On Sun, May 12, 2019 at 09:28:40PM -0000, Christian Weisgerber wrote: > On 2019-05-11, Christian Weisgerber <[email protected]> wrote: > > > Splitting gettext into separate ports imposes a small maintenance > > burden: The gettext-tools build needs to be tweaked to pick up > > pre-installed libintl and libtextstyle. > > It also reaches into libtextstyle and wants to copy some header > files from there, some of which are only generated by a configure > run in libtextstyle. And since we tell gettext-tools to use the > pre-installed libintl, for its regression tests it now wants to run > a battery of tests against this OS-provided libintl, but those tests > expect the gettext functions to be in libc and not in a separate > libintl. > > All of this can be fixed, but I decided to go with multi-packaging. > > A notable side-effect of multi-packaging is that you need to build > the whole thing, even though many basic ports just require the small > runtime component. Then again, most interesting ports I want to > build nowadays depend on llvm+rust, so I'm not too worried about > gettext bloat.
The priority should still be bulk builds. gettext is very high in dpb's ranking. It always gets built as soon as its dependencies are done, e.g., as soon as dpb recomputes its queues when libiconv is finished.
