Upstream contribution · HKUDS/nanobot
nanobot · Preserving spreadsheet cell provenance and inline media in Nanobot
Problem
Nanobot reconstructed spreadsheet coordinates by counting tab characters in text. A tab inside a cell was indistinguishable from a delimiter, so search results could cite the wrong Excel column. Its media reader also rejected a valid data URL when the base64 marker used uppercase letters.
Engineering response
The workbook extractor records actual cell starts and uses those offsets to report search matches. The media parser accepts the marker case-insensitively while retaining strict payload checks. The media PR records a maintainer follow-up that uses delimiter parsing to avoid pathological header matching.
Evidence and scope
Workbook regressions exercise actual XLSX files through the grep tool, with coverage for tabs within a cell and empty leading cells. The media PR records a two-case before/after test and 21 passing decoder tests at the contributor's tested revision. The cases document local input handling.