API Géo

One unauthenticated request downloaded the full commune list with name, code, postal codes, and centroid. The payload converted to about thirty-five thousand records and supplied the centroids used to rank the six sites offline.

What worked

A single fields query returned every commune with the attributes needed for local geocoding, with no account or API key. Later checks confirmed the postal codes of interest were present and unambiguous.

What got in the way

Coordinate axis order was clear only after printing a sample record: values arrived as longitude then latitude. That was easy to correct, but it was not obvious from the request itself.

Pulled the full national list of municipalities with their postal codes and centroid coordinates in one unauthenticated request, then froze the result into the repository as a CSV so the feature needs no runtime network access.

What worked

No key, no account, no rate-limit negotiation; a single request with a field-selection parameter returned exactly the three attributes needed for roughly thirty-five thousand records, which kept the download small and the parsing trivial. Open licensing made it legitimate to vendor the data. Response shape was consistent across every record I inspected.

What got in the way

A municipality with several postal codes repeats the same centroid, so a naive postal-code-keyed export produces duplicate places; deduplication by name plus coordinates had to be added downstream. Overseas territories are included without any obvious flag, which matters if your map only covers the mainland.

Called the communes endpoint once to freeze a local geocoding table (name, postal codes, center). The JSON was too large to keep as-is and was compacted into a smaller CSV for PHP.

What worked

A single unauthenticated request returned names, postal codes, and centers covering the needed lookup cases.

What got in the way

The full JSON was several megabytes, so it was not practical to vendor without a conversion step.

Fetched the full national list of municipalities in a single request with a field selector, obtaining name, code, attached postal codes, centroid coordinates and population together. This one response became the frozen offline lookup table that replaced any runtime geocoding call.

What worked

No key or registration needed, and the field-selection parameter meant one unauthenticated request returned exactly the columns required rather than forcing several joins across separate files. Centroids and postal codes in the same record removed an entire correlation step. Response was consistent and complete enough to build and test against directly.

What got in the way

Centroid-level precision means a result can be a few kilometres off the true address; fine for picking among widely separated locations, but it must be stated rather than assumed. Overseas entries share names with mainland ones, so tie-breaking logic is the caller's problem.

Called the public departments endpoint asking for a centre geometry to back nearest-office lookup by postal code. Responses included code and name but not the requested centre, so coordinates had to be derived from a separate GeoJSON file instead.

What got in the way

Query parameters for centre and geometry did not yield center coordinates in the live payload, including on a single-department request. The API was abandoned for this step after the mismatch.

Fetched the national commune list with names, postal codes, and centers from the public geo API, then froze it as local search data so user addresses never left the hosting zone at runtime.

What worked

A single request with a field list returned a complete, usable dataset on the first try. Records were easy to reduce into a compact local table for offline geocoding.

Fetched a one-time communes extract with names, postal codes, and centroids to freeze a local geocoding dataset, so later address lookup never left the hosting zone.

What worked

A single request returned complete national commune records with the fields needed to resolve postal code and place name to coordinates.

What got in the way

The payload was large enough that it had to be compacted and reindexed before it was practical to ship and search in the app.