Our git download backend switches the remote of our local clone, and
fetches all refs and tags from that remote.
When the local clone has a tag fetched from another remote, and the new
remote also has a tag by the same name, and that tag points to another
commit, then git refuses to fetch the new tag and exits in error, as it
considers that the new tag would clobber the existing one. This is safe
and sane behaviour when run interactively with a human that can take a
decision.
However, in our case, we don't care about any tags that were present
before, as only the last one makes sense in our case: the one from the
remote the user has requested for the current build.
Tell git to forcefully pull tags, even if they would clobber existing
ones.
Note that, although this changes the git backend, it does not change the
content of generated archives, so we do not need to bump the suffix
version.
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Signed-off-by: Arnout Vandecappelle <arnout@mind.be>
Files in a git repository can be given attributes, like the usual eol
that can convert to-from crlf, cr, lf; those are applied when comitting
or checking-out a file.
There are also two attributes that are meant to be used when generating
an archive (with git archive): export-subst, and export-ignore, that
respectively substitutes format placeholders in a file, and excludes a
file from the archive.
Some package (e.g. pcm-tools, luajit) use the export-subst attribute
to generate versioning information. luajit, specifically, uses the UNIX
timestamp of the commit as the patch-level for its semantic versioning.
We don't use git-archive, because we need to get submodules and LFS
blob, which git-archive does not handle. So, our git backend tries to
impersonate git-archive as much as possible, but the support for git
attributes was lost when we converted it from using git-archive to
manually creating the tarball in 3abd5ba424 (support/download/git: do
not use git archive, handle it manually) in preparation for f109e7eeb5
(support/download/git: add support for submodules) (arguably, a long
time ago...)
Extend the git backend to handle the export-subst attribute. There is
no git tool (that we could find) that does that automatically, except
git-archive, which we can't use; "git check-attr" however can report
whether a file has a specific attribute (and git check-attr can work
with \0-delimited fields and records).
So, we iterate over all the files in the repository, and filter those
that have the export-subst attribute set. Then for each file, we use a
bit of awk to do the replacement:
- for each line (managed natively by awk), we iterate over each
format placeholder,
- for each placeholer, we query "git log" with the requested format,
- we emit the replacement.
When doing the replacement, we decided to force abbreviating short
hashes to 40 chars, which is the length of a full sha1, rather than
actually abbreviating them:
- letting git decide of the length is not reproducible over time:
- as new commits are added, the short length will increase to avoid
collisions,
- newer git versions may decide on a different heuristic to shorten
hashes,
- users may have local settings with an arbitrary length (in their
~/.gitconfig for example);
- deciding on our side of an "small" arbitrary value would not be
viable long term either, as it might be too large to be minimum, or
too short to avoid collisions.
The only reproducible solution is to use unabbreviated hashes.
Handling git-attributes also implies that the format of the generated
archives has changed, since we now expand placeholders, so we bump our
git format version.
Hash files for all git-downloaded packages will be updated in followup
commits.
Of all our git-downloaded packages, 5 are affected, and their hashes
will be updated in a followup commit too:
- pcm-tools, which was known, and the one that triggered this commit;
since we now expand placeholders, we can drop the post-extract hook;
switching to a full hash in replacements also changes the hash of
the generated archive;
- qt5knx, qt5location, qt5mqtt, and qt5opcua: the file .tag at the
repository root, contains only the full hash placeholder; that file
is not used at all during the build (AFAICS);
Finally, a sixth package, luajit, uses export-subst; it currently relies
on the github-generated archive (because it happens to currently use a
format that is reproducible); it will also be converted in a follow-up
patch.
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: Woody Douglass <wdouglass@carnegierobotics.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Francois Perrad <fperrad@gmail.com>
Cc: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Arnout Vandecappelle <arnout@mind.be>
The quoting around the expansion of ${relative_dir} was indeed incorrect
since it was introduced back in 8fe9894f65 (suport/download: fix git
wrapper with submodules on older git versions): it is in fact already
quoted as part of the whole sed expression.
