The dependency tree created by parsing the output of "apt-cache depends" always
generates an array of array. The old representation of a simple dependency as
an hash is gone in favor of an OR with a single entry (so the hash is the only
element within the array).
So all the code in add_missing() that was dealing with the case of the hash
was never triggered.
I tweaked the OR case so that the generated log messages actually
mention "OR" only if it's a true OR dependency (i.e. with multiple
entries in the array).
Everywhere where we call add_missing() on Depends, we must do it also
on Recommends and Suggests based on $add_rec and $add_sug.
Tweak the code that is supposed to stop the recursion to use
the list that we have built but don't be strict about version
constraints on Recommends/Suggests since they can't always
be respected.
Closes: #601203
There's no point in passing them as parameters everywhere.
This changes the behaviour in some places:
1/ When we process the non-explicitly included packages (for COMPLETE=1),
we used to never take Recommends/Suggests into account, now we do. But
the only thing that will be changed is the ordering with some
Recommended packages being included before their depending package.
2/ When we process the EXTRANONFREE image, this one was always including
Recommends and Suggests, and it's now respecting the setting too.
It's required by debian-installer and cryptsetup >= 2.4.0 no longer
recommends it, so it ends up missing in the netinst image.
base-installer then fails to honor the apt-setup request and the
resulting system is unbootable because the initramfs doesn't contain
anything to unlock the encrypted partitions. :-(
Using the new generate_firmware_patterns script, generate a patterns
file in /firmware/dep11/<package>.patterns as we add a firmware
package onto media.
Deal with size changes as these are added/removed, and add them to the
md5sum.txt file on media too.
The source code doesn't use the MODULE_FIRMWARE() macro so the machinery
implemented in src:firmware-nonfree wouldn't help. Extract alias info
manually once, and make note to keep an eye on further updates (e.g. via
a crontab on d-i.debian.org).
The related code seems to have been useful a long time ago (the need to use
an RLE file and to reduce the number of colours to 16 has been dropped as of
syslinux version 3.62). So replace obsolete code with a simple cp command.
Signed-off-by: Wolfgang Schweer <wschweer@arcor.de>
tools/make_disc_trees.pl: Apply patch by Daniel Dickinson to create Packages
files for local/debian-installer. Closes: #523017
Signed-off-by: Holger Levsen <holger@layer-acht.org>
tools/start_new_disc: Apply patch by JH Chatenet to include local Release file
for debian-installer. Closes: #968671
Signed-off-by: Holger Levsen <holger@layer-acht.org>
Shift the extra_image helper function from boot-x86 to common.sh to
help.
Call extra_image with || true for now on boot-arm64 so we don't fail
builds straight away. Remove that || true later once we're happy
things work.
I was just bitten by this problem in Kali where the ISO was not
containing the expected packages because tools/sort_deps was accepting
"xserver-xorg-input-libinput-udeb" as a solution to
"xserver-xorg-input-all | xserver-xorg-input-driver" and the
udeb was among the package already selected for inclusion on the
ISO.
With this change, tools/sort_deps will look for udeb to solve
udeb dependencies and it will look for non-udeb to solve non-udeb
dependencies.
FTR, here's the sort_deps.log extract proving my assertion:
xserver-xorg Dep: (OR xserver-xorg-input-all xorgxrdp xserver-xorg-input-aiptek xserver-xorg-input-elographics xserver-xorg-input-evdev xserver-xorg-input-evdev-udeb xserver-xorg-input-joystick xserver-xorg-input-kbd xserver-xorg-input-libinput xserver-xorg-input-libinput-udeb xserver-xorg-input-mouse xserver-xorg-input-mtrack xserver-xorg-input-multitouch xserver-xorg-input-mutouch xserver-xorg-input-synaptics xserver-xorg-input-void xserver-xorg-input-wacom xserver-xorg-input-xwiimote ) soft_depend 1
xserver-xorg-input-all not included in a useful version, check_backports 0
xorgxrdp not included in a useful version, check_backports 0
xserver-xorg-input-aiptek not included in a useful version, check_backports 0
xserver-xorg-input-elographics not included in a useful version, check_backports 0
xserver-xorg-input-evdev not included in a useful version, check_backports 0
xserver-xorg-input-evdev-udeb not included in a useful version, check_backports 0
xserver-xorg-input-joystick not included in a useful version, check_backports 0
xserver-xorg-input-kbd not included in a useful version, check_backports 0
xserver-xorg-input-libinput not included in a useful version, check_backports 0
xserver-xorg-input-libinput-udeb is included already, acceptable version 0.29.0-1
Add the option to switch to format v2 which uses sha256 everywhere
instead of md5. We can't actually turn this on yet, need to wait for
users to upgrade a bit and get new v2-aware jigdo clients.
But this works as-is for generating v1 until we hit that switch - see
the new JIGOO_CHECKSUM option in CONF.sh
Mostly copied from the arm64 support, but also copy in the images like
previously for non-efi platforms.
Move the old boot-arm-common code back into boot-armel now it's no
longer common.
If COMPLETE=0 is set, the content should rather be
"bluray/not_complete" instead of just "not_complete". This would allow
one to build offline usable Edu BDs.
Steve McIntyre
Raphaël Hertzog
Samuel Thibault
bauen1
Cyril Brulebois
Wolfgang Schweer
Holger Levsen
John Paul Adrian Glaubitz