You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
csv_output silently overwrites distinct fields when flattened keys collide #3865
When csv_output is enabled, list-tool results are converted to CSV by flattening nested objects with .-joined column names. A literal field name containing . can produce the same column name as a nested field. One value is silently overwritten; Go map iteration makes the surviving value unpredictable.
Reproduction (from the current implementation)
Convert a list response containing this JSON row through jsonTextToCSV:
appendFlattenedCSVFields maps both the top-level user.login and the nested user → login to the same row["user.login"]. The resulting CSV has only one user.login column, containing either literal or nested, depending on map iteration order. The other value is lost.
The conversion is used by withCSVOutput for default list_* tools whenever the csv_output feature flag is enabled (including insiders mode). This is a code-path demonstration, not a claim of a live GitHub API response with these exact fields.
Expected behavior
Distinct JSON fields remain distinguishable in CSV, or the conversion explicitly reports that it cannot represent this row without loss. The chosen output should be deterministic.
Suggested fix
Escape/encode path components when building flattened column names, or detect collisions and use unambiguous names/fail clearly. Add a regression test with both a literal dotted key and an equivalent nested path (and verify neither value is silently dropped).
Source
pkg/github/csv_output.go: appendFlattenedCSVFields, csvColumnName, csvRowsFromArray, and withCSVOutput.
pkg/github/csv_output_test.go: existing flattening tests cover nested objects, but not colliding flattened names.
Activity