Migrate raster:bands -> bands fails with earth_search S2_MSI_L2A_COG

[remi-braun]

Describe the bug
Migrate raster:bands -> bands fails with earth_search provider and S2_MSI_L2A_COG collection

Code To Reproduce

import os
from eodag import EODataAccessGateway, setup_logging

os.environ["EODAG__EARTH_SEARCH__AUTH__CREDENTIALS__AWS_ACCESS_KEY_ID"] = "xxx"
os.environ["EODAG__EARTH_SEARCH__AUTH__CREDENTIALS__AWS_SECRET_ACCESS_KEY"] = "yyy"
os.environ["EODAG_PROVIDERS_WHITELIST"] = "earth_search"

if __name__ == "__main__":
    setup_logging(3)  # DEBUG level

    # Create gateway
    dag = EODataAccessGateway()
    dag.set_preferred_provider("earth_search")

    # Query all
    products = dag.search_all(
        start='2025-02-08T00:00:00',
        end='2025-02-11T00:00:00',
        **{
            "collection": "S2_MSI_L2A_COG",
            "grid:code": 'MGRS-32ULU'
        }
    )

Output

2026-04-30 14:57:56,374 eodag.config                     [DEBUG   ] Loading configuration from C:\Users\rbraun\Anaconda3\envs\eodownload\Lib\site-packages\eodag\resources\providers.yml
2026-04-30 14:57:56,918 eodag.plugins.manager            [DEBUG   ] EcmwfApi plugin skipped, eodag[ecmwf] or eodag[all] needed
2026-04-30 14:57:56,942 eodag.plugins.manager            [DEBUG   ] CSWSearch plugin skipped, eodag[csw] or eodag[all] needed
2026-04-30 14:57:56,957 eodag.provider                   [INFO    ] Loading user configuration from: C:\Users\rbraun\.config\eodag\eodag.yml
2026-04-30 14:57:56,973 eodag.provider                   [DEBUG   ] Loading configuration from environment variables
2026-04-30 14:57:56,985 eodag.core                       [INFO    ] Locations configuration loaded from C:\Users\rbraun\.config\eodag\locations.yml
2026-04-30 14:57:57,145 eodag.core                       [INFO    ] Searching on provider earth_search
2026-04-30 14:57:57,177 eodag.search.base                [DEBUG   ] Validate request
2026-04-30 14:57:57,244 eodag.search.base                [INFO    ] earth_search is configured with default sorting by 'start_datetime' in ascending order
2026-04-30 14:57:57,244 eodag.search.base                [DEBUG   ] Mapping eodag collection to provider collection
2026-04-30 14:57:57,245 eodag.search.qssearch            [DEBUG   ] Building the query string that will be used for search
2026-04-30 14:57:57,245 eodag.product.metadata_mapping   [DEBUG   ] Retrieving queryable metadata from metadata_mapping
2026-04-30 14:57:57,246 eodag.search.qssearch            [INFO    ] Sending search request: https://earth-search.aws.element84.com/v1/search
2026-04-30 14:57:57,246 eodag.search.qssearch            [DEBUG   ] Query parameters: {"query": {"mgrs:utm_zone": {"eq": "32"}, "mgrs:latitude_band": {"eq": "U"}, "mgrs:grid_square": {"eq": "LU"}}, "datetime": "2025-02-08T00:00:00.000Z/2025-02-11T00:00:00.000Z", "collections": ["sentinel-2-l2a"], "sortby": [{"field": "properties.datetime", "direction": "asc"}], "limit": 500, "page": "1"}
2026-04-30 14:57:57,246 eodag.search.qssearch            [DEBUG   ] Query kwargs: {}
2026-04-30 14:57:57,833 eodag.search.qssearch            [DEBUG   ] Next page merge collected and set for the next search
2026-04-30 14:57:57,851 eodag.search.qssearch            [DEBUG   ] Adapting 1 plugin results to eodag product representation
2026-04-30 14:57:57,873 eodag.core                       [ERROR   ] Error while searching on provider earth_search (ignored):
Traceback (most recent call last):
  File "C:\Users\rbraun\Anaconda3\envs\eodownload\Lib\site-packages\eodag\api\core.py", line 1869, in _do_search
    search_result = search_plugin.query(prep, **search_params)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\rbraun\Anaconda3\envs\eodownload\Lib\site-packages\eodag\plugins\search\qssearch.py", line 1848, in query
    eo_products_normalize = self.normalize_results(provider_results, **kwargs)
                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\rbraun\Anaconda3\envs\eodownload\Lib\site-packages\eodag\plugins\search\qssearch.py", line 1863, in normalize_results
    normalized = super().normalize_results(results, **kwargs)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\rbraun\Anaconda3\envs\eodownload\Lib\site-packages\eodag\plugins\search\qssearch.py", line 1339, in normalize_results
    product._normalize_bands()
  File "C:\Users\rbraun\Anaconda3\envs\eodownload\Lib\site-packages\eodag\api\product\_product.py", line 434, in _normalize_bands
    normalize_bands(self.assets[key])
  File "C:\Users\rbraun\Anaconda3\envs\eodownload\Lib\site-packages\eodag\api\product\metadata_mapping.py", line 1870, in normalize_bands
    band = processed_bands[index]
           ~~~~~~~~~~~~~~~^^^^^^^
IndexError: list index out of range

