On Mon, Mar 28, 2022 at 03:59:07PM +0800, Teng Long wrote: > Although all MIDX bitmap are scanned, only one of them will eventually work > , and there seems to be no guarantee that the last MIDX that works is the > most appropriate one? (with the same comfusion applies to non-MIDX > bitmap's behavour...) I think your "which is the most appropriate one?" exactly highlights the problem. Since Git can only handle one .bitmap file at a time, we should tell the user when we found more than one candidate. In this case, they may be expecting or depending on a particular choice for performance reasons, and so may be rather surprised or upset if Git chose a different available .bitmap. The one exception to this is when we have both a single- and multi-pack bitmap available, in which case it almost always makes more sense to pick the multi-pack bitmap. So we don't complain in this case (i.e., if open_midx_bitmap() succeeds, we immediately return success from open_bitmap() and proceed to have prepare_bitmap_git() call load_bitmap() on the previous result without even bothering to check for pack bitmaps). Thanks, Taylor