${GIT} can contain more than one item, but we don't care about splitting
on spaces when we just print it for debug, so we can just quote it
rather than add an exception.
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Since commit b7efb43e86 (download/git: try to recover from
utterly-broken repositories), we catch errors through an ERR
trap, so we can try and recover from a broken repository. In
that commit, we switched from using "set -e" to "set -E", so
that trap is inherited in functions, command substitutions,
and subshells.
However, the trap is not defined until we have parsed the
options, created the cache directory, and eventually chdir()ed
into it. Athough improbable, it is possible for the git helper
to fail in any of those steps, and that would not get caught.
Fix that
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
The original patch for commit cff428fe31 ("download/git: support Git
LFS") included a call to "git lfs install" but this was a problem as it
could modify ~/.gitconfig outside the dl/ tree. When this was
updated it was thought that the modification to gitconfig was
unnecessary because the LFS fetch and checkout steps are performed
manually.
Unfortunately, this is not correct and the LFS checkout fails with:
Cannot checkout LFS objects, Git LFS is not installed.
Add the call to "git lfs install", with the --local option so that only
the repository's .git/config is modified and not the user's global
~/.gitconfig.
This is also required for submodules as the parent repository's config
is not inherited.
Cc: Arnout Vandecappelle <arnout@mind.be>
Cc: Vincent Fazio <vfazio@xes-inc.com>
Signed-off-by: John Keeping <john@metanate.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Git Large File Storage replaces large files with text pointers in the
Git repository while storing the contents on a remote server. If a
repository is using this extension, then git-lfs must be used to
checkout the large files before the source archive is generated.
Signed-off-by: John Keeping <john@metanate.com>
[vfazio:
- add git-lfs to DL_TOOLS_DEPENDENCIES
- fixup for 5a0d681394
("infra/pkg-download: make the DOWNLOAD macro fully parameterised")
]
Signed-off-by: Vincent Fazio <vfazio@xes-inc.com>
[Arnout:
- don't "git lfs install";
- recurse into submodules.
]
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Commit 54d3d94b6e ("support/download: print
command used for download") broke the git and svn download helpers, because
these helpers have invocations of the _git/_svn commands where the exact
output matters.
For example for git, this would result in:
date: invalid date ‘GIT_DIR=.../dl/libyuv/git/.git git log -1 --pretty=format:%ci \n2019-04-12 17:48:45 +0000’
Detected a corrupted git cache.
Removing it and starting afresh.
Fix by splitting the _git function in two: _git and _plain_git.
The former echoes the command, and then calls the latter.
Most invocations use _git as before, but those cases where the output should
not be disturbed, directly call _plain_git.
For symmetry, all download helpers are aligned, even though only the git and
svn helpers were broken.
Fixes: #13631
Fixes:
http://autobuild.buildroot.org/results/c2f/c2fcd4aa6660e3c2f9c6f85646ca7dfe0db56040/
Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
[yann.morin.1998@free.fr: add bug report and autobuild failure]
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Even though that most download commands actually print some output, like
progress indication or other messages, the actual command used is not. This
makes it hard to analyze a build log when you are not fully familiar with
the typical output of said log.
Update the download helpers to do just that, respecting any quiet/verbose
flag so that a silent make (make -s) does not get more verbose.
Note: getting rid of the duplication of the command in the script is not
straightforward without breaking support for arguments with spaces.
Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
[yann.morin.1998@free.fr: use printf, not echo]
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Most 'verbose' variable inside the download helpers actually mean 'quiet'.
I.e. they are assigned in case quiet operation is requested, and empty in
case of non-quiet operation. Using the name 'verbose' for such a variable is
confusing, especially when you want to test the variable on emptiness or
non-emptiness (in a subsequent commit).
Signed-off-by: Thomas De Schampheleire <thomas.de_schampheleire@nokia.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Switch to using the tarball helper, that can generate reproducible
archives whatever the tar version >= 1.27.
However, those archives are not identical to the previous ones generated
in the (now-broken) gnu format.
To avoid any clashing between old and new archives, and new and old
Buildroot versions, we need to name the new generated archives
differently from the existing ones.
