There was some manual check to know if the local repository was really a clone
of the one specified in PKGBUILD. This check has been removed because it is
buggy and not necessary.
It is buggy because this check needs to be semantic, not a simple string
comparison. For example, I was blocked from building a PKGBUILD because Bazaar
was returning two different strings for the same location (for HTTP one was
url-encoded while the other was not, and for local paths one was absolute while
the other was relative). While this may be a bug in Bazaar, the check is
unreliable since the comparison is not semantic (http://foo.com/%2Bplus and
http://foo.com/+plus obviously refer to the same location for example).
Specially, it is useless because the intention is updating the existing local
clone. However, if the local clone is not a real clone of the repository
specified in PKGBUILD (which was what this buggy check tried to tell), next step
which is a pull operation will fail anyway. This is because bzr pull does not
perform merges, it just makes one branch into a mirror of another.
There was a reason provided when this manual check was added for Git, but no
reason provided for copying such check when Bazaar support was added, see
commits below. In fact, Mercurial lacks such manual check.
* c926c39b04
* 3b02f80dcb
Fixes FS#43448.
Signed-off-by: Allan McRae <allan@archlinux.org>
Commit 9e5e86aa was supposed to fix this. Instead I picked another [[ -f ]]
statement in the same region and added the hardlink test to it instead, thus
not fixing the bug.
Signed-off-by: Allan McRae <allan@archlinux.org>
This fixes the issue where if the user explicitly set the name of the cloned source
to eg. foo.git, the directory name in $SRCDEST would be foo.git as expected, but the
clone in $srcdir would be stripped of the .git suffix.
Signed-off-by: Lukáš Jirkovský <l.jirkovsky@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
If an already used array is passed array_build, some entries from the old
array could be carried over if the old array was longer than the new one.
Clear the destination array before adding elements to it to prevent this
issue.
Fixes: https://bugs.archlinux.org/task/43387
Signed-off-by: Allan McRae <allan@archlinux.org>
I suspect this is just wrong -- you never need to quote the replacement
side of a PE. In bash 4.3, this is essentially a no-op, but because of
a bug in bash 4.2, we get embedded quotes as a result of this
replacement. The relevant changelog item in bash is:
Fixed a bug that caused single quotes that resulted from $'...' quoting
in the replacement portion of a double-quoted ${word/pat/rep} expansion
to be treated as quote characters.
But this doesn't apply to us. Let's just drop the excessive quoting...
Signed-off-by: Allan McRae <allan@archlinux.org>
When a shared library uses an absolute symlink for its its .so file, the check
if the shared version of a static library exists fails. Test for the presence
of a broken symlink too.
Signed-off-by: Allan McRae <allan@archlinux.org>
This bombs out when "$trusted" expands to the empty string. We're
better off passing the var by name and letting bash default to "0" when
the var is empty
Fixes: https://bugs.archlinux.org/task/43269
Signed-off-by: Allan McRae <allan@archlinux.org>
We fixed this up to check architecture specific sources in ec679e09b2,
but fudged the array name in the in_array call.
Signed-off-by: Allan McRae <allan@archlinux.org>
Previously, we used a single boolean value to determine correlation of
sources to checksums. Since the introduction of arch-specific sources,
this is no longer sufficient, as we must ensure that we have checksums
for (potentially) multiple source arrays.
This change inlines the logic of have_sources to build an associative
array of source array names, unsetting them as we discover their
checksums. The error condition then becomes a non-empty correlation
array.
Fixes: https://bugs.archlinux.org/task/43192
Signed-off-by: Allan McRae <allan@archlinux.org>
We validated all sources when making a source package, whether or not they
are included in the tarball.
Signed-off-by: Allan McRae <allan@archlinux.org>
Before this, we'd see bizzare behavior of:
-> Adding changelog file (systemd.install)...
And, changelog files in the global section would not be added at all.
The code is clearly wrong here, as it references 'install' within a
loop of 'changelog' and 'install'. Let's use parameter indirection to
ensure that the proper file is identified and added.
Signed-off-by: Allan McRae <allan@archlinux.org>
File in noextract should still be symlinked into $srcdir so that they
can be accessed without using $SRCDEST. Using noextract on VCS files
makes no sense as these are not being extracted, so now this does
nothing.
Signed-off-by: Allan McRae <allan@archlinux.org>
This matches the behaviour with non-VCS sources. It also allows incremental
builds when subversion is used to obtain sources.
Signed-off-by: Lukáš Jirkovský <l.jirkovsky@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
bzr does not recognize bare ssh:// urls.
Fixes FS#41811
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
The local changes are discarded when updating. This matches the behaviour
when non-VCS sources are used. It also allows incremental builds.
This also changes the checkout during bzr source "extraction" to a heavyweight
checkout so that pulling a specific revision does not alter the original
download.
