| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| ... | |
| |/ |
|
| |\ |
|
| | |
| |
| |
| |
| |
| | |
This way we can be sure we're not attempting to build when instead
we should be bailing out (i.e. when the project's sources are
not available).
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The following will effectively disable any previous 'set -e':
foo_fail() {
(
set -e
false
true
)
}
if (set +e; foo_fail); then
echo "foo_fail() failed, unsuccessfully"
fi
Creating a subshell within an 'if' statement makes 'set -e'
non-functional.
|
| | |
| |
| |
| |
| |
| |
| | |
"update" actions were failing ever since revision b7fcc477 for
projects with git sources due to an error in
project_update_check_git() not returning 1 when it determined that the
project's sources was a git repo (to force an update, always).
|
| |\ \ |
|
| | | |
| | |
| | |
| | |
| | |
| | | |
The Libreboot build system is not fully automated, doesn't help with
the installation process per-se, and doesn't provide ROMs inside the
repository itself.
|
| |/ /
| |
| |
| |
| |
| | |
This fork of Paper is simply called the "Libreboot build system".
Addresses issue #613
|
| |\| |
|
| | |
| |
| |
| |
| | |
If the $install_path directory does not exist or a file which should
be present in the install directory isn't, fail.
|
| |/
|
|
|
| |
If the $build_path directory does not exist or a file which should be
present in the build directory isn't, fail.
|
| |\ |
|
| |/
|
|
|
|
|
|
|
| |
project_action_check() is there to avoid redundant actions from being
performed (e.g., if SeaBIOS is already built, don't build it
again). Since this is its primary purpose, we want to avoid erroring
out when it returns with a non-zero return code.
This addresses issue #614
|
| | |
|
| |\ |
|
| | |
| |
| |
| |
| |
| | |
If cd fails, nothing else should be done. This is to prevent a
potentially destructive action from being performed on the Libreboot
repository (if the script is executed from the same directory).
|
| | |
| |
| |
| |
| | |
Having a specific function for checking whether a patch would apply
isn't that useful if git_am and git_apply call it internally anyway.
|
| | |
| |
| |
| |
| |
| | |
git_diff_check() was added for future use since it's complementary to
git_diff_staged_check(). It could be used to check for a dirty
worktree, for example.
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
The recursive call in git_project_patch_recursive() is now performed
last so that if a patching attempt fails, the calling functions won't
attempt to apply more patches. The consequence of this change is that
patches are applied in (nearly) opposite order from before.
git_project_prepare_patch() is now called after
git_project_prepare_blobs() so that if patching fails and the user
chooses to perform additional project actions anyway, such as
building, the project is already deblobbed and not a risk to their
freedom. I had mistakenly reordered these function calls in a prior
commit.
|
| | |
| |
| |
| |
| |
| |
| |
| | |
Executing the project action within its own subshell makes it
easy to abort the action with a simple "exit" while still
allowing for the printing of the footer describing
failure/success. Prevously, if "exit" were called in a project
action the footer would not print.
|
| |\ \
| |/
|/| |
|
| | |
| |
| |
| |
| | |
The "let" builtin returns 1 if it's given an argument that evaluates
to zero.
|
| | | |
|
| |/ |
|
| |\ |
|
| | |
| |
| |
| |
| | |
LC_ALL is already set to "C.UTF-8" and exported if libfaketime path is
given in libreboot.conf
|
| | | |
|
| | | |
|
| |/
|
|
|
|
|
| |
Previous patches made certain tests unnecessary due to the setting of
$BRANCH_PREFIX in $branch. By adding the branch prefix right before
it's actually needed we can retain the usefulness of those
conditional tests.
|
| |\ |
|
| | |
| |
| |
| |
| | |
Cryptsetup is now using LUKS2 by default.
https://gitlab.com/cryptsetup/cryptsetup/commit/ae90497762bc4e3f04064e0ebbbde8c64bf27c4a
|
| |/ |
|
| |\ |
|
| | |
| |
| |
| |
| |
| |
| |
| |
| | |
The subshell, changing directory, and checking out was unecessary
considering all of these are done within the functions called by
git_patch()
git_am() was added in order to mirror the same wrapper style as used
by git_apply() and git_commit()
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
This simplifies git_project_patch_recursive() and provides a unified
method for patching, regardless of the patch format (mailbox or
diff).