So, we bump the git-specific format-version to -br1.
The %ci date has been supported by git back to 1.6.0, released August
2008); it is not strictly ISO8601, but is still accepted as a PAX date
header. The strict ISO8601 placeholder, %cI, was only introduced with
2.2.0, release in November 2014, so too recent to be widely available.
As the format and the names of the archives changes, we need to update
all the hash files with the new names and hashes.
Of all the bootloaders that have a git download method, vexpress-firmware
is the only one to have a hash. Others have no hash files, or they have
explicitly set BR_NO_CHECK_HASH_FOR.
For the packages, linux-headers is the special snowflake, as the git
download is only for custom git tree, so it is excluded from the hash
verification with BR_NO_CHECK_HASH_FOR.
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: Vincent Fazio <vfazio@xes-inc.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Reviewed-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Reviewed-by: Vincent Fazio <vfazio@xes-inc.com>
---8<------8<------8<------8<---
#!/bin/sh
# Find and download all packages using git as backend.
# Manually fix hashes for affected packages.
# Packages that only have a host variant
HOST_ONLY='imx-mkimage|mxsldr|netsurf-buildsystem|opkg-utils|prelink-cross|qoriq-rcw|vboot-utils'
# Packages that have a non-git main _SOURCE, and/or which
# have BR_NO_CHECK_HASH_FOR for the git _SOURCE
NOT_GIT='aufs|aufs-util|xenomai|linux-headers'
export BR2_DL_DIR=$(pwd)/temp-dl-dir
make defconfig
make $( git grep -l -E 'SITE_METHOD[[:space:]]*:?=[[:space:]]*git\>|_SITE[[:space:]]*:?=[[:space:]]*git:' \
boot/vexpress-firmware/ package/ \
|sed -r -e 's,.*/([^/]+)\.mk,\1,' \
|sed -r -e '/^('"${NOT_GIT}"')$/d;' \
-e 's/^('"${HOST_ONLY}"')/host-\1/;' \
-e 's/$/-legal-info/;'
)
---8<------8<------8<------8<---
Older versions of git store the absolute path of the submodules'
repository as stored in the super-project, e.g.:
$ cat some-submodule/.git
gitdir: /path/to/super-project/.git/modules/some-submodule
Obviously, this is not very reproducible.
More recent versions of git, however, store relative paths, which
de-facto makes it reproducible.
Fix older versions by replacing the absolute paths with relative ones.
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
This reverts commit 6f35d96756.
Repeat after me: on the master branch you will not work. On the master
branch you will not work.
This definitely shouldn't have been pushed. Sorry about that.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Running "git fetch origin ${cset}:${cset}" to create a local ref
${cset} from the remote ref ${cset} causes Git to issue a warning like
the below, when the version is a full commit hash:
===
warning: refname '49eb4ecb1ef9879ebc6789a1bdb536ab2b1d9871' is ambiguous.
Git normally never creates a ref that ends with 40 hex characters
because it will be ignored when you just specify 40-hex. These refs
may be created by mistake. For example,
git switch -c $br $(git rev-parse ...)
where "$br" is somehow empty and a 40-hex ref is created. Please
examine these refs and maybe delete them. Turn this message off by
running "git config advice.objectNameWarning false"
===
This warning is very confusing for users, and is caused by the fact
that Git doesn't like our local ref name to look like a commit hash.
So, this commit proposes to fix the issue by having the local ref
named buildroot-${cset}, i.e
buildroot-${version-specified-by-the-package}.
The generated tarballs are exactly identical, nothing changes, it is
really just internally the local ref we are using to checkout the
correct version that is different. And it avoids the confusing
warning.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
'.' should be at the end of the sentence, not the beginning of a new
line.
Signed-off-by: John Keeping <john@metanate.com>
Cc: Yann E. MORIN <yann.morin.1998@free.fr>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
This file uses leading spaces, not TABs.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
We currently attempt a shallow clone, as tentative to save bandwidth and
download time.
However, now that we keep the git tree as a cache, it may happen that we
need to checkout an earlier commit, and that would not be present with a
shallow clone.
