Hi Junio, On Mon, 28 Jan 2019, Junio C Hamano wrote: > Johannes Schindelin <Johannes.Schindelin@xxxxxx> writes: > > > The big difference is that gettext is needed to build Git and run its test > > suite. While gnu-time is only needed if you want to run the perf tests, > > which is not a part of the CI configuration we have, neither Travis nor > > Azure Pipelines. > > > > So as long as we do not run the perf tests as part of the CI runs, that > > optional dependency should *not* be included in *CI_TYPE* specific > > sections of the code. > > Ah, in that case, I do not think it makes much sense to even keep > that comment. As you said, ci/ is about running tests under CI > platform, and the scripts are not designed to be run manually with > tweaks, and none of our CI integration runs the perf thing, there is > no point to even mention it. Well, yes and no. Yes, it does not make much sense to cater to interactive usage when we target a CI system. But yes, it *does* make sense to let users run those ci/ scripts in case they want to investigate test failures. Take, for example, a Windows user contributing patches that end up failing on Ubuntu with clang. Then they can install Ubuntu from the Windows Store, run those scripts in ci/ to painlessly get the necessary packages (without spending extra time to research which ones, and how to install them on Ubuntu in the first place), and they have a pretty good head start at debugging this interactively. Likewise, I take the comment about perf testing as a seedling for testing performance in a controlled, repeatable and automated manner. As such, it would probably make less sense to step on that seedling and crush it; Instead, we will want to keep the comment there because it does no evil and at the same time is kind of a placeholder for something beautiful to come. Ciao, Dscho
