This partially reverts changes done in commit 72cc44958e ("treewide:
mark selected packages nonshared") as it removes the nonshared flag, but
keeps the PKG_RELEASE as the PKG_RELEASE bump while adding nonshared
flag was incorrect.
Unmark uci, ubus, libubox, lua, libnl-tiny and libjson-c as nonshared
packages as this fix attempt didn't worked out. Currently the
imagebuilder is broken again:
openwrt-imagebuilder-21.02.0-rc3-ipq40xx-generic.Linux-x86_64$ make image PROFILE=avm_fritzbox-7530 PACKAGES=luci-ssl-openssl
...
Collected errors:
* pkg_hash_check_unresolved: cannot find dependency libiwinfo20210430 for luci-mod-status
* pkg_hash_fetch_best_installation_candidate: Packages for luci-mod-status found, but incompatible with the architectures configured
* pkg_hash_check_unresolved: cannot find dependency libiwinfo20210430 for rpcd-mod-iwinfo
* pkg_hash_fetch_best_installation_candidate: Packages for rpcd-mod-iwinfo found, but incompatible with the architectures configured
* satisfy_dependencies_for: Cannot satisfy the following dependencies for luci-ssl-openssl:
* libiwinfo20210430
* opkg_install_cmd: Cannot install package luci-ssl-openssl.
Everything because iwinfo's ABI was changed two times since rc3 release:
+IWINFO_ABI_VERSION:=20210430
+IWINFO_ABI_VERSION:=20210420
Since iwinfo is marked as nonshared, it wasn't built by phase2 builders, but
luci-mod-status was already updated 2 times since rc3 and was thus rebuilt by
phase2 builders:
d1d452ed2fb3 luci-mod-status: don't set '-' hostname when creating static lease
95b3633055c1 luci-mod-status: switch to html table for wlan channel analysis
So now luci-mod-status depends on libiwinfo20210430 but only
libiwinfo20210106 can be downloaded. This is first part of the fix, in
the upcoming commit Jo is going to remove nonshared flag from iwinfo
package as well.
References: https://lists.infradead.org/pipermail/openwrt-devel/2021-July/035736.html
References: https://lists.infradead.org/pipermail/openwrt-devel/2021-July/035741.html
Acked-by: Jo-Philipp Wich <jo@mein.io>
Reported-by: Nick Hainke <vincent@systemli.org>
Signed-off-by: Petr Štetiar <ynezz@true.cz>
Mark uci, ubus, libubox, lua, libnl-tiny and libjson-c
as nonshared packages. This helps to keep coherent dependencies
if these ABI versioned packages are later updated.
Before this commit it is possible to get missing dependencies
in target-specific nonshared packages (like iwinfo) that depend
on these shared ABI versioned packages. If these are later updated
and rebuilt, only the new ABI version will be available for download,
while the target-specific packages in releases continue to depend on
the old ABI version.
After this commit the packages are built along the other nonshared
packages by the phase1 images buildbot and will be available at the
target/ download directories instead of packages/base dir. That will
help to keep a coherent set available.
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
For glibc, lua needs an explicit link to libdl as glibc has it separate
Fixes the following error in at least collectd:
ld: usr/lib/liblua.so: undefined reference to `dlopen'
ld: usr/lib/liblua.so: undefined reference to `dlclose'
ld: usr/lib/liblua.so: undefined reference to `dlerror'
ld: usr/lib/liblua.so: undefined reference to `dlsym'
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Compilation of liblua itself works, but when other packages link against
it, the linker starts throwing undefined references to a bunch of math
functions in libm.
First discovered in a failed attempt to transition a package to uClibc++.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
[fix commit title capitalization]
Signed-off-by: David Bauer <mail@david-bauer.net>
Since the binaries for both lua as well as lua5.3 contain the version
number, invocations of the "lua" binary are failing, as it's not created
anymore for the host package.
Fixes: fe59b46 ("lua: include version number in installed files")
Signed-off-by: David Bauer <mail@david-bauer.net>
This is necessary to build PowerDNS authoritative and recursor against
OpenWRT, and may avoid packages depending on lua/host unnecessarily.
Signed-off-by: James Taylor <james@jtaylor.id.au>
Lua's LNUM patch currently doesn't parse properly certain numbers as
it's visible from the following simple tests.
On x86_64 host (stock Lua 5.1.5, expected output):
$ /usr/bin/lua -e 'print(0x80000000); print(0x80000000000); print(0x100000000)'
2147483648
8796093022208
4294967296
On x86_64 host:
$ staging_dir/hostpkg/bin/lua -e 'print(0x80000000); print(0x80000000000); print(0x100000000)'
-2147483648
0
0
On x86_64 target:
$ lua -e 'print(0x80000000); print(0x80000000000); print(0x100000000)'
-2147483648
0
0
On ath79 target:
$ lua -e 'print(0x80000000); print(0x80000000000); print(0x100000000)'
-2147483648
8796093022208
4294967296
It's caused by two issues fixed in this patch, first issue is caused by
unhadled strtoul overflow and second one is caused by the cast of
unsigned to signed Lua integer when parsing from hex literal.
Run tested on:
* Zidoo Z9S with RTD1296 CPU (aarch64_cortex-a53)
* qemu/x86_64
* qemu/armvirt_64
* ath79
Signed-off-by: Liangbin Lian <jjm2473@gmail.com>
[commit subject/message touches, fixed From to match SOB, fixed another
unhandled case in luaO_str2i, host Lua, package bump]
Signed-off-by: Petr Štetiar <ynezz@true.cz>
Cleanup to prepare for changing STAGING_DIR_HOSTPKG. The actual change of
STAGING_DIR_HOSTPKG (i.e., moving the host packages back into a common, not
target-specific directory) will be done after the first LEDE release, but
the cleanup will also be useful for projects like Gluon.
Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net>
