The git helper uses gzip to compress the intermediate tarball. But gzip
removes the source file, and create a new file named by appending .gz to
the original file name.
Thus, we end up with output.gz, while the download wrapper expects jsut
output, and thus believes the downlaod failed.
Fix that by storing the tar from git to a temporary file, then pipe this
file to gzip's stdin, and redirect gzip's stdout to the output file.
Reported-by: Graham Newton <gnewton@peavey-eu.com>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Peter Seiderer <ps.report@gmx.net>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
This drastically simplifies the git helper, as it no longer has to deal
with atomically saving the downloaded archive.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Tested-by: Thomas De Schampheleire <thomas.de.schampheleire@gmail.com>
(Tested by running 'make fmc-fsl-sdk-source')
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
When switching the git helper over to a shell script, a special case was
not carried over: in case the remote has the required reference, we
attempt a shallow clone, using --depth 1. However, this is not supported
when the remote is accessed with the http protocol.
Therefore, the download fails.
What happened before the conversion to a shell script was that the helper
in the Makefile would fallback to doing a full-clone.
This is the case and behaviour that were lost in the conversion.
To avoid making the script too complex, we only attempt a full clone if
needed. And we decide that a full clone is needed by default; we decide
it is unnecessary if the remote has the needed reference *and* the
shallow clone was successful.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
DL_DIR can be a very precious place for some users: they use it to store
all the downloaded archives to share across all their Buildroot (and
maybe non-Buildroot) builds.
We do not want to trash this location with our temporary downloads (e.g.
git, Hg, svn, cvs repository clones/checkouts, or wget, bzr tep tarballs).
Turns out that we already have some kind of scratchpad, the BUILD_DIR.
Although it is not really a disposable location, that's the best we have
so far.
Also, we create the temporary tarballs with mktemp using the final tarball,
as template, since we want the temporary to be on the same filesystem as
the final location, so the 'mv' is just a plain, atomic rename(2), and we
are not left with a half-copied file as the final location.
Using mktemp ensures all temp file names are unique, so it allows for
parallel downloads from different build dirs at the same time, without
cloberring each downloads.
Note: we're using neither ${TMP} nor ${TMPDIR} since they are shared
locations, sometime with little place (eg. tmpfs), and some of the
repositories we clone/checkout can be very big.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Samuel Martin <s.martin49@gmail.com>
Cc: Arnout Vandecappelle <arnout@mind.be>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Cc: Thomas De Schampheleire <patrickdepinguin@gmail.com>
Tested-by: Thomas De Schampheleire <thomas.de.schampheleire@gmail.com>
[tested a particular scenario that used to fail: two separate builds
using a shared DL_DIR, ccache enabled, so that they run almost
synchronously. These would download the same file at the same time,
corrupting each other. With the patches in this series, all works
fine.]
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
The git download helper is getting a bit more complex. Fixing it in the
Makefile when it breaks (like the recent breakage with a non-existing
sha1-cset) proves to be challenging, to say the least.
Move it into a shell script in support/download/git, which will make
it much easier to read, maintain, fix and enhance in the future.
[Peter: redirect pushd/popd output to /dev/null]
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Peter Korsgaard <jacmet@uclibc.org>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Acked-by: Luca Ceresoli <luca@lucaceresoli.net>
Cc: Arnout Vandecappelle <arnout@mind.be>
Reviewed-by: Samuel Martin <s.martin49@gmail.com>
Reviewed-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
