go-go-golems/esp32-s3-m5/esp-idf-sdkconfig-variant-builds

ESP-IDF sdkconfig variant builds

Use this skill when an ESP-IDF project needs multiple build variants and you must be sure a config flag really changed.

Core rule

If the project has a shared sdkconfig file, do not rely on -DSDKCONFIG_DEFAULTS=... alone for variant experiments.

Use all three:

1. A dedicated build directory per variant.
2. A dedicated build-local sdkconfig file via -DSDKCONFIG="$BUILD_DIR/sdkconfig.variant".
3. A base + variant defaults chain via -DSDKCONFIG_DEFAULTS="sdkconfig.defaults;sdkconfig.<variant>".

This is the repo-proven pattern that fixed the allocator-variant leakage in 0082-papers3-wamr-allocator-control.

Bad pattern

This can silently inherit choices from the project's shared sdkconfig:

idf.py -B build-variant \
  -DSDKCONFIG_DEFAULTS="sdkconfig.defaults;sdkconfig.internal_pool" \
  build

If a prior build or checked-in sdkconfig selected another choice item, this may not produce the variant you think it does.

Good pattern

PROJECT_DIR=/abs/project
BUILD_DIR="$PROJECT_DIR/build-internal-pool"
 
idf.py -C "$PROJECT_DIR" -B "$BUILD_DIR" \
  -DSDKCONFIG="$BUILD_DIR/sdkconfig.variant" \
  -DSDKCONFIG_DEFAULTS="sdkconfig.defaults;sdkconfig.internal_pool" \
  reconfigure build

The dedicated sdkconfig.variant isolates the build from the shared project sdkconfig.

Workflow

1. Define the Kconfig surface

• Put the variant choice in main/Kconfig.projbuild or the appropriate component Kconfig.
• Keep the choices explicit and mutually exclusive when possible.
• Add small defaults files such as:
• sdkconfig.defaults
• sdkconfig.headless
• sdkconfig.system_allocator
• sdkconfig.internal_pool

Example choice pattern:

choice PAPERS3_WAMR_ALLOCATOR_BACKING
    prompt "WAMR allocator backing"
    default PAPERS3_WAMR_ALLOCATOR_POOL_SPIRAM
 
config PAPERS3_WAMR_ALLOCATOR_POOL_SPIRAM
    bool "Pool allocator backed by external SPIRAM"
 
config PAPERS3_WAMR_ALLOCATOR_POOL_INTERNAL
    bool "Pool allocator backed by internal RAM"
 
config PAPERS3_WAMR_ALLOCATOR_SYSTEM
    bool "System allocator"
endchoice

2. Build each variant in its own build directory

• build-headless
• build-system-allocator
• build-internal-pool

Do not reuse one build directory for multiple variants unless you intentionally want that churn.

3. Force a build-local sdkconfig file

Always pass:

-DSDKCONFIG="$BUILD_DIR/sdkconfig.variant"

That is the critical guardrail against config leakage.

4. Chain defaults explicitly

Use:

-DSDKCONFIG_DEFAULTS="sdkconfig.defaults;sdkconfig.<variant>"

Notes:

• Keep sdkconfig.defaults first.
• Keep the variant file second so it overrides the base.
• Use semicolon-separated paths because CMake expects a list.

5. Reconfigure before build

Prefer:

idf.py ... reconfigure build

Use reconfigure when:

• you changed any sdkconfig.* file
• you changed Kconfig definitions
• you are switching variants
• you do not fully trust the current build cache

6. Verify generated config, not just source files

After configure, inspect the generated files inside the build directory:

• build-variant/config/sdkconfig.json
• build-variant/config/sdkconfig.cmake
• build-variant/config/sdkconfig.h

Use the bundled verifier:

skills/esp-idf-sdkconfig-variant-builds/scripts/check_variant_flags.sh \
  /abs/project/build-internal-pool \
  PAPERS3_WAMR_ALLOCATOR_POOL_INTERNAL \
  PAPERS3_WAMR_ALLOCATOR_SYSTEM

What to look for:

• sdkconfig.json should show the boolean you expect.
• sdkconfig.cmake should show the selected config as "y" and the opposite choice empty.
• sdkconfig.h should contain the active #define.

7. Add runtime proof

Generated config proof is necessary but not always sufficient. Add runtime logging or status output that surfaces the effective mode.

Examples:

• allocator=system-allocator
• allocator_backing=system
• host_api.display=disabled
• wamr.pool_buffer=0x0
• wamr.pool_size=0

If runtime status disagrees with the build-time files, trust the discrepancy and investigate before drawing conclusions.

Fast diagnosis checklist

When a variant result looks suspicious:

1. Confirm the command used a unique -B build directory.
2. Confirm it also used -DSDKCONFIG="$BUILD_DIR/sdkconfig.variant".
3. Inspect build-variant/config/sdkconfig.json.
4. Inspect build-variant/config/sdkconfig.cmake.
5. Inspect build-variant/config/sdkconfig.h.
6. Check runtime status strings/logs.
7. Only then interpret the experiment.

Repo anchors

Open these first in this repo:

• 0082-papers3-wamr-allocator-control/main/Kconfig.projbuild
• 0082-papers3-wamr-allocator-control/sdkconfig.defaults
• 0082-papers3-wamr-allocator-control/sdkconfig.system_allocator
• 0082-papers3-wamr-allocator-control/sdkconfig.internal_pool
• ttmp/2026/03/23/ESP-42-PAPERS3-WAMR-ALLOCATOR-CONTROL--papers3-minimal-wamr-allocator-control-firmware-to-isolate-instantiate-vs-psram-contamination/scripts/flash_and_probe_allocator_system.sh
• ttmp/2026/03/23/ESP-42-PAPERS3-WAMR-ALLOCATOR-CONTROL--papers3-minimal-wamr-allocator-control-firmware-to-isolate-instantiate-vs-psram-contamination/scripts/flash_and_probe_allocator_internal.sh

These scripts encode the corrected variant-build pattern.

Pseudocode

for each variant:
  choose BUILD_DIR unique to variant
  choose VARIANT_DEFAULTS file
 
  run idf.py with:
    -B BUILD_DIR
    -DSDKCONFIG=BUILD_DIR/sdkconfig.variant
    -DSDKCONFIG_DEFAULTS=sdkconfig.defaults;VARIANT_DEFAULTS
    reconfigure build
 
  verify generated config files in BUILD_DIR/config/
  run firmware
  verify runtime status matches intended variant