Environment:

• Python version: Python 3.11.14
• EODAG version: eodag (Earth Observation Data Access Gateway): version 4.2.0

[lhoupert]

I encountered similar issue . An other problem is that the searching for data on a specific provider will return 0 results but data are available.

I wrote some reproducible steps for my use case below.

Summary

Searching for COP_DEM_GLO30_DGED on the earth_search provider returns 0 results since v4.2.0.
The STAC API call succeeds and returns the expected tiles, but normalize_bands() (introduced in #2050) raises an IndexError when a product has raster:bands but no eo:bands. The exception is silently swallowed by _do_search, so the CLI reports "Returned 0 products" with no warning.

Environment

• eodag version: 4.2.0
• Python: 3.14
• OS: macOS 15 (arm64)
• Provider: earth_search
• Collection: COP_DEM_GLO30_DGED → STAC collection cop-dem-glo-30

Steps to reproduce

uvx eodag search -p earth_search -c COP_DEM_GLO30_DGED \
  --box -3 41 5 44 \
  --all \
  --storage dem_search

Output:

Returned 0 products
Results stored at 'dem_search.geojson'  # empty FeatureCollection

Minimal Python reproducer

Regression introduced in eodag 4.2.0 by PR #2050 which added normalize_bands().
Versions ≤ 4.1.x are not affected — normalize_bands did not exist and the search
returned results normally.

Both snippets below run with a transient env (no pip install; network required).

---

eodag 4.1.0 — working ✓

uvx --with 'eodag==4.1.0' python - <<'PY'
import json
import urllib.request
from importlib.metadata import version

print(f"eodag version: {version('eodag')}")

# --- 1. Query the STAC catalog -----------------------------------------

LIMIT = 5
BBOX  = [-3, 41, 5, 44]
URL   = "https://earth-search.aws.element84.com/v1/search"

payload = json.dumps({
    "collections": ["cop-dem-glo-30"],
    "bbox": BBOX,
    "limit": LIMIT,
}).encode()
req = urllib.request.Request(
    URL,
    data=payload,
    headers={"Content-Type": "application/json"},
)
with urllib.request.urlopen(req) as resp:
    data = json.loads(resp.read())

n       = len(data.get("features", []))
matched = data.get("context", {}).get("matched")

print(f"\n[1] STAC search returned {n} item(s) (page limit={LIMIT}).")
if matched is not None:
    print(f"    Total matches for this bbox: {matched}.")
else:
    print("    context.matched not present (some servers omit it).")

# --- 2. Inspect the first returned item --------------------------------

feature     = data["features"][0]
fid         = feature.get("id", "<missing>")
asset_names = sorted(feature["assets"].keys())

print(f"\n[2] First item id : {fid!r}")
print(f"    Asset names   : {asset_names}")

# --- 3. Check band metadata on the first asset -------------------------

asset_key, asset = next(iter(feature["assets"].items()))
has_eo     = "eo:bands"     in asset
has_raster = "raster:bands" in asset

print(f"\n[3] Asset {asset_key!r} — field keys: {sorted(asset.keys())}")
print(f"    eo:bands present     : {has_eo}")
print(f"    raster:bands present : {has_raster}")
if has_raster:
    rb = asset["raster:bands"]
    print(f"    raster:bands count   : {len(rb)}")
    if rb:
        print(f"    first band keys      : {list(rb[0].keys())}")

# --- 4. Confirm normalize_bands does not exist in 4.1.0 ----------------

try:
    from eodag.api.product.metadata_mapping import normalize_bands
    print("\n[4] normalize_bands exists — unexpected for 4.1.0")
except ImportError:
    print("\n[4] normalize_bands does not exist in this version.")
    print("    The search pipeline never calls it, so no crash occurs.")
PY

Output:

eodag version: 4.1.0

[1] STAC search returned 5 item(s) (page limit=5).
    Total matches for this bbox: 29.

[2] First item id : 'Copernicus_DSM_COG_10_N43_00_W003_00_DEM'
    Asset names   : ['data']

[3] Asset 'data' — field keys: ['href', 'raster:bands', 'roles', 'title', 'type']
    eo:bands present     : False
    raster:bands present : True
    raster:bands count   : 1
    first band keys      : ['sampling', 'data_type', 'spatial_resolution', 'unit']

[4] normalize_bands does not exist in this version.
    The search pipeline never calls it, so no crash occurs.

