fix(gdal): st_read_meta returns empty results for /vsi* paths

[jatorre]

Summary

st_read_meta() silently returns 0 rows when called with GDAL virtual filesystem paths (/vsizip/, /vsicurl/, /vsigs/, etc.). This is because the Bind function uses MultiFileReader::CreateFileList() which doesn't understand /vsi* prefixes and returns an empty list (masked by ALLOW_EMPTY).

This is inconsistent with st_read(), which correctly handles /vsi* paths by passing them directly to GDAL via GetPrefix().

The Fix

When the input path starts with /vsi, bypass MultiFileReader and pass the path directly to GDAL — matching the existing behavior of st_read(). Non-/vsi paths continue to use MultiFileReader for glob expansion.

Before:

-- Returns 0 rows silently
SELECT * FROM st_read_meta('/vsizip//vsicurl/https://example.com/data.zip');

After:

-- Returns full metadata (layers, geometry types, CRS, fields)
SELECT * FROM st_read_meta('/vsizip//vsicurl/https://example.com/data.zip');

Root Cause

In gdal_module.cpp, the ST_Read_Meta::Bind function:

const auto mfreader = MultiFileReader::Create(input.table_function);
const auto mflist = mfreader->CreateFileList(context, input.inputs[0], FileGlobOptions::ALLOW_EMPTY);
return make_uniq_base<FunctionData, BindData>(mflist->GetAllFiles());

MultiFileReader::CreateFileList tries to resolve /vsizip/... through DuckDB's VFS layer, fails, and returns an empty list. The ALLOW_EMPTY flag suppresses the error.

Meanwhile, st_read avoids this by using GDALClientContextState::GetPrefix() which already has explicit /vsi handling (line 446):

if (StringUtil::StartsWith(value, "/vsi")) {
    return value;  // Pass through to GDAL unchanged
}

Use Case

This enables listing layers in remote archives without downloading them:

-- Discover layers in a remote shapefile zip
SELECT unnest(layers) FROM st_read_meta('/vsizip//vsicurl/https://storage.googleapis.com/bucket/data.zip');

-- Works with signed URLs using the {braces} syntax
SELECT * FROM st_read_meta('/vsizip/{/vsicurl/https://...?X-Goog-Signature=...}/layer.shp');

Previously the only workaround was calling ogrinfo as an external process, or downloading the file locally first.

Related Issues

• How to query a layer of OpenFileGDB dataset? #45 — Original request for metadata reading (led to st_read_meta)
• Please standardize file access and basic capabilities #729 — Reports st_read_meta crashes/returns 0 rows for some formats
• calling st_read_meta() crash duckdb #544 — st_read_meta() crash reports