Office Document Extraction Fixed: ODT Text Loss, Empty ODS and ODP Documents, and Missing PPT and CSV Support

· Data Ingestion · Bug Fix · All updates

Office documents were being indexed incompletely, or not at all, and in most cases nothing said so. Three separate faults in the extraction layer have been fixed: truncated OpenDocument text, empty spreadsheets and presentations, and file types that were accepted for indexing and then quietly rejected.

01 ODT paragraphs stopped at the first styled word

A paragraph in an .odt file was read only as far as its first styled run. A line reading "Safety instructions for high voltage" was indexed as just "Safety". Because bold, italics and links appear in almost every real document, entire paragraphs were disappearing from indexes across whole document sets.

Before

Reading stopped at the first bold, italic or linked run. "Safety instructions for high voltage" was indexed as "Safety".

After

The full text of the paragraph is extracted, styled runs included. The same fix applies to spreadsheet cells and presentation slides picked up by the Web Crawler.

02 Spreadsheets and presentations indexed with no text

An .ods or .odp file passed every check, was accepted, and produced a document with an empty body. No error was raised, so there was nothing to notice: the file appeared in the index and simply never matched anything. Document extraction now reads both formats.

A file type with no extractor behind it now fails loudly instead of quietly producing an empty document. Silent success was the worst part of this bug.

03 Legacy PowerPoint and CSV are now indexed

Legacy PowerPoint (.ppt) and CSV files were being selected for indexing, sent for extraction, and then rejected as an unsupported file type, so they never reached the index at all. Both formats are now read, through the Data Ingestion API and the crawler alike.

Before .odt .ods .odp .ppt .csv first word only empty document, no error rejected as unsupported After all five formats extracted in full; unsupported types now fail loudly
Each format failed in a different way, and only the last one told you about it.
View the full changelog