summaryrefslogtreecommitdiffstats
path: root/TODO
diff options
context:
space:
mode:
authorHolger Levsen <holger@layer-acht.org>2015-01-23 12:50:30 +0100
committerHolger Levsen <holger@layer-acht.org>2015-01-23 12:50:30 +0100
commit1f3cacc6bf14171909cf5c99fb9376722237c032 (patch)
tree2038af13a0cf4d977dc63589ad43a0ffbd489c9f /TODO
parent0f7d8e7e0b1b560e82abb739e8af1f8f56e75368 (diff)
downloadjenkins.debian.net-1f3cacc6bf14171909cf5c99fb9376722237c032.tar.xz
reproducible: add notes how to test variations in kernel version, date and cpu type
Diffstat (limited to 'TODO')
-rw-r--r--TODO11
1 files changed, 11 insertions, 0 deletions
diff --git a/TODO b/TODO
index a8f7a8ce..2830094e 100644
--- a/TODO
+++ b/TODO
@@ -141,6 +141,17 @@ properties:
* reproducible: dont delete userContent/$pkg on build begin..
* missing tests: variation in kernel and date
+** http://www.bstern.org/libuname/ - add "-ldl" to the linker flags...
+** lunar suggests to use qemu with -r and -cpu and -rtc flags
+** https://wiki.debian.org/qemubuilder - part of cowdancer
+ < Lunar^> I would rather say we do not do full uname and CPU variation that use an unreliable LD_PRELOAD
+ < Lunar^> because as soon as the package unset or reset the variable it won't be there anymore
+ * | h01ger nods. its also ok to leave some variations for when we go "for real" (aka rebuilds in the field)
+ <h01ger> | Lunar^: if that happens. if that doesnt we still get some more variations
+ <h01ger> | so i'd say such an LD_PRELOAD would be useful, even though it wouldnt be complete
+ <h01ger> | but we will find more problems "before going into the fields"
+ * | h01ger guesstimates that we would go from 300-600 builds per day to 100-200 with qemubuilder
+ <h01ger> | but those qemubuilders would only get 4 or 8gb ram... maybe 6. and that will have a huge impact
* cleanup+rename:
** unschedule_from_db( (defined in build.sh...)