meson: decouple session tracking mechanism from unit and sysusers #417
Conversation
bluca
changed the title
meson.build
Outdated
| systemd_systemdsystemunitdir = systemd_dep.get_pkgconfig_variable('systemdsystemunitdir') | ||
| endif | ||
|
|
||
| systemd_sysusers_dir = systemd_dep.get_pkgconfig_variable('sysusers_dir', default: '/usr/lib/sysusers.d') |
There was a problem hiding this comment.
This should at least provide an option to manually specify the directory like for system units, in case there is no libsystemd.
There was a problem hiding this comment.
This is unrelated to libsystemd, it's a separate and independent pkg-config file
There was a problem hiding this comment.
Ah, then in case there's no systemd.pc.
There was a problem hiding this comment.
Then it's a bug in the package/build script/etc and it needs to be provided
There was a problem hiding this comment.
It seems unnecessarily unkind to packagers to add a new required dependency just to get a single path.
There was a problem hiding this comment.
But why is it acceptable to fall back to the default value if the .pc file is present but doesn't contain the required option, but it's not acceptable to fall back to the default value if the .pc file is not present?
There was a problem hiding this comment.
Good point, that can be dropped too
There was a problem hiding this comment.
I don't mean to ad fuel to the fire but, TBH, configuring 'dependency' if 'dependency' is not (and probably never will be) present on the system (be it systemd or whatever) seems like abomination to me (/me just shot himself in foot, probably), even though we use prepared and good-looking system like pkgconf. @bluca, this is nothing personal, I just believe there are cleaner ways than dropping unused config files in possibly-yet-non-existent directories on systems that have right to go different ways. But maybe I'm still a purist in this topic :D
There was a problem hiding this comment.
Please don't force a stub .pc file. The existence of systemd.pc (even a stub) can imply that systemd exists on the system, and there will be programs that will make that assumption and configure themselves incorrectly.
(speaking as a maintainer of a downstream distro that happens to not use most systemd components)
There was a problem hiding this comment.
I don't mean to ad fuel to the fire but, TBH, configuring 'dependency' if 'dependency' is not (and probably never will be) present on the system (be it systemd or whatever) seems like abomination to me (/me just shot himself in foot, probably), even though we use prepared and good-looking system like pkgconf. @bluca, this is nothing personal, I just believe there are cleaner ways than dropping unused config files in possibly-yet-non-existent directories on systems that have right to go different ways. But maybe I'm still a purist in this topic :D
I disagree, but it's your project, so as you wish - the pkgconf is now optional, and the well-known default installation directories are used as a fallback if missing.
|
Please be aware that in the current state this commit is in you would be breaking 042897e again. |
bluca
force-pushed
the
meson_fix
branch
2 times, most recently
from
c00bf1f
to
9e1a903
Compare
…pfiles Session tracking build options are about whether logind or other APIs are used in the polkit code, at compilation and linking time. units, sysusers and tmpfiles files have nothing to do with code changes or APIs, they simply install config files, that can just be ignored if they are not needed. Use pkg-config if available, but otherwise have a hard-coded fallback with the well-known defaults. This also fixes another bug, 'systemdsystemunitdir' is specified as an option the systemd_dep variable is not defined, but the sysusers.d directory lookup uses it, causing a build failure: dh_auto_configure -- \ -Dexamples=false \ -Dintrospection=true \ -Dman=true \ -Dsystemdsystemunitdir=/usr/lib/systemd/system \ -Dtests=true \ -Dgtk_doc=true -Dsession_tracking=libsystemd-login cd obj-x86_64-linux-gnu && DEB_PYTHON_INSTALL_LAYOUT=deb LC_ALL=C.UTF-8 meson setup .. --wrap-mode=nodownload --buildtype=plain --prefix=/usr --sysconfdir=/etc --localstatedir=/var --libdir=lib/x86_64-linux-gnu -Dpython.bytecompile=-1 -Dexamples=false -Dintrospection=true -Dman=true -Dsystemdsystemunitdir=/usr/lib/systemd/system -Dtests=true -Dgtk_doc=true -Dsession_tracking=libsystemd-login The Meson build system Version: 1.3.1 Source dir: /builds/bluca/polkit/debian/output/source_dir Build dir: /builds/bluca/polkit/debian/output/source_dir/obj-x86_64-linux-gnu Build type: native build Project name: polkit Project version: 124 <...> Run-time dependency libsystemd found: YES 255 Checking for function "sd_uid_get_display" with dependency libsystemd: YES Checking for function "sd_pidfd_get_session" with dependency libsystemd: YES ../meson.build:222:37: ERROR: Unknown variable "systemd_dep". Follow-up for 24f1e0a
|
I got a message that tools like tmpfiles.d and systemd-sysusers can also exist separately from systemd, hence it make sense (since we use those tools) to provide configuration files for them. IF they are not desired on a specific distro, a package maintainer for that distro can always exclude those files. Agreed? |
Yes there are separate implementations in bash for both. Thanks. |
Session tracking build options are about whether logind or other APIs
are used in the polkit code, at compilation and linking time.
units, sysusers and tmpfiles files have nothing to do with code changes
or APIs, they simply install config files, that can just be ignored if
they are not needed. Use pkg-config if available, but otherwise have
a hard-coded fallback with the well-known defaults.
This also fixes another bug, 'systemdsystemunitdir' is specified as
an option the systemd_dep variable is not defined, but the sysusers.d
directory lookup uses it, causing a build failure:
Follow-up for 24f1e0a