Timeline

Many hyperlinks are disabled.
Use anonymous login to enable hyperlinks.

50 most recent check-ins related to "2.3.x"

2013-03-17
18:41
This will become 2.4 check-in: fc671b80db user: bdrewery tags: 2.4.x
2013-03-15
00:29
MFT: Fix detection of IPv4 support. net.inet is not a good check. kern.features.inet[6] is supported on all supported FreeBSD branches. Reported by: flo Closed-Leaf check-in: 6aa34cc5e3 user: bdrewery tags: 2.3.x
2013-03-13
10:18
Deprecate poudriere-cron check-in: 7fac17b152 user: bdrewery tags: 2.3.x
2013-03-12
22:03
Remove csup method for ports check-in: 61133bfd15 user: bapt tags: 2.3.x
15:10
MFT: Fix compatibilty with 8.3 make(1). It returns blank with the -V ${} syntax check-in: 0183726495 user: bdrewery tags: 2.3.x
11:16
MFT: testport -i: Packages are at /packages now check-in: d5e223f89b user: bdrewery tags: 2.3.x
03:05
MFT: testport: Respect NO_MTREE by not populating PREFIX. 'install-mtree' will do this correctly already. This at least fixes mail/qmail (PREFIX=/var/qmail) incorrectly show bin/ was removed. This also fixes testport to match bulk -t as it was not populating PREFIX itself. check-in: 482647249e user: bdrewery tags: 2.3.x
02:17
MFT: Fix leftovers having extra './' in removed dirs, causing the PREFIX to not be stripped out check-in: 6bb2091694 user: bdrewery tags: 2.3.x
00:52
MFT: testport -n: Respect the port's advertised PREFIX if not already set in env or poudriere.conf This fixes 2 issues. 1) bulk was already respecting the port's PREFIX. testport would report different failures than bulk -t would. 2) 75 ports define their own PREFIX. This may be worth changing in the ports. While they are doing this, we should test them with their advertised default PREFIX. This change does not impact the default of PREFIX=/prefix/ for testport check-in: fdf44de3db user: bdrewery tags: 2.3.x
00:07
Rename MYBASE -> LOCALBASE (matches trunk) check-in: ccb1314304 user: bdrewery tags: 2.3.x
2013-03-11
11:46
Disable SIGINFO during index building as it was adding junk into the INDEX Fixes: [43e10c89914] check-in: f97b8ec4f4 user: bdrewery tags: 2.3.x
2013-03-10
11:36
Add a comment into the generated make.conf show the source file check-in: d76aced7c4 user: bdrewery tags: 2.3.x
2013-03-08
20:29
MFT: Fix NO_RESTRICTED due to packages being mounted read-only in the jail check-in: 041efc595d user: bdrewery tags: 2.3.x
04:09
MFT: Give hint about ctrl+t check-in: 7106f85cf3 user: bdrewery tags: 2.3.x
2013-03-07
23:20
MFT No need to save_wrkdir() when failed to extract check-in: 7da2619e30 user: bdrewery tags: 2.3.x
2013-03-06
11:22
Show make.conf in buildlog check-in: 2c82751520 user: bdrewery tags: 2.3.x
03:58
MFT Fix build_queue() exiting early while builds are still occurring. Also fixes... 1. The last ports built would forever be in PORT:done status, instead of idle:. This was especially noticeable in the new HTML page. 2. The stats would not get updated due to the race condition, noticeable on the HTML page 3. The bulk would exit while children were still running, resulting in busy mounts and failures cleaning up 4. build_queue() was not closing the &6 socket when done since it would 'return 0' on empty queue, instead of break. This regressed due to multiple problems. Most of all, the splitting of the building pool from pool/PKG/.building into building/ causes the code to see the build pool is empty, so it immediately returns, regardless if any builders are still active. Now it will keep looping until all builders are inactive. check-in: f5ab1f380d user: bdrewery tags: 2.3.x
2013-03-03
15:31
Pass VERSION to subcommands check-in: 4b2b5a608b user: bdrewery tags: 2.3.x
14:11
Ignore realpath errors when cleaning up This was due to recursive cleanup already deleting the deps/ that were now expected in rdeps/ of a dependent pkg. This was a harmless warning check-in: 33a13b7941 user: bdrewery tags: 2.3.x
00:59
Show poudriere version in buildlog check-in: c72c403ba7 user: bdrewery tags: 2.3.x
2013-03-02
16:20
ports: Fix svn+https Submitted by: antoine@freebsd.org Ticket: [c63bc9066d2511d6] check-in: b3ca524cfd user: bdrewery tags: 2.3.x
12:07
MFT: priority pool. Based on patch from: Matthew Dillon <dillon@apollo.backplane.com> check-in: b00f66b60a user: bdrewery tags: 2.3.x
2013-03-01
14:15
Explicitly disable compression in builder zfs clones check-in: e3157b0093 user: bdrewery tags: 2.3.x
13:07
When cloning ZFS builder, disable atime/sync as they are not needed check-in: 51a9053ef2 user: bdrewery tags: 2.3.x
12:23
Speedup clean.sh more by calling mv(1) less and passing more files at once check-in: f2e6b4e0c2 user: bdrewery tags: 2.3.x
04:33
Fix always running mark_preinst. [ -n ] needs its param quoted or it is always true. check-in: 43a159c5e2 user: bdrewery tags: 2.3.x