Furthermore, the shallow fetch is already really broken, and just
happens to work by chance. Consider the following actions, which are
basically what happens today:
mkdir git
git init git
cd git
git remote add origin https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
git fetch origin --depth 1 v4.17-rc1
if ! git fetch origin v4.17-rc1:v4.17-rc1 ; then
echo "warning"
fi
git checkout v4.17-rc1
The checkout succeeds just because of the git-fetch in the if-condition,
which is initially there to fetch the special refs from github PRs, or
gerrit reviews. That fails, but we just print a warning. If we were to
ever remove support for special refs, then the checkout would fail.
The whole purpose of the git cache is to actually save bandwidth and
download time, but in the long run. For one-offs, people would
preferably use a wget download (e.g. with the github macro) instead of
a git clone.
We switch to always doing a full clone. It is more correct, and pays off
in the long run...
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Cc: Arnout Vandecappelle <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
When a git tree has had sub-dir <-> sub-module conversions, or has had
submodules added or removed over the course of time, checking out a
changeset across those conversions/additions/removals may leave
untracked files, or may fail because of a conflict of type.
So, before we checkout the new changeset, we forcibly remove the
submodules. The new set of submodules, if any, will be restored later.
Ideally, we would use a native git command: git submodule deinit --all.
However, that was only introduced in git 1.8.3 which, while not being
recent by modern standards, is still too old for some enterprise-grade
distributions (RHEL6 only has git-1.7.1).
So, instead, we just use git submodule foreach, to rm -rf the submodules
directory.
Again, we would ideally use 'cd $toplevel && rm -rf $path', but
$toplevel was only introduced in git 1.7.2. $path has always been there.
So, instead, we just cd back one level, and remove the basename of the
directory.
Eventually, we need to get rid of now-empty and untracked directories,
that were parents of a removed submodule. For example. ./foo/bar/ was a
submodule, so ./foo/bar/ was removed, which left ./foo/ around.
Yet again, recent-ish git versions would have removed it during the
forced checkout, but old-ish versions (e.g. 1.7.1) do not remove it with
the forced checkout.
Instead we rely on the already used forced-forced clean of directories,
untracked, and ignored content, to really get rid of extra stuff we are
not interested in.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Cc: Arnout Vandecappelle <arnout@mind.be>
Reviewed-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Tested-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Force the checkout to ignore and throw away any local changes. This
allows recovering from a previous partial checkout (e.g. killed by
the user, or by a CI job...)
git checkout -f has been supported since the inception of git, so we
can use it without any second thought.
Also do a forced-forced clean, to really get rid of all untracked stuff.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Arnout Vandecappelle <arnout@mind.be>
Reviewed-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Tested-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
In some cases, the repository may be in a state we can't automatically
recover from, especially since we must still support oldish git versions
that do not provide the necessary commands or options thereof.
As a last-ditch recovery, delete the repository and recreate the cache
from scratch.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Cc: Arnout Vandecappelle <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Check that the given cset is indeed something we can checkout. If not,
then exit early.
This will be useful when a later commit will trap any failing git
command to try to recover the repository by doing a clone from scratch:
when the cset is not a commit, it does not mean the repository is broken
or what, and re-cloning from scratch would not help, so no need to trash
a good cache.
Reported-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Cc: Arnout Vandecappelle <arnout@mind.be>
Reviewed-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Tested-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
That way, we can pushd earlier, which will help with last-ditch recovery
in a followup commit.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Cc: Arnout Vandecappelle <arnout@mind.be>
Reviewed-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Tested-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
We really want the user not to use our git cache manually, or their
changes (committed or not) may eventually get lost.
So, add a warning file, not unlike the one we put in the target/
directory, to warn the user not to use the git tree.
Ideally, we would have carried this file in support/misc/, but the git
backend does not have access to it: the working directory is somewhere
unknown, and TOPDIR is not exported in the environment.
So, we have to carry it in-line in the backend instead.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Cc: Arnout Vandecappelle <arnout@mind.be>
Reviewed-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Tested-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Older versions of tar (e.g. 1.27.1) incorrectly interpret the escaping
of the regexp separator, and generate broken tarballs.
