| Age | Commit message (Collapse) | Author |
|---|
|
we currently distinguish, by virtue of logic, multi-
and single-tree projects. this introduces a lot of
complexity, so the following patch has removed this
distinction. all single-tree projects are now multi
tree, but with only one tree.
other design changes:
config/git removed. rev/tree now handled with mkhelper
and target configs. this allows per-tree remote handling,
so we could (for example) add the mrchromebox coreboot
fork easily, for specific chromebook models.
config/project is now config/module/project. this
cleans up the main config/ directory.
cfgname variable: .config no longer hardcoded when
doing configs in projects. it is configurable with
this variable. if unset, the previous .cfg is used
noconfig variable: set this in target/mkhelper
configs, and config/module/project/tree/config/ is
unused; instead, make is called without config copying.
a "noconfig" placeholder is used. this enables cleaner
usage of previously single-tree projects in the new
multi-tree-only structure.
cfgvars: variables reinitialised in project config are
cleaned more reliably. in particular, several variables
e.g. autogenargs were not being reset between projects,
although the autogenargs variable is currently unused.
general cleanup has been conducted across the build
system, as part of this redesign. several parts of the
build system more over-engineered as a result of the
previous design, and the new design enabled a severe
reduction in sloccount (75 lines reduced in lbmk).
yes, this change *adds features* but there are fewer
lines of code than before. this is the best kind of
change. way better than the 3-line-diff. yes.
rev= and tree= are set more centrally in configs. the
mkhelper config can be used to set url/url_bkup, but
individual trees can override them.
configs have been cleaned up, with several variables
now much more heavily centralised. config/git/ was
used before to first download the bare repository, and
then the real clone was made with patches. this allows
git repositories to be cached for future re-builds.
with config/git/ gone, the download logic for projects
is more centrally implemented. a separate call to ./mk -f
is no longer performed, and in fact ./mk -f never needs
to be used anymore as a result. ./mk -f now is basically
the same thing as -b, just without any action performed.
this change will allow rapid expansion of the libreboot
project. i want to add hundreds of chromebooks to the
project, using mrchromebox, but the previous design did
not permit this. this change means that i can easily have
unlimited upstreams per project. coreboot will, in some
follow-up changes, have the following upstreams:
coreboot/default: coreboot.org
coreboot/fam15h: 15h.org fork
coreboot/chromebook: mrchromebox fork
this can now be done freely without massive changes to
the build system. under the previous design, i worked on
a now-abandoned patch to do N remotes per project, reading
repo links from a file. this meant that i'd have coreboot.org
first (main and backup), then mrchromebox. this meant that
lbmk would failover through to mrchromebox, wasting time
for the user.
this new design is much cleaner, and will enable the
project to be maintained much more easily.
Signed-off-by: Leah Rowe <leah@libreboot.org>
|
|
libreboot's policy is to thoroughly clean sources per
build, because we cache builds out of tree and avoid
re-builds if they exist; these and the sources are
deleted if metadata in config/ changes per project,
tree and/or target.
the problem is that this change only affects lbmk.git
it doesn't affect release archives, where project sources
do not have git-init applied to them. this will be
addressed.
can't use it on coreboot, because it will delete xgcc
tarballs in a way currently not handled, and handling
it will mean adding a project-specific hack.
the xgcc tarballs could be built out-of-tree, and this
will have to be done, otherwise coreboot will have to be
made gitclean="y" for now.
everything else is fair game.
Signed-off-by: Leah Rowe <leah@libreboot.org>
|
|
yes, make it generalised. use a new variable in
mkhelper.cfg, gitclean and gitcleanargs
not enabled by default, but u-boot enables it. i'm
considering whether to use this by default, as it
is the general policy of lbmk to clean everything
just after a build, as builds are cached out of
tree. this way, we can do re-builds safely for each
project. there are probably a lot of subtle bugs that
might be fixed if we used this
the problem is that this is currently only used on
lbmk.git, but not on libreboot release archives which
lack .git directories; the main one is initialised in
lbmk, but not in upstream sources. we could just initialise
those.
in fact, that is probably what i'm going to do.
Signed-off-by: Leah Rowe <leah@libreboot.org>
|
|
This reverts commit 5b4c9158e5a79f8d7e776c8c4ece69dda5aa8690.
|
|
On coreboot for example, as Mate has told me, if you're
making Kconfig changes and re-compiling, sometimes the
actual image that you build might still have the old one
in it, due to how coreboot's build system works.
To mitigate this, you can just always run distclean before
doing the build, but lbmk was doing just clean.
In practise, we did not find any issues, but this change should
be harmless, and might prevent such issues in the future. It's
even possible that we might have already encountered this before
and not realised, and we were just lucky that no noticeable issues
were caused.
It's *also* possible that the reverse is true: an issue that
was previously covered up, then that issue will now be exposed.
However, if that turns out to be true, then that is good because
we are exposing said bugs and then we will know to fix them!
Anyway, the variable in target.cfg is:
cleancmd="whatever_you_want"
e.g.
cleancmd="distclean"
You may also specify this in global mkhelper.cfg files, per
project; I've already done this for SeaBIOS, coreboot
and U-Boot, since all of these use Kconfig files.
Signed-off-by: Leah Rowe <leah@libreboot.org>
|
|
NOTE: Support added for xarch target x86_64-elf,
but U-Boot failed to build with this error:
OBJCOPY lib/efi_loader/helloworld.efi
x86_64-elf-objcopy: lib/efi_loader/helloworld_efi.so: invalid bfd target
make[2]: *** [scripts/Makefile.lib:476: lib/efi_loader/helloworld.efi] Error 1
Since I'm building U-Boot for x86_64 *on* an x86-64
host, and since that is currently the recommended type
of machine to use for lbmk development, and since the
other x86 payloads currently don't cross compile anyway,
this is an acceptable compromise for now. This is because
at present, I'm not making U-Boot the primary payload on x86,
instead preferring to chain it from GRUB and SeaBIOS.
The target.cfg file for x86 u-boot shows xarch/xtree commented.
Uncomment these to compile on crossgcc instead of hostcc.
I mention 64-bit because I initially did this first, but decided
to do 32-bit first. I'll work on the 64-bit one next (SPL).
It's only enabled in QEMU for now.
Signed-off-by: Leah Rowe <leah@libreboot.org>
|
|
certain code checks for build.list, to skip it, for
example in items()
we already use config/data/grub to store grub config data
that applied to all trees
create these directories too:
config/data/coreboot
config/data/u-boot
config/data/seabios
move the respective build.list files in here, and also
to config/data/grub
now multi-tree projects contain, per directory, just the
target.cfg file and the patches directory. this is much
cleaner, because some of the logic can be simplified more
Signed-off-by: Leah Rowe <leah@libreboot.org>
|
