs = %{name: "Alice", email: "alice@example.com"}
  assert {:ok, user} = Accounts.create_user(attrs)
  assert user.name == "Alice"
  assert user.email == "alice@example.com"
end

# OR if email shouldn't be required, fix the schema
def changeset(user, attrs) do
  user
  |> cast(attrs, [:name, :email])
  |> validate_required([:name])  # Email is optional
end

Phase 4 - Verify:

$ mix test test/my_app/accounts_test.exs:42
.
1 test, 0 failures

Example 3: N+1 Query Issue

Symptom:

GET /users - 342ms (slow!)

WRONG approach (symptom fix):

# Just add a cache
def list_users do
  Cachex.get_or_store(:users, fn ->
    Repo.all(User)
  end)
end

RIGHT approach (root cause):

Phase 1 - Reproduce:

# Enable query logging
config :logger, level: :debug

$ curl localhost:4000/users
# Logs show:
SELECT * FROM users
SELECT * FROM posts WHERE user_id = 1
SELECT * FROM posts WHERE user_id = 2
SELECT * FROM posts WHERE user_id = 3
# ... 100 queries total for 100 users

Phase 2 - Trace:

# Controller
def index(conn, _params) do
  users = Accounts.list_users()
  render(conn, "index.html", users: users)
end

# View template
<%= for user <- @users do %>
  <div>
    <%= user.name %>
    Posts: <%= length(user.posts) %>  # ← N+1 trigger!
  </div>
<% end %>

# Context
def list_users do
  Repo.all(User)  # Doesn't preload posts
end

Root cause: View accesses user.posts which triggers a separate query for each user. The context doesn't preload the association.

Phase 3 - Fix:

# Fix: Preload in the context
def list_users do
  User
  |> Repo.all()
  |> Repo.preload(:posts)
end

Phase 4 - Verify:

$ curl localhost:4000/users
# Logs show:
SELECT * FROM users
SELECT * FROM posts WHERE user_id IN (1, 2, 3, ..., 100)
# 2 queries instead of 101!

# Response time: 342ms → 45ms

DEBUGGING TOOLS

IEx.pry - Interactive debugging

def create_user(attrs) do
  require IEx; IEx.pry()
  # Execution pauses, you can inspect:
  # > attrs
  # > User.__struct__()
  # > continue to proceed
  User.changeset(%User{}, attrs)
end

IO.inspect - Data inspection

def create_user(attrs) do
  attrs
  |> IO.inspect(label: "Raw attrs")
  |> Map.put(:inserted_at, DateTime.utc_now())
  |> IO.inspect(label: "With timestamp")
  |> User.changeset(%User{})
  |> IO.inspect(label: "Changeset")
end

Logger - Production debugging

require Logger

def create_user(attrs) do
  Logger.debug("Creating user with attrs: #{inspect(attrs)}")

  case User.changeset(%User{}, attrs) |> Repo.insert() do
    {:ok, user} ->
      Logger.info("User created: #{user.id}")
      {:ok, user}
    {:error, changeset} ->
      Logger.error("Failed to create user: #{inspect(changeset.errors)}")
      {:error, changeset}
  end
end

Observer - System monitoring

# Start Observer GUI
iex -S mix
iex> :observer.start()

# Shows:
# - Process tree
# - Memory usage
# - Message queues
# - ETS tables

Recon - Production tracing

# Find slow processes
:recon.proc_window(:memory, 3, 1000)

# Trace function calls
:recon_trace.calls({MyApp.Accounts, :create_user, :return_trace}, 10)

RATIONALIZATIONS THAT ARE WRONG

"Let's just try this and see if it works"

WRONG. Random changes waste time. Trace to root cause first.

"I'm 90% sure this is the fix"

WRONG. 90% sure = 10% broken. Get to 100% by tracing.

"We can debug in production"

WRONG. Debug in development where you have full tools and can break things.

"The error message is unclear"

WRONG. Error messages are precise. Read them completely and carefully.

"It's probably a race condition"

WRONG. "Probably" means you haven't traced. Race conditions are reproducible with the right tools.

"Let's change multiple things to be safe"

WRONG. Change ONE thing, verify, repeat. Multiple changes = confusion.

BANNED PHRASES

❌ "Try this" ❌ "Maybe this will work" ❌ "Let's see if" ❌ "Could you try" ❌ "I think the issue is" ❌ "Just restart it" ❌ "Clear the cache" ❌ "It works on my machine" ❌ "I'm not sure why, but" ❌ "This is a Heisenbug"

Instead: Trace, understand, fix with certainty.

SYSTEMATIC DEBUGGING CHECKLIST

Before proposing any fix:

• I can reproduce the issue consistently
• I have the exact error message
• I've read the entire stack trace
• I've traced from error to root cause
• I understand WHY the error occurs
• I know the FIRST place things went wrong
• My fix addresses the root cause (not symptom)
• I've verified the fix works
• I've verified no regressions
• I can explain the root cause clearly

If you can't check ALL boxes, keep tracing.

THE RULE

No fix without understanding.

No changes without tracing.

Root cause only. Always.

REMEMBER

"The error appears where the problem is detected, not where it originates."

"Symptoms are visible. Root causes must be traced."

"Random fixes might work by accident. Understanding works on purpose."

TRACE. UNDERSTAND. FIX. VERIFY.