For example, given the following transform expression:
--transform="s/^\.\//squashfs-e38956b92f738518c29734399629e7cdb33072d3\//"
the resulting paths in the generated tarball would be:
squashfs-e38956b92f738518c29734399629e7cdb33072d3\/
i.e. a directory which last character is indeed a '\'.
We fix that by using a separator which is very unlikely to occur in a
filename.
Fixes:
http://autobuild.buildroot.org/results/742/7427f34e5c9f6d043b0fe6ad2c66cc0f31d2b24f/
and probably a slew of others as well...
Take this opportunity to fix indentation on the following line
(leading spaces, not TABs).
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
The different versions of git will behave in different ways when
fetching remote references, as summarised by the table below:
| ancient git | new git
--------------------------------------------------------------------
git fetch | fetch all refs but tags | fetches all refs but tags
git fetch -t | fetches only tags | fetch all refs and tags
(git-fetch may still fetch tags, but only if reachable from a branch)
So, to cover all the bases, we do a simple fetch, to be sure we have
branches, followed by the existing fetch -t, to get extra tags.
Fixes:
http://autobuild.buildroot.net/results/0a2/0a238a7f55ea56c33b639ad03ed5796143426889/build-end.log
Reported-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: Arnout Vandecappelle <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
There are cases where a repository might be broken, e.g. when a previous
operation was killed or otherwise failed unexpectedly.
We fix that by always initialising the repository, as suggested by
Ricardo. git-init is safe on an otherwise-healthy repository:
Running git init in an existing repository is safe. It will not
overwrite things that are already there. [...]
Using git-init will just ensure that we have the strictly required files
to form a sane tree. Any blob that is still missing would get fetched
later on.
Reported-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Reported-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: Arnout Vandecappelle <arnout@mind.be>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Acked-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Tested-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
git always look directories up until it finds a repository. In case
the git cache is broken, it may no longer be identified as a repository,
and git will look higher in the directories until it finds one.
In the default conditions, this would be Buildroot's own git tree
(because DL_DIR is a subdir of Buildroot), but in some situations may
very well be any repository the user has Buildroot in, like a
br2-external tree...
So, we force git to use our git cache and never look elsewhere, as
Suggested by Ricardo.
Use GIT_DIR, as it has been there for ages now, while --git-dir was
only introduced later (even if most distros ship an later version),
as suggested by Arnout.
Also fix the one call to git that was not using the wrapper.
Reported-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: Arnout Vandecappelle <arnout@mind.be>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Acked-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Tested-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
In case the git backend gets killed right in-between it finished
initialising the repository, but before it could add the remote,
we'd end up with a repository without the 'origin' remote, so we
would not be able to change its URL.
Another case that may happen (like in the build failure, below),
is that the repository was initialised with a previous version
of Buildroot, before the commit e17719264b (download/git: don't
require too-recent git) was applied, and that trepository was
still lying around...
Fixes:
http://autobuild.buildroot.org/results/25a/25aae054634368fadb265b97ebe4dda809deff6f/
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Peter Korsgaard <peter@korsgaard.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
git has supported -C only since 1.8.5, and some distros have not yet
caught up after more than 4 years...
Fall back to entering the directory.
Fixes:
http://autobuild.buildroot.net/results/35f9f7a4adc6c2cad741079e4afdf1408c94703b
Reported-by: André Hentschel <nerv@dawncrow.de>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: André Hentschel <nerv@dawncrow.de>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
When a package contains a relative symlink which first component is '..'
(thus pointing one directory higher), for example package 'meh' contains
this symlink:
foo/bar -> ../buz
then it would be stored as 'meh-version./buz' because of the
transform-name pattern replacement.
Fix it to only match the leading './'.