Original-work-by: Lukáš Jirkovský <l.jirkovsky@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
The local changes are discarded when updating. This matches the behaviour
when non-VCS sources are used. It also allows incremental builds.
Signed-off-by: Lukáš Jirkovský <l.jirkovsky@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
Previously the sources were dowloaded in HEAD revision in the download_svn().
If a specific revision was requested in fragment, the code was updated to that
revision in extract_svn(). However, because SVN is a centralized system,
this means that the changed sources has to be downloaded again.
By moving the fragment handling to download_svn(), we get the correct revision
without having to download it later in extract_svn().
Signed-off-by: Lukáš Jirkovský <l.jirkovsky@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
The local changes are discarded when updating. This matches the behaviour
when non-VCS sources are used. It also allows incremental builds.
Signed-off-by: Lukáš Jirkovský <l.jirkovsky@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
Similar to .PKGINFO, .SRCINFO provides structured metadata from the
PKGBUILD to be included with source packages.
The format is structured such that it contains a "pkgbase" and one to
many "pkgname" sections. Each "pkgname" section represents an "output
package", and inherits all of the attributes of the "pkgbase" section,
and then can define their own additive fields.
For example, a simple PKGBUILD:
pkgbase=ponies
pkgname=('applejack' 'pinkiepie')
pkgver=1.2.3
pkgrel=1
arch=('x86_64' 'i686')
depends=('friendship' 'magic')
build() { ...; }
package_applejack() {
provides=('courage')
...;
}
package_pinkiepie() {
provides=('laughter')
...;
}
Would yield the following .SRCINFO file:
pkgbase = ponies
pkgdesc = friendship is magic
pkgver = 1.2.3
pkgrel = 1
arch = x86_64
arch = i686
depends = friendship
depends = magic
pkgname = applejack
provides = courage
pkgname = pinkiepie
provides = laughter
The code to generate this new file is taken a project which I've been
incubating[0] under the guise of 'mkaurball', which creates .AURINFO
files for the AUR. AURINFO is the exactly same file as .SRCINFO, but
named as such to make it clear that this is specific to the AUR.
Because we're parsing shell in the packaging functions rather than
executing it, there *are* some limitations, but these only really crop
up in more "exotic" PKGBUILDs. Smoketesting[1] for accuracy in the Arch
repos yields 100% accuracy for [core] and [extra]. [community] clocks in
at ~98% accuracy (.3% difference per PKGBUILD), largely due to silly
haskell packages calling pacman from inside the PKGBUILD to determine
dependencies. [multilib] currently shows about 92% accuracy -- a
statistic which can be largely improved by utilizing the recently merged
arch-specific attribute work. This is also a smaller repo so the numbers
are somewhat inflated. In reality, this is only a .8% variance per
PKGBUILD.
Together, we can make PKGBUILD better.
[0] https://github.com/falconindy/pkgbuild-introspection
[1] https://github.com/falconindy/pkgbuild-introspection/blob/master/test/smoketest
Signed-off-by: Allan McRae <allan@archlinux.org>
We can avoid setting a default value for epoch since we intend to mean
unset and "0" as the same thing. This is also a more consistent default
as the display of epoch=0 is no epoch at all in the full package
version.
The extra paranoia in get_full_version can be removed due to lint_epoch
guarding against non-integer values of epoch.
Signed-off-by: Allan McRae <allan@archlinux.org>
This bug isn't currently exposed by any of the existing codepaths, but
an upcoming patch to introduce SRCINFO files to makepkg will expose
this.
Signed-off-by: Allan McRae <allan@archlinux.org>
This regression snuck in during some reviewing of 963f7fe02f
(arch-specific sources). We must always check the source=() array for
sources.
Signed-off-by: Allan McRae <allan@archlinux.org>
Interesting attributes created with 'local' or 'declare' won't be
surfaced in .PKGINFO, so we shouldn't try to look for them.
Signed-off-by: Allan McRae <allan@archlinux.org>
Rather than implementing suffix matching, which might clash, let's just
print the full fingerprint of the err'ing key so that the user can
copy/paste it into validpgpkeys. Also, make it clear in the manpage
that validpgpkeys needs full fingerprints, and nothing else.
Signed-off-by: Allan McRae <allan@archlinux.org>
I found this feature confusing, and the documentation wasn't any help.
It was pointed out to me on IRC that validpgpkeys expects full
fingerprints, and won't accept shorter forms. This makes the
documentation insufficient, and the variable name itself misleading.
This patch bolsters the documentation to explain more about what the
contents should be, and implements suffix matching to allow matching on
shorters fingerprint suffices. Now, when makepkg tells you that a key
ID isn't valid, it's sufficient to manually check the key ID against
the known good ID, and add it as is to validpgpkeys.