---

eodag 4.2.0 — broken ✗

uvx --with 'eodag==4.2.0' python - <<'PY'
import json
import urllib.request
from importlib.metadata import version
from eodag.api.product.metadata_mapping import normalize_bands

print(f"eodag version: {version('eodag')}")

# --- 1. Query the STAC catalog -----------------------------------------

LIMIT = 5
BBOX  = [-3, 41, 5, 44]
URL   = "https://earth-search.aws.element84.com/v1/search"

payload = json.dumps({
    "collections": ["cop-dem-glo-30"],
    "bbox": BBOX,
    "limit": LIMIT,
}).encode()
req = urllib.request.Request(
    URL,
    data=payload,
    headers={"Content-Type": "application/json"},
)
with urllib.request.urlopen(req) as resp:
    data = json.loads(resp.read())

n       = len(data.get("features", []))
matched = data.get("context", {}).get("matched")

print(f"\n[1] STAC search returned {n} item(s) (page limit={LIMIT}).")
if matched is not None:
    print(f"    Total matches for this bbox: {matched}.")
else:
    print("    context.matched not present (some servers omit it).")

# --- 2. Inspect the first returned item --------------------------------

feature     = data["features"][0]
fid         = feature.get("id", "<missing>")
asset_names = sorted(feature["assets"].keys())

print(f"\n[2] First item id : {fid!r}")
print(f"    Asset names   : {asset_names}")

# --- 3. Check band metadata on the first asset -------------------------

asset_key, asset = next(iter(feature["assets"].items()))
has_eo     = "eo:bands"     in asset
has_raster = "raster:bands" in asset

print(f"\n[3] Asset {asset_key!r} — field keys: {sorted(asset.keys())}")
print(f"    eo:bands present     : {has_eo}")
print(f"    raster:bands present : {has_raster}")
if has_raster:
    rb = asset["raster:bands"]
    print(f"    raster:bands count   : {len(rb)}")
    if rb:
        print(f"    first band keys      : {list(rb[0].keys())}")

# --- 4. Trigger the bug ------------------------------------------------
# normalize_bands() should merge raster:bands into the STAC 1.1 bands list.
# When eo:bands is absent, processed_bands starts empty and the index read
# on the first iteration raises IndexError.

print("\n[4] Calling normalize_bands() ...")
normalize_bands(asset)   # IndexError: list index out of range
PY

Output:

eodag version: 4.2.0

[1] STAC search returned 5 item(s) (page limit=5).
    Total matches for this bbox: 29.

[2] First item id : 'Copernicus_DSM_COG_10_N43_00_W003_00_DEM'
    Asset names   : ['data']

[3] Asset 'data' — field keys: ['href', 'raster:bands', 'roles', 'title', 'type']
    eo:bands present     : False
    raster:bands present : True
    raster:bands count   : 1
    first band keys      : ['sampling', 'data_type', 'spatial_resolution', 'unit']

[4] Calling normalize_bands() ...
Traceback (most recent call last):
  File "<stdin>", in normalize_bands
    band = processed_bands[index]
IndexError: list index out of range

Root cause

normalize_bands (added in #2050, metadata_mapping.py ~line 1867) merges raster:bands
into processed_bands, which was expected to have been pre-populated by the eo:bands loop
above it. When a product has only raster:bands (no eo:bands), processed_bands is
empty and processed_bands[index] at the start of the loop crashes immediately.

The else: processed_bands.append(band) branch a few lines below (line 1879) was clearly
intended to handle this case, but is unreachable because the crash happens first.

# migrate raster:bands -> bands
if len(bands["raster:bands"]) > 0:
    index = 0
    for item in bands["raster:bands"]:
        band = processed_bands[index]   # ← IndexError when processed_bands is empty
        ...
        if index < len(processed_bands):
            processed_bands[index] = band
        else:
            processed_bands.append(band)  # ← intended fallback, never reached
        index += 1

Suggested fix

Guard the read with the same condition used for the write:

if len(bands["raster:bands"]) > 0:
    index = 0
    for item in bands["raster:bands"]:
        band = processed_bands[index] if index < len(processed_bands) else {}
        for key in item:
            if key in UNPREFIX_BAND_FIELDNAME:
                band[key] = item[key]
            else:
                band["raster:{}".format(key)] = item[key]
        if index < len(processed_bands):
            processed_bands[index] = band
        else:
            processed_bands.append(band)
        index += 1

Additional notes

• The COP_DEM_GLO30_DGED → cop-dem-glo-30 mapping is explicitly present in
  providers.yml under earth_search, so this collection is an intended target.
• The STAC API itself works correctly (verified: 29 tiles returned for the bbox above).
• The error is caught and ignored in core.py:_do_search, which is why the CLI gives
  no warning — making this particularly hard to diagnose without -vvv.