Reported-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Cc: Peter Korsgaard <peter@korsgaard.com>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Reviewed-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Tested-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Commit "6d938bcb52 download: git: introduce cache feature" introduced a
typo that makes the tarball to contain files without the package
basename:
$ tar -tvf good-a238b1dfcd825d47d834af3c5223417c8411d90d.tar.gz
-rw-r--r-- 0/0 8 2017-10-14 02:10 ./file
Historically, all tarballs are generated with the basename:
$ tar -tvf good-a238b1dfcd825d47d834af3c5223417c8411d90d.tar.gz
-rw-r--r-- 0/0 8 2017-10-14 02:10 good-a238b1dfcd825d47d834af3c5223417c8411d90d/file
The hashes in the tree were calculated with the basename.
In the most common scenario, after the download ends the tarball is
generated, the hash mismatches and the download mechanism falls back to
use the tarball from http://sources.buildroot.net .
The problem can be reproduced by forcing the download of any git package
PKG that has a hash file to check against:
$ make defconfig
$ ./utils/config --set-str BR2_BACKUP_SITE ""
$ BR2_DL_DIR=$(mktemp -d) make PKG-dirclean PKG-source
Fix the typo so the basename is really added to the files, that was
clearly the intention of the code.
Signed-off-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Cc: Arnout Vandecappelle <arnout@mind.be>
Cc: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: Peter Korsgaard <peter@korsgaard.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Yann E. MORIN <yann.morin.1998@free.fr>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Now we keep the git clone that we download and generates our tarball
from there.
The main goal here is that if you change the version of a package (say
Linux), instead of cloning all over again, you will simply 'git fetch'
from the repo the missing objects, then generates the tarball again.
This should speed the 'source' part of the build significantly.
The drawback is that the DL_DIR will grow much larger; but time is more
important than disk space nowadays.
Signed-off-by: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Currently all download helpers accepts the local output file, the remote
locations, the changesets and so on... as positional arguments.
This was well and nice when that's was all we needed.
But then we added an option to quiesce their verbosity, and that was
shoehorned with a trivial getopts, still keeping all the existing
positional arguments as... positional arguments.
Adding yet more options while keeping positional arguments will not be
very easy, even if we do not envision any new option in the foreseeable
future (but 640K ought to be enough for everyone, remember? ;-) ).
Change all helpers to accept a set of generic options (-q for quiet and
-o for the output file) as well as helper-specific options (like -r for
the repository, -c for a changeset...).
Maxime:
Changed -R to -r for recurse (only for the git backend)
Changed -r to -u for URI (for all backend)
Change -R to -c for cset (for CVS and SVN backend)
Add the export of the BR_BACKEND_DL_GETOPTS so all the backend wrapper
can use the same option easily
Now all the backends use the same common options.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Maxime Hadjinlian <maxime.hadjinlian@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Reviewed-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Reviewed-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Force gzip compression level 6 when calculating hash of a downloaded GIT repo.
To make sure the tar->gzip->checksum chain always provides consistent result.`
The script was relying on the default compression level, which must not be
necessarily consistent among different gzip versions. The level 6 is gzip's
current default compression level.
Signed-off-by: Petr Kulhavy <brain@jikos.cz>
Acked-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The removal of the .git dir before creating the tarball is not anymore
just an optimization. It is necessary to make the tarball reproducible.
Also, without the removal, large tarballs (gigabytes) would be created
for some linux trees.
Update the comment accordingly.
Reported-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
On most distros, the tar format defaults to GNU. However, at build time
the default format may be changed to posix. Also, future versions of
tar will default to posix.
Since we want the tarballs created by the git download method to be
reproducible (so their hash can be checked), we should explicitly
specify the format. Since existing tarballs on sources.buildroot.org
use the GNU format, and also the existing hashes in the *.hash files
are based on GNU format tarballs, we use the GNU format.
In addition, the Posix format encodes atime and ctime as well as mtime,
but tar offers no option like --mtime to override them. In the GNU
format, atime and ctime are only encoded if the --incremental option is
given.
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Cc: Peter Seiderer <ps.report@gmx.net>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Currently, there are two failure paths in the wrapper:
- if the tar fails, then the error is ignored because it is on the
left-hand-side of a pipe;
- if the find fails, then the error is ignored because it is a
process substitution (and there is a pipe, too).
While the former could be fixed with "set -o pipefail", the latter can
not be fixed thusly and we must use an intemediate file for it.