04:23
bulk: Don't mark preinstall if not running with -t check-in: d4b7ec40c9 user: bdrewery tags: 2.3.x
01:32
Speedup cleaning by running rm(1) much less. This is especially noticeable when running 'clean_pool ports-mgmt/pkg' with -a check-in: 176ad60340 user: bdrewery tags: 2.3.x
2013-02-28
20:57
No longer need to remove from pool/ when cleaning since it is now moved to building/ check-in: 11612a66b9 user: bdrewery tags: 2.3.x
15:50
Remove the queue lock(1) This was originally done since the pool/ dir had a combination of 'ready to build', 'building' and 'needs deps'. Now that all are separated, the lockf is not needed. The pkgdirs are atomically moved between each and atomically cleaned up. next_in_queue() also only runs from 1 process always. The cleanup is still in its own clean.sh, but could now be moved into common.sh. This change is reported to fix the build getting stuck up to 60 seconds looking for another port to build while others are cleaning up. If there are X builders and X finish at the same time, they must serially cleanup, while the master also waits for the lock to be released. Now the master can pick a new pkg to build much quicker. check-in: 692b8241d0 user: bdrewery tags: 2.3.x
15:50
Instead of touching pool/<pkgname>/.building, move it to building/<pkgname> This avoids a race condition when cleaning from child and master is looking for work in pool/. The child cleanup removes the .building and then the pool/<pkgname> dir while the master then sees pool/<pkgname> is empty and ready to build, so it picks it to build again. By moving the pkgname to building/ instead, fully atomic operations can be done, avoiding the need for lockf (coming next) check-in: 142ce6e257 user: bdrewery tags: 2.3.x
2013-02-24
18:07
When showing crashed builders, trim away trailing / check-in: 467129f218 user: bdrewery tags: 2.3.x
2013-02-23
18:18
Fix modified file detection incorrectly picking up directories after 'type' was removed check-in: 1c419719f5 user: bdrewery tags: 2.3.x
18:17
Fix leftover detection The mod variable was being shadowed, which caused the file 'M' to be written to. Since the filesystem is now read-only, this caused the check to error out and not finish. check-in: 88ba4a3116 user: bdrewery tags: 2.3.x
17:50
Keep leftover tmpfiles in the jail (long lost jailbase var was not updated) check-in: baa22aefe4 user: bdrewery tags: 2.3.x
16:22
job_msg: Redirect to msg() if MY_JOBID is not set check-in: 5c5866a31b user: bdrewery tags: 2.3.x
2013-02-20
02:58
Run log_stop in exit_handler before killing children. Not doing so causes the children kill to immediately exit the parent process (even though it is NOT being killed) check-in: 080c9c8242 user: bdrewery tags: 2.3.x
2013-02-17
22:58
bulk: Detect cycles before starting build check-in: b4d7352fe6 user: bdrewery tags: 2.3.x
22:52
Remove whitespace check-in: 6914425552 user: bdrewery tags: 2.3.x
22:48
Give smarter error messages when hitting deadlocked queue Ticket: [f8d40363c4] check-in: 79ae4dd484 user: bdrewery tags: 2.3.x
18:57
bulk: Fix creating pkg repo.txz if the package dir is a symlink check-in: 71e3d4ca1d user: bdrewery tags: 2.3.x
2013-02-16
18:41
Check for deadlocked queue before reading queue as it may already be known to be dead, thus not needing the timeout. check-in: 6e308f8a47 user: bdrewery tags: 2.3.x
18:29
bulk: Fix silent ignoring of ports that depend on each other. This is fixed by checking both the ready-to-build pool (pool) and need-to-build pool (deps) to be empty before considering the bulk done. By doing this, if 2 ports depend on each other and are all that remain, this will trigger the 'unprocessable queue' assertion and break out. check-in: 46660377e0 user: bdrewery tags: 2.3.x
18:24
Add a timeout to the queue read(1) so queue deadlocks can be detected due to dependency issues or crashed builders that to not notify the master pipe check-in: 211cddecda user: bdrewery tags: 2.3.x
2013-02-15
23:57
testport: Fix -c to run before depends are checked/built/installed as it may change which depends are needed check-in: 4e168d4d8d user: bdrewery tags: 2.3.x
23:28
Merge sandboxed/atomic (per-port) packaging from trunk check-in: 1d5089129d user: bdrewery tags: 2.3.x
23:18
mnt is not set on 2.3.x, use JAILMNT Closed-Leaf check-in: 8e3d4c757d user: bdrewery tags: port-new_packages
23:13
Fix DISTFILES changes breaking out of jail due to 'mnt' not being set on 2.3.x check-in: 06dcdb06ce user: bdrewery tags: 2.3.x
20:16
Remove testport -d zsh completion check-in: 626d8728d8 user: bdrewery tags: 2.3.x
14:22
Mount packages as read only in jails, and only add new created packages in the package directory if they are ok check-in: 50ba04b15b user: bdrewery tags: port-new_packages