Found a surplus "pop state" which somehow causes a regression (fill color doesn't get properly reset) between 0.94 and 0.95. Shows in Barcode4J's fop-extension-demo.fo. The line has already been there in 0.94 but only shows in 0.95 probably due to an additional save/restore state pair added in PDFRenderer for reference areas after the block-container area refactoring.