Categories: Development

TODO list for CRUX 3.8

Ports

toolchain updates: gcc 14, binutils, glibc, python3 (done)
decide whether to move libbsd and libmd to core (shadow enables it by default)
- resolution: moved both to core (commit, commit)
add a new group 'plugdev' and user 'sgx' to satisfy udev (done)
reconsider the default Mail Transport Agent: something more lightweight than exim will satisfy the majority of use cases.
- What can we use instead? Proposals?
- Or don't include any MTA in core, since our default cron daemon has the -M switch to obviate the need for /usr/sbin/sendmail (but this feature needs to be documented if we ship crontabs with our core ports).
- If I had to say something, I'd go with none as well probably. (tb)
rewrite rc scripts to target dash instead of bash? (Side effect: users with UPGRADE ^etc/rc.conf$ NO in their pkgadd.conf will have to be notified of breaking changes, as dash lacks support for bash-style arrays)
in '/etc/rc' use a better algorithm to seed /dev/urandom (FS#1921)
in '/etc/rc' consider protecting all mount commands behind a /bin/mountpoint test, for users who configure a multi-stage boot process (done)
in '/etc/rc.conf' consider introducing a variable for the screen blanking time (FS#1922) (done)
glibc now keeps libcrypt disabled by default (see 2.38 release notes)
- libxcrypt seems to be the way forward
- resolution: imported libxcrypt to core (commit)

drop python2 (done)
replace cdrkit with cdrtools? Every toolchain update seems to introduce new patches for cdrkit, and there's no upstream maintainer taking responsibility for merging distro patches. (Side effect: system/iso/Makefile will need to use mkisofs instead of genisoimage)
decide whether to close FS#1313 as a "wontfix", or to update the git driver to act more like rsync and httpup
formulate a better definition of what belongs in opt, and what should be moved to contrib (mostly done)
glib and gobject-introspection circular dependency
- resolution: imported glib-introspection and introduced as a dependency (commit history)

move xorg-libpixman to the opt collection, for easier setup of a headless machine with the xorg collection deactivated (FS#1924)
build mesa with dynamically-configured value for vulkan-drivers (currently hard-coded as amd,intel), similar to how PKGMK_MESA_GALLIUM and PKGMK_MESA_PLATFORMS are defined
make x11 optional, maybe opt for a wayland default if hardware support is available?

consider adhering more closely to the guidelines for placing config files into '/etc' versus '/usr/etc'. New repo maintainers are confused by the apparent contradiction between the Handbook and actual practice (system daemons sometimes storing their configs under '/usr/etc', userland apps sometimes storing their configs under '/etc'). We should either give them a better example to follow, or rewrite the guidelines to reflect a new consensus.
- How do you all feel about reconsidering our guideline? Having just one etc folder might simplify, especially with a system wide XDG_CONFIG_DIRS setting

recognize zstd as a valid compression mode for built packages
verify the signature of Pkgfile before sourcing it (FS#1851)
disallow Pkgfile or pkgmk.conf overriding command-line options like --ignore-signature (current counterintuitive behaviour rediscovered by r0ni)
consider supporting simple overlays
consider the possibility to handle git sources natively
fix the symlink handling in pkgadd (FS#15)
update man-pages to encourage future-proof Pkgfiles and to warn against perverse configurations
respect UPGRADE NO rules even if the footprint of the new package no longer has the matching file (FS#1074)

allow transactions to use the Optional line in the Pkgfile header (current behaviour is to consider only hard dependencies)
improve the argument parsing when a mix of installed and not-installed ports is given (FS#1910)
allow users to configure auto-removal of build logs when uninstalling a package (patch by farkuhar)
consider enabling dependency resolution by default, and letting the user disable it with the --nodeps switch (in June 2008 jw provided some historical notes about managing dependencies to justify such a change)
update the man-page: show practical uses of pipelines (which would justify closing FS#849 and FS#1605 as "wontfix"), and delete from the examples any references to dropped ports
ensure that whitespace separates the port name from the version, when 'diff' detects version differences in ports of more than 20 characters (FS#1930)
respect --install-root when pre- or post-install scripts are enabled (bug report by emmett1)

update the language in the dialog presented by prtrej to be less misleading. Right now, "older" and "newer" are applied without bothering to check timestamps. (patch by ukky)
allow prtrej to recognize SYSROOT, for users who cross-compile (patch by ukky)
switch to Perl implementations of prtwash and prtsweep (rewrites by farkuhar)(Done)

bump Linux to 6.6 and ship a corresponding modular kernel config (done, SiFuh)
edit the init script so that crux media is detected when booting from ventoy (patch by ukky) - Seems like this is no longer needed, tested ventoy 1.0.99 with both 3.7 updated and 3.8 rc ISOs (No longer needed)
include zfs?
use setup-helper to rename ports that refer to an outdated soversion (done)

move the discussion of rejmerge to a more prominent place in the handbook
mention in the handbook the cron jobs that are enabled by default
warn users about the behaviour of the git driver if FS#1313 is closed as a "wontfix"
decide what to do about upstream ports that do not bundle pre-generated man-pages