On 12/02/26 2:10 pm, Helmut Grohne wrote: >> Do you mind rewording this and sending across a MR? I'll be happy to quickly >> merge that. > > There are several cases here and we may want to consider the most common > patterns. I think by far the most common pattern is not having installed > tests. At that point, on ignores noinsttest and just goes with nocheck. > > The second most common pattern probably is not having any tests. > > Neither of these are relevant to the tag as it only comes up once you > skip installing anything based on nocheck. So when the developer reads > the tag they are quite definitely dealing with installed tests. Most > commonly, installed tests are also run. So my expectation is that the > common case here is needing both. That's why I specifically added it.
I do understand your reasoning here. The problem occurs in cases where there are B-D that need to be marked with just !nocheck and B-D that need to be marked with !nocheck and !noinsttest both in the same source package; dbus for example. I find it problematic if the maintainer sees the explanation and ends up marking more than required B-D with both the tags. It'd be another problem if maintainer carries down this (wrong) understanding to other packages. > Would the following rephrasing improve the situation? > > | If a build dependency is required both for running build-time tests > | and for building installed tests, ... There are packages that install tests also in the same binary package along with the resulting program. It is not ideal, and should be split (which is another issue). However, I also think that !noinsttest would not apply in such a situation? What do you think about this? | If a build dependency is needed both to run build-time tests and to build | installed tests that reside in a separate binary package with !noinsttest profile, ... Best, Nilesh