Logic is included for dealing with patching failures due to
git_apply_check() not being enough to catch certain edge cases, such
as 'git apply --check' returning 0 but a subsequent 'git apply'
failing.
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Since changes from an applied diff need to be added to the index and
committed (vs. mailbox files with the git_patch function) new
functions were necessary to account for this difference.
This also fixes a bug where diffs were applied but not committed,
at minimum dirtying the work tree and potentially causing problems for
any later patches.
|
| |/
|
|
|
|
| |
The previous function, diff_patch_file, was more error-prone and
fragile and did not take into consideration patching of more than one
file at a time(!).
|
| |\ |
|
| | |
| |
| |
| |
| |
| |
| |
| |
| | |
Patching before removing blobs is cleaner since patches are applied
against a freshly created and checked out branch at the specified
revision. Applying patches first also seems to reduce the likelihood
of failed patching attempts due to a nebulous error about files not
matching the git index (but doesn't eliminate it). More debugging
required.
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Previously the only way patching would work is if for every target
there existed a corresponding directory under $project/$PATCHES, e.g:
"./libreboot update coreboot x200 8MiB corebootfb grub"
would require the "grub" directory and its parents:
"projects/coreboot/patches/x200/8MiB/corebootfb/grub"
Now you only need target-specific patch directories if you need them;
I believe this was the intended behavior.
Superfluous quotes were removed for readability.
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
There was a bug where only $project was used as the branch name due to
some faulty logic which checked for the existence of a revision file
before setting the variables "prepare_branch" and "prepare_path".
The bug in this case was that prepare_branch would be set to the value
of $branch in only the first iteration of the for loop since in almost
all cases there is only one revision file that exists for a given
project, regardless of the arguments provided. Explained another way,
in order for the proper branch name to be used a revision file would
have had to exist in every target directory.
This was an issue because only one branch was ever operated on (named
$project), meaning if actions were performed on many targets then only
the last to run would be represented in the project's
repository--making tracking down some bugs a bit harder.
With this fixed we now create a branch for every possible project
or tool configuration and leaves us with a log for each.
|
| | |
| |
| |
| |
| |
| |
| | |
This is a slight mitigation against branch name collisions. More
robust handling should be added in the future but this is simple
enough and provides some clear indication which branches have been
created by the build system.
|
| |/
|
|
|
| |
A patch failing to apply shouldn't leave the repository in such a
state where manual intervention is necessary to reset it.
|
| |\ |
|
| | |
| |
| |
| |
| |
| |
| |
| |
| | |
The 16MiB ROM is non-functional at the moment so an 8MiB target has
been added in order to avoid losing QEMU support while the bug is
identified and fixed.
Files were copied from the 16MiB directory with the exception of the
config file located at 8MiB/config which is new.
|
| | | |
|
| | |
| |
| |
| |
| |
| |
| |
| | |
The config for text mode doesn't actually have to be there (since
text mode is the default in the base config) but, as in other cases,
this configuration snippet is included for the sake of
documenting what part of the base config is related to the
target ("textmode" in this case).
|
| | |
| |
| |
| |
| | |
16MiB is the largest flash size for most of our supported boards so
any emulated boards should stick with that upper limit.
|
| | |
| |
| |
| |
| |
| |
| |
| |
| | |
The 16MiB ROM is non-functional at the moment so an 8MiB target has
been added in order to avoid losing QEMU support while the bug is
identified and fixed.
Files were copied from the 16MiB directory with the exception of the
config file located at 8MiB/config which is new.
|
| | | |
|