So, fix both issues by using intermediate files, both to generate the
list of files to include in the archive, and generate the archive in a
temporary tarball.
Fixes the following build issue, where the find is failing for whatever
unknown reason:
http://autobuild.buildroot.net/results/20f/20fd76d2256eee81837f7e9bbaefbe79d7645ae9/
And this one, where the process substitution failed, also for an unknown
reason:
http://autobuild.buildroot.org/results/018/018971ea9227b386fe25d3c264c7e80b843a9f68/
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
This adds support to pass options to the underlying command that is used
by downloader. Useful for retrieving data with server-side checking for
user login or passwords, use a proxy or use specific options for cloning
a repository via git and hg.
Signed-off-by: Romain Perier <romain.perier@free-electrons.com>
Acked-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The "--no-patch" option used by the git downloader appeared on git
1.8.4. Systems with older git versions show an error and fall back to
the wget downloader, which isn't suitable for all the cases.
Signed-off-by: Enrique Ocaña González <eocanha@igalia.com>
Tested-by: Matthew Weber <matthew.weber@rockwellcollins.com>
Tested-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Acked-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Some git repositories may be split into a master repository and
submodules. Up until now, we did not have support for submodules,
because we were using bare clones, in which it is not possible to
update the list of submodules.
Now that we are using plain clones with a working copy, we can retrieve
the submdoules.
Add an option to the git download helper to kick the update of
submodules, so that they are only fetched for those packages that
require them. Also document the existing -q option at the same time.
Submodules have a .git file at their root, which contains the path to
the real .git directory of the master repository. Since we remove it,
there is no point in keeping those .git files either.
Note: this is currently unused, but will be enabled with the follow-up
patch that adds the necessary parts in the pkg-generic and pkg-download
infrastructures.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Tested-by: Matt Weber <matt@thewebers.ws>
Reviewed-by: Matt Weber <matt@thewebers.ws>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
We currently use git-archive to generate the tarball. This is all handy
and dandy, but git-archive does not support submodules. In the follow-up
patch, we're going to handle submodules, so we would not be able to use
git-archive.
Instead, we manually generate the archive:
- extract the tree to the requested cset,
- get the date of the commit to store in the archive,
- store only numeric owners,
- store owner and group as 0 (zero, although any arbitrary value would
have been fine, as long as it's a constant),
- sort the files to store in the archive.
We also get rid of the .git directory, because there is no reason to
keep it in the context of Buildroot. Some people would love to keep it
so as to speed up later downloads when updating a package, but that is
not really doable. For example:
- use current Buildroot
- it would need foo-12345, so do a clone and keep the .git in the
generated tarball
- update Buildroot
- it would need foo-98765
For that second clone, how could we know we would have to first extract
foo-12345 ? So, the .git in the archive is pretty much useless for
Buildroot.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Tested-by: Matt Weber <matt@thewebers.ws>
Reviewed-by: Matt Weber <matt@thewebers.ws>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Currently, we are using bare clones, so as to minimise the disk usage,
most notably for largeish repositories such as the one for the Linux
kernel, which can go beyond the 1GiB barrier.
However, this precludes updating (and thus using) the submodules, if
any, of the repositories, as a working copy is required to use
submodules (becaue we need to know the list of submodules, where to find
them, where to clone them, what cset to checkout, and all those is
dependent upon the checked out cset of the father repository).
Switch to using /plain/ clones with a working copy.
This means that the extra refs used by some forges (like pull-requests
for Github, or changes for gerrit...) are no longer fetched as part of
the clone, because git does not offer to do a mirror clone when there is
a working copy.
Instead, we have to fetch those special refs by hand. Since there is no
easy solution to know whether the cset the user asked for is such a
special ref or not, we just try to always fetch the cset requested by
the user; if this fails, we assume that this is not a special ref (most
probably, it is a sha1) and we defer the check to the archive creation,
which would fail if the requested cset is missing anyway.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Tested-by: Matt Weber <matt@thewebers.ws>
Reviewed-by: Matt Weber <matt@thewebers.ws>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The way we use it, gzip will store the current time in the header, which
leads to unreproducible archives.
