From: Pablo Date: Mon, 23 Mar 2026 02:33:25 GMT Subject: Re: [GSoC PATCH WIP RFC v3 1/3] graph: add --graph-lane-limit option Message-ID: In-Reply-To: Junio C Hamano () writes: > If you have active branches, whether they are merged to some very > small number of integration branches or they are left updating > without getting merged for a long time, you'd end up getting very > wide output from "log --all --graph"? Or is this a problem only > when these active branches are merged and having to show merge > commits create the need for wider output? I cannot quite see which > from the above description. The graph gets wide based on the branches active at the same time, each one occupies two columns, merges don't create new lanes. I'll improve the problem description. > The implementation detail that you happened to choose a C > proprocessor macro instead of hardcoded constatnt to write the lower > limit is not something readers of the log message needs to know. > They can see that in the "log -p" output easily. > > What is more helpful for readers to know is what you mean by > "graph-lane", what you are counting, and why you want its lower > limit to 1 (instead of 0 or 2). These reasoning behind the design > is much more important to record to help future developers who want > to fix bugs in this code or who want to extend the feature this code > adds, without violating the underlying assumption and design goals > of the original author (i.e., you). Ok, I'll focus more on why, rather than how. the minimum is set to 1 to have at least 1 visible lane, even though it could accept 0 it's the same as not using this option, 0 it's treated as no limit and I found it better to not give the users the option to place it because no other option seems to behave like this, so I thought the best would be to force the input to be >= 1 to be valid. in v4 I'll make sure that this is clear for others. > This is "unsigned int"; don't we want the other places (like the > on-stack local variable handle_revision_opt() uses to parse the > value from the command line) and the parameter used in > graph_needs_truncation() helper function all consistently use the > same type? I saw that other options like max_count, min/max_parents use int instead of unsigned int, so yes the most consistent would be to have it as int also, but it would make no sense to have -1 visible lanes. I thought it would be a good idea to keep it explicit that it can't be negative. this examples, max_count, etc were the closest examples I saw, but parse_count does return an int so i can't cast it to unsigned without checking if its neg that's why this > int max_lanes = parse_count(optarg); > if (max_lanes < MINIMUM_GRAPH_COLUMNS) > die(_("minimum lanes is %d, cannot be set to %d"), > MINIMUM_GRAPH_COLUMNS, max_lanes); > revs->graph_max_lanes = max_lanes; where it checks if it's < 1, now max_lanes has to be > 1 and it fits in an unsigned int. But i do understand to keep the consistency and the coding guidelines, i'll make it an int. Thanks for the feedback!