Signed-off-by: Allan McRae <allan@archlinux.org>
This commit changes the few remaining instances of:
[[ ! $foo = "$bar" ]]
to the more common:
[[ $foo != "$bar" ]]
Signed-off-by: Allan McRae <allan@archlinux.org>
This implements support for declarations such as:
arch=('i686' 'x86_64')
...
source=("somescript.sh")
source_i686=("http://evilmonster.com/i686/ponies-9001-1.i686.bin")
source_x86_64=("http://evilmonster.com/i686/ponies-9001-1.x86_64.bin")
md5sums=('d41d8cd98f00b204e9800998ecf8427e')
md5sums_i686=('e4ca381035a34b7a852184cc0dd89baa')
md5sums_x86_64=('4019740e6998f30a3c534bac6a83f582')
Just the same as the "untagged" sources, multiple integrity algorithms
are supported. The manpage is updated to reflect support for these
suffices.
This commit also refactors download_sources slightly:
1) to use the otherwise preferred convention of lowercase local variable
names, and to make the handling of $1 more clear.
2) rename the "fast" parameter to "novcs", to make it more clear what
this token does.
3) add a new possible token "allarch" to ensure that download_sources
will fetch all sources, for all architectures.
This also fixes a "bug" in which a PKGBUILD without any source array
would generate "md5sums=()". While not technically wrong, we can easily
do better and emit nothing at all.
We rely on values in the arch array to be valid as part of variable
names, so extend the arch lint check to catch this.
This also cleans up lint_arch to restrict the use of "lint" only to the
package-specific architecture checks. It previously had an odd
declaration with a conditional expansion that would never be true.
Since source package creation is architecture independent, we should
ignore architecture-dependent behaviors such as the lint check which
will halt execution when the host machine is not a supported arch.
https://github.com/falconindy/pkgbuild-introspection/issues/15
This eval enables the following in a PKGBUILD to "just work":
source=('$pkgname-$pkgver.tar.gz'::'https://host/$pkgver.tar.gz')
This has at least two problems:
- It violated the principle of least surprise.
- It could be a security issue since URLs are arbitrary input.
Instead, expand the dlagent command line into an array, replace the %o,
%u place holders, and run the resultant command line as is.
Embedded spaces in the DLAGENTS entry can be escaped with a backslash.
Fixes FS#41682
Signed-off-by: Allan McRae <allan@archlinux.org>
Git has the ability to use helper applications for interfacing with hg,
and from what we had before, the following url::
foo::git+hg::http://foo.bar/foobar
would get converted to something along the lines of:
filename: foo
URL: http://foo.bar/foobar
and the 'git+hg' part would essentially be ignored when it's getting set
up in the 'get_protocol' and 'get_downloadclient' functions. With this
patch it is possible to have a source link with '::' in it, however it
is not possible to have a filename with '::', which is the current
behavior.
Signed-off-by: Allan McRae <allan@archlinux.org>
This introduces support for architecture-specific conflicts, depends,
optdepends, makedepends, replaces, and conflicts by appending "_$CARCH"
to the array name. For example, in the global section:
arch=('i686' 'x86_64')
depends=('foo')
depends_x86_64=('bar')
This will generate depends of 'foo' and 'bar' on x86_64, but only 'foo'
on i686. Moreover, this is supported in the package functions with the
same heuristics as the generic names, e.g.
...
arch=('i686' 'x86_64')
depends=('foo')
...
package_somepkg() {
depends_x86_64=('bar')
...
}
Again, will cause x86_64 to have depends of 'foo' and 'bar', but only
'foo' for i686.
Break apart each of the blocks into their own separate functions. And,
instead of the hand crafted eval statements, reuse the logic from
pkgbuild-introspection[0] to abstract away the complexities of parsing
bash.
This commit fixes at least 3 bugs in check_sanity:
1) The wrong variable is shown for the error which would be thrown
when, e.g. pkgname=('foopkg' 'bar^pkg')
2) The "arch" variable is not sanity checked when the PKGBUILD has
an arch override, but only one output package.
3) https://bugs.archlinux.org/task/40361
Lastly, there's some string changes here which should help to clarify
a few errors emitted in the linting process.
[0] https://github.com/falconindy/pkgbuild-introspection
Dependencies are now handled with --nobuild unless specificially skipped.
Using --verifysource will skip dependency checks unless --syncdeps is
specified.
Fixes FS#35057 and FS#36999.
Signed-off-by: Allan McRae <allan@archlinux.org>
Add an array VCSCLIENTS to makepkg.conf that matches vcs source protocols
to the package containing the software needed for handling the source.
Signed-off-by: Allan McRae <allan@archlinux.org>
The documentation very clearly states that the buildfile has to be in
$PWD, but the error thrown by makepkg reference some mysterious "build
directory". Simplify this check so that we more directly check that the
file being referred to is in fact in our $PWD. Revise the error message
when the check fails to more plainly point out the problem.
Signed-off-by: Allan McRae <allan@archlinux.org>
