Le 2019-12-16 12:52, Fabio Valentini a écrit :
On Mon, Dec 16, 2019 at 12:46 PM Petr Viktorin
Yeah, I think using file lists works OK-ish for some things
(translations with %find_lang, Java packaging, etc.), but it doesn't
really map to the way python packages work in fedora.
Practically, automating the packaging of any upstream distribution
format, that does not resumes itself to a single kind of file in a
single filesystem directory, will end up as file lists.
File lists are the only mechanism provided by rpm to communicate between
%prep, %build, %install and %deploy (aka %files) sections. So as soon as
you automate anything in any previous section, that has consequences on
the deploy phase, you have to resort to file lists (shell variables do
not pass the section barrier, rpm variables are evaluated before the
spec is executed)
And even within an rpm section, the counter-productive decision to use
bash in sh mode, means you do not have access to all the array
manipulating facilities built-in bash (mapfile…). Since manipulating
lists in legacy shell mode is horrible, you may as well resort to file
lists even within a section.