In the migration to double precision (seen in aw080 but also in master) values we have places in vcl where based on double translation and scale the integer Point and Size need to be created. Up to now a direct basegfx::fround was used for this, but this can be wrong for the Size. It is more precise (and may give a different integer result) when using basegfx::fround(translation + scale) and substractiong the already rounded Point. This is because the target position where the double addition lands is the correct position to round where the pure scale values may round based on different (and in this case wrong) unit buckets (always from ]v-0.5 .. v+0.5] when using basegfx::fround).
This visualizes in sometimes appearing single-pixel sized 'gaps' in the page shadow (just open new draw/impress and play with zoom, watch page shadow, esp. in the edges since there different bitmaps get glued).