Am 02.08.19 um 21:14 schrieb Junio C Hamano: > I can see how this would help, but it somehow feels a bit brittle > to rely on where the decorations get loaded. Right. > I wonder if it would help to move the ability to handle decoration > filter down from the log layer to revisions.c API layer. > > It looks to me that this caller of setup_revisions() can prepare > decoration_filter before it calls setup_revisions(); we can let the > revisions.c layer call load_ref_decorations() in setup_revisions() > if that is the case, no? Having cmd_log_init_finish() call load_ref_decorations() before setup_revisions() would indeed solve the issue as well. But we need to call the latter to check if --pretty=raw was given and avoid loading decorations in that case, don't we? > Other two callers of load_ref_decorations() are deep inside pretty.c > but I wonder in the longer term if we would want to turn them into > an "a lot higher level should have already loaded decorations" > assert. This would require that higher level to parse the user format to check if %d or %D is present before formatting the first item. Hmm. René