Fix that by telling gzip to not store the name and date of the file it
compresses, with the -n option. Since it compresses its stdin, there was
already no filename stored; now there's even no date stored.
Note: gzip has had -n since at least 1.2.4, released in 1993, so
virtually every gzip out there nowadays has it.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Some users may provide custom download commands with spaces in their
arguments, like so:
BR2_HG="hg --config foo.bar='some space-separated value'"
However, the way we currently call those commands does not account
for the extra quotes, and each space-separated part of the command is
interpreted as separate arguments.
Fix that by calling 'eval' on the commands.
Because of the eval, we must further quote our own arguments, to avoid
the eval further splitting them in case there are spaces (even though
we do not support paths with spaces, better be clean from the onset to
avoid breakage in the future).
We change all the wrappers to use a wrapper-function, even those with
a single call, so they all look alike.
Note that we do not single-quote some of the variables, like ${verbose}
because it can be empty and we really do not want to generate an
empty-string argument. That's not a problem, as ${verbose} would not
normally contain space-separated values (it could get set to something
like '-q -v' but in that case we'd still want two arguments, so that's
fine).
Reported-by: Thomas De Schampheleire <patrickdepinguin@gmail.com>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Thomas De Schampheleire <patrickdepinguin@gmail.com>
Reviewed-by: Thomas De Schampheleire <thomas.de.schampheleire@gmail.com>
Tested-by: Thomas De Schampheleire <thomas.de.schampheleire@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
When specifying BR2_LINUX_KERNEL_CUSTOM_REPO_VERSION, a user may want to
specify the SHA of a reference different than a branch or tag.
For instance, Gerrit stores the patchsets under refs/changes/xx/xxx, and
Github stores the pull requests under refs/pull/xxx/head.
When cloning a repository with --bare, you don't fetch these references.
This patch uses --mirror for a full clone, in order to give the user
access to all references of the Git repository.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Reviewed-by: "Maxime Hadjinlian" <maxime.hadjinlian@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
In 50c8b7e (support/download: support -q in all download backends), the
backend were made to respect the quietness of the main Makefile, when -s
is poassed on the 'make' command line. In doing so, they were all made
to be verbose by default.
However, the verbosity of some of the tools, like scp, is very high, and
is in fact intended for debug purposes.
Drop being verbose by default, just use whatever each tool deems normal
output. Only respect the quietness requested by the user.
Reported-by: Thomas De Schampheleire <patrickdepinguin@gmail.com>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Reviewed-by: Thomas De Schampheleire <thomas.de.schampheleire@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Following commit 95a572282e (pkg-infra: move the git download helper to a
script, 2014-07-02), move the comment describing the shallow clone trickery as
well. Merge this comment with the existing helper comment that was added in
7e40a1103a (support/download: convert git to use the wrapper, 2014-08-03).
Rename $($(PKG)_DL_VERSION) to ${cset} to match the helper code context.
Cc: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
We expresely call printf in the git helper, calls which were not
addresed in the previous silent-build patchset.
Just redirect stdout to oblivion when being silent.
Reported-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Acked-by: Fabio Porcedda <fabio.porcedda@gmail.com>
Tested-by: Fabio Porcedda <fabio.porcedda@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Add an option flag to all backends, as well as the check-hash script, so
as to silence download helpers when the user wants a silent build.
Additionaly, make the default be verbose.
Inspired by Fabio's patch on git/svn.
[Thomas: fix a typo "Environemnt" -> "Environment"
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Fabio Porcedda <fabio.porcedda@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Not all systems have /bin/bash (e.g. NixOS[1] doesn't). Buildroot
already uses /usr/bin/env shebangs for other interpreters (perl,
python), so why not bash?
This changes only the shebangs used by Buildroot itself; stuff installed
to the target system is left unchanged.
With this applied I can run Buildroot unmodified on NixOS.
[1]: http://nixos.org/
Signed-off-by: Bjørn Forsman <bjorn.forsman@gmail.com>
Acked-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
