I hesitate to even bring it up, thinking I should be able to figure it out, but I've never seen behavior like this. Keep reading if you want to twist your brain in knots like mine.
TL/DR - stop now if you'd rather be productive for your day job. Somehow, I have a consistent state where a file originating in Excel, modified through POI, then saved and opened in Excel causes not just an Excel repair, but incorrect/old data as well. So far, just a coding error, I figure. But the weirdness starts when I use the debugger to check on things. A specific row that I know comes out wrong has the right cell values when I use wb.getSheet(#).getRow(#).getCell(#).toString(), but when I check the CTRow and CTCell objects, then I find the old values still. So somehow, I've got the XSSFRow._cells collection out of sync with the CTRow _row contents. Anyone know how I could have done that? I may be calling workbook.write() in multiple places for different purposes/copies. It looks to me like XSSFRow.onDocumentWrite() explicitly replaces the CTRow array of CTCell beans with new ones due to bug, #56170, but it is replacing the CTCell references in the XSSFCell objects with the new copies, . I haven't found a smoking gun yet, but any thoughts are welcome. Perhaps the issue may lie with the formula evaluator created prior to the save event? It may be related to that bug, or even directly tied to it, as that was about updating tables, and that's part of the updates my process is doing as well. Sorry for the long document, normally I just figure these out on my own, but I've been at this a week so far, which has never happened before. If it comes down to black box behavior by XMLBeans, that would help explain it, as I know next to nothing about that library's details. Greg