If set, binaries will have rpath set to $ORIGIN/../lib. That means
that even using defaults (--prefix=/usr), you can install
it via DESTDIR anywhere in your system and binaries will work.
This is in preparation for BSD support.
Closes: #87 [via git-merge-pr]
Use number of processors as reported by /proc/cpuinfo,
otherwise just set it to 1 (default).
This makes running the test suite in less than 16s on my
i5 ivybridge CPU.
This now works as expected again, see https://i.imgur.com/pr2XO70.png
for the full dependency graph of the sway pkg installed on my system.
Closes: #80 [via git-merge-pr]
Splitting the socks5 request is causing issues when running against Tor.
It causes Tor to fail with an address unknown error. Assembling the
request and sending it in its entirety solves the issue.
An strace of torsocks, that functions correctly when used with xbps, shows
that the request is being sent all at once:
recvfrom(3, "\5\0", 2, 0, NULL, NULL)
sendto(3, "\5\1\0\3>lysator7eknrfl47rlyxvgeamrv7ucefgrrlhk7rouv3sna25asetwid.onion\0P", 69, 0, NULL, 0)
recvfrom(3, "\5\0\0\1\0\0\0\0\0\0", 10, 0, NULL, NULL)
Closes: #54 [via git-merge-pr]
The previous idea was to use virtual packages in the users configuration
to satisfy dependencies by mapping them to existing installed packages.
Using virtual packages for it doesn't work as expected and trying to make
it work would break other functionalities of virtual packages, like the
version satisfaction checks for `provides` and the ability to replace
virtual packages with real packages. The virtual package functionality
should be used exclusively for virtual packages.
This allows users to specify packages packages that should be ignored.
Ignored packages in dependencies are always satisfied without installing
the package, while updating or installing a package that depends on an
ignored package.
This does NOT ignore the shlib checks, ignoring a package that provides
required shared libraries will abort the transaction as if there was no
package that provides the required shared library.
Connect to the addresses from `getaddrinfo(3)`,
alternating between address family,
starting with ipv6 and wait `fetchConnDelay`
between each connection attempt.
If a connection is established within the attempts,
use this connection and close all others.
If `connect(3)` returns `ENETUNREACH`, don't attempt more
connections with the failing address family.
If there are no more addresses to attempt,
wait for `fetchConnTimeout` and return the first established
connection.
If no connection was established within the timeouts,
close all sockets and return -1 and set errno to
`ETIMEDOUT`.
faster: use a hash table with pkg names on the transaction dict,
the process of collecting and sorting is now 50x faster or
even more (kde5).
bugs: this now detects cyclic deps and returns with an appropropiate
return value: ELOOP and ENOENT in xbps-query(1) --fulldeptree.
Ping me if you need more details :-)
Close https://github.com/void-linux/xbps/issues/16
Close https://github.com/void-linux/xbps/issues/5
The order of entries in files.plist is inconsistent
between different releases of xbps. Hence they must be
ordered before removal to, so sub directories are removed
before their parents.
A segfault in xbps-create(1) was found in:
https://github.com/void-linux/void-packages/pull/7524
xbps-create adds a file type string to the xentry for every file it
finds in `destdir`, files other than symlinks, regular files and
directories weren't handled and resulted in a segfault later when
when processing the xentry structures.
This commit adds checks for sockets and fifos and exits with an
appropriate error message or with a generic error message for every
other unhandled file.
Closes#66
