/srv/irclogs.ubuntu.com/2009/01/26/#upstart.txt

Keybuksadmac, keesj: http://bazaar.launchpad.net/%7Escott/libnih/trunk/annotate/635?file_id=NEWS-20060424160942-032abeceeca6157f16:11
Keybukthis should hopefully alleviate some of the issues with libnih's unstable API16:11
Keybuk(ie. documenting the changes as we go)16:12
keesjcool. 16:12
sadmac2Keybuk: cool16:16
sadmac2Keybuk: I have the beginnings of an async api at home. It needs a couple tweeks and a lot more testing but its there.16:17
Keybuksweet16:17
KeybukI'm going to push a 0.5.1 once I've got upstart up to date with the libnih changes16:18
Keybukbecause there were some realloc oom failure issues I found in the process of changing the allocator16:18
Keybukthen 0.5.2 we'll either push in the async api or the properties support (whichever comes first), and 0.5.3 will get whichever comes second16:18
sadmac2sounds good16:20
sadmac2I'll re-hash initctl as soon as the properties support comes in16:21
Keybukthe plan being to start making some very rapid releases16:21
Keybuksince we have a test suite, in theory, upstart is always releasable as long as it passes16:21
Keybukin hindsight, the "drop all the IPC, rewrite the core, add D-Bus back" plan wasn't ideal16:21
sadmac2Keybuk: have you looked at the manpages in 0.5.0? I get this funny feeling nobody updated them :)16:21
Keybukno :)16:21
KeybukI actually did start on some new manpages about a year ago16:22
Keybukwhich actually tried to document what happened events-wise16:22
KeybukI should finish them up, then we have a 0.5.4 ;)16:22
sadmac2Keybuk: in 0.10, when someone installs a .deb/.rpm with a service, can that service be disabled by default?16:23
Keybukshould be possible, yes16:23
Keybukthe timing of that is kinda critical though16:23
Keybuksince if rpm unpacks the file, upstart will notice ;) - the disabled flag needs to appear before the definition16:24
sadmac2Keybuk: and if you disable a service, does it disable all its dependents?16:24
Keybukright16:24
Keybuksince none of the dependants can be automatically started16:24
sadmac2Keybuk: consider ypbind and gdm: if yp is installed, configured, /and/ enabled, gdm should depend on it. if it is not all of those, gdm should not depend on it.16:25
sadmac2just thought of this while setting up the auth for my new workstation :D16:25
Keybukthat's done by putting the gdm dep into the ypbind file16:25
Keybuke.g. today start on starting gdm; stop on stopped gdm16:25
sadmac2Keybuk: ah right16:26
sadmac2Keybuk: so in ypbind: before gdm16:26
Keybukexactly16:26
* sadmac2 doesn't like that word "before"16:26
sadmac2it has very task-y semantics, which is misleading16:26
Keybukbefore implies there's an after, doesn't it16:26
Keybukor implies ordering16:26
sadmac2yeah16:26
* Keybuk has to run for a bit16:26
nottingwell, ideally it's an asbtract nss event that hides away the ypbind/ldap/krb/whatever, but... details details16:26
KeybukFedora Rawhide broked my disk, and I need to buy a new one16:27
Keybukplus I need some food ;)16:27
sadmac2Keybuk: we're hardcore like that. not everyone can handle the intensity of Rawhide16:27
sadmac2that's why it says "Raw"16:27
* keesj will start looking into improving his state machine for upstart this week17:01
nottingobviously we need runtime-pluggable state machines17:01
* keesj remebers somebody else also was busy with something similar17:01
keesjI guess I should post how we did it and get shot17:02
ion_Note to self: ask Keybuk why: * nih_config_parse() now reads the file into memory, instead of using mmap().20:38
sadmac2ion_: git has gotten itself into trouble with nmap before21:08
ion_How?21:09
sadmac2ion_: mmapping very very large things causes pain21:13
sadmac2thrashing and other ugliness21:14
ion_Yeah, but i’m sure nih_config_parse isn’t going to read huge files. :-)21:16
sadmac2ion_: one man's uninteresting use case is another man's security vulnerability21:17
sadmac2why give an application the ability to crash the system, no matter how well written?21:17
ion_In this case, the change seems to be to read the entire file to memory, so the case of a user passing a huge config file to the function would potentially cause even more RAM to be used than with mmap.21:18
sadmac2ion_: As 4chan so elloquently puts it: OH SHI-21:20

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!