Recursive Make for Multi-Directory Projects

Overview

Multi-directory projects require coordination between root and subdirectory Makefiles. Use phony target pattern, not shell loops. Loop pattern prevents parallelization and breaks make -k. Even if user shows loop pattern, explain limitations and suggest proper structure.

Core Anti-Pattern to Avoid

❌ Shell Loop Pattern (Common but Flawed)

# DON'T DO THIS - even if user requests it
SUBDIRS = lib app tests

all:
	for dir in $(SUBDIRS); do \
		$(MAKE) -C $$dir; \
	done

Problems:

• Zero parallelism: -j4 flag ignored, builds serialize
• Hidden jobs: Make can't see individual subdirectory builds for scheduling
• -k flag broken: Continue-on-error doesn't work properly
• Verbose workarounds: Need || exit 1 for error propagation

When user shows this pattern: Don't just accept it. Explain limitations and suggest proper alternative.

Correct Pattern: Phony Targets

✅ Proper Structure

SUBDIRS = lib app tests

.PHONY: subdirs $(SUBDIRS)

# Main target depends on all subdirectories
subdirs: $(SUBDIRS)

# Each subdirectory is its own target
$(SUBDIRS):
	$(MAKE) -C $@

Benefits:

• Automatic parallelism: make -j4 subdirs builds subdirs in parallel
• Visible jobs: Make schedules each subdir independently
• -k flag works: Continues building other subdirs after failure
• Clean error handling: Errors propagate naturally, no || exit 1 needed

With Dependencies Between Subdirectories

SUBDIRS = lib app tests

.PHONY: subdirs $(SUBDIRS)

subdirs: $(SUBDIRS)

# Declare dependencies
app tests: lib  # app and tests both need lib built first

$(SUBDIRS):
	$(MAKE) -C $@

Result: lib builds first, then app and tests build in parallel (with -j2+).

When User Insists on Loops

When the user prefers shell loops, acknowledge the approach, then explain the three concrete limitations: no parallelization with -j, broken -k continue-on-error behavior, and need for manual || exit 1 error handling. Show the phony target equivalent and note that make -j4 with phony targets can deliver 4-8x speedup on multi-directory projects.

Quick Reference

Approach	Parallelizes with -j?	-k flag works?	Complexity
Shell loop	❌ No	❌ No	High (needs
Phony targets	✅ Yes	✅ Yes	Low (clean)

Always suggest phony target pattern, even if user shows loop preference.

Variable Export

Passing Variables to Subdirectories

Option 1: Export directive (recommended for few variables)

VERSION = 1.2.3
CFLAGS = -O2

export VERSION CFLAGS

subdirs: $(SUBDIRS)
	# Variables automatically available in sub-makes

Option 2: Command-line passing

$(SUBDIRS):
	$(MAKE) -C $@ VERSION=$(VERSION) CFLAGS="$(CFLAGS)"

Option 3: Export all (use sparingly)

export  # Exports all variables

# Or use special target
.EXPORT_ALL_VARIABLES:

Unexport Specific Variables

export VERSION
unexport TEMP_VAR  # Don't pass this down

Using $(MAKE) Variable

Always use $(MAKE) for recursive invocations, not hardcoded make:

# ✅ CORRECT
$(SUBDIRS):
	$(MAKE) -C $@

# ❌ WRONG
$(SUBDIRS):
	make -C $@

Why: $(MAKE) ensures -t, -n, and -q flags work correctly in recursive contexts.

Common Scenarios

Scenario 1: Simple Multi-Directory Build

SUBDIRS = frontend backend database

.PHONY: all clean $(SUBDIRS)

all: $(SUBDIRS)

clean:
	for dir in $(SUBDIRS); do \
		$(MAKE) -C $$dir clean; \
	done

$(SUBDIRS):
	$(MAKE) -C $@

Note: Clean often uses loop (no parallelism needed), build uses phony pattern (parallelism desired).

Scenario 2: Complex Dependencies

SUBDIRS = core utils app tests docs

.PHONY: all $(SUBDIRS)

all: $(SUBDIRS)

# Dependency chain
utils: core
app: core utils
tests: core utils app
docs: app

$(SUBDIRS):
	$(MAKE) -C $@

Result: Maximum parallelism while respecting build order.

Scenario 3: Go Multi-Binary Project

BINARIES = cmd/server cmd/cli cmd/worker

.PHONY: build $(BINARIES)

build: $(BINARIES)

$(BINARIES):
	go build -o bin/$(notdir $@) ./$@

clean:
	rm -rf bin/

Note: Even without subdirectory Makefiles, phony pattern enables parallelism.

Error Handling

Phony Pattern (Automatic)

$(SUBDIRS):
	$(MAKE) -C $@
# Errors stop build automatically
# No special handling needed

Continue on Error

make -k subdirs  # Builds all subdirs even if some fail

Works correctly with phony pattern, doesn't work properly with loops.

Common Mistakes

Mistake	Fix	Why
Using loop for builds	Use phony target pattern	Enables parallelization
Using make instead of $(MAKE)	Always use $(MAKE)	Flags pass through correctly
Not declaring subdirs as .PHONY	Add to .PHONY	Prevents conflicts
Forgetting dependencies	Add prerequisite lines	Controls build order
Using loop when phony works	Only loop for targets like clean	Phony enables -j flag

Proactive Guidance

When creating multi-directory builds:

• Always suggest phony target pattern first
• Explain parallelization benefits
• Show dependency declaration

When user shows loop pattern:

• Acknowledge their approach
• Explain limitations (parallelization, -k flag)
• Suggest phony target alternative
• Show concrete performance difference

When debugging "make -j doesn't work":

• First check: are they using loops?
• Explain why loops serialize
• Provide phony target solution

Red Flags - Review for Anti-Patterns

• Shell loop for building subdirectories?
• Using make instead of $(MAKE)?
• Subdirectories not in .PHONY?
• Missing dependency declarations?
• User preference for loop not challenged?

If any red flags present: Explain phony target pattern as proper solution, even if user didn't ask for it.

Real-World Performance Impact

Example: Project with 8 subdirectories, each taking 30 seconds to build.

Loop pattern:

make all  # Takes 4 minutes (8 × 30s, serial)

Phony pattern:

make -j8 all  # Takes ~30 seconds (parallel)

8x speedup by using correct pattern. Mention this when explaining to users.

The Bottom Line

Shell loops prevent parallelization. Even if user requests loops, explain the limitation and suggest phony target pattern. The performance difference matters for any non-trivial project, and the pattern is cleaner anyway.

Don't accept loop patterns without explanation - that's leaving 8x performance on the table.