18:00:03 #startmeeting FESCo (2014-11-12) 18:00:03 Meeting started Wed Nov 12 18:00:03 2014 UTC. The chair is nirik. Information about MeetBot at http://wiki.debian.org/MeetBot. 18:00:03 Useful Commands: #action #agreed #halp #info #idea #link #topic. 18:00:03 #meetingname fesco 18:00:03 #chair dgilmore jwb kalev mattdm mitr mmaslano nirik sgallagh stickster t8m thozza 18:00:03 #topic init process 18:00:03 The meeting name has been set to 'fesco' 18:00:03 Current chairs: dgilmore jwb kalev mattdm mitr mmaslano nirik sgallagh stickster t8m thozza 18:00:10 hi 18:00:13 hi 18:00:14 Hello 18:00:25 hi 18:00:38 morning everyone. 18:01:01 hello! 18:02:15 ok, I guess lets dive in. A pretty short agenda today... 18:02:24 #topic #1365 A unique system-wide TMP directory for all programs and sane ways to retrieve the default 18:02:24 .fesco 1365 18:02:24 https://fedorahosted.org/fesco/ticket/1365 18:02:26 nirik: #1365 (A unique system-wide TMP directory for all programs and sane ways to retrieve the default) – FESCo - https://fedorahosted.org/fesco/ticket/1365 18:02:34 this sort of stalled out after last week... 18:03:04 firefox folks removed their hack... 18:03:31 so they use /tmp again in rawhide. 18:04:14 that sounds like a regression 18:04:35 regression? 18:04:46 downloading multi gigabyte files into /tmp would probably easily fill it up 18:04:48 well, that's going to fill up with big downloads 18:04:59 well, only if people don't say where to download them too... 18:05:05 hola 18:05:07 hi, sorry for being late 18:05:15 it only uses that until the save dialog not answered. 18:05:46 so I guess somone could save an iso and not answer the dialog and fill up space. 18:05:47 While I have questions around that, I’m not sure we should be micromanaging individual component’s behavior 18:05:49 my understanding is that they should continue using /var/tmp for that purpose 18:06:12 If we ended up researching Firefox and proposing Firefox changes today, that would get us no closer to resolving 1365 18:06:18 https://bugzilla.redhat.com/show_bug.cgi?id=1161110#c4 18:06:30 * nirik is with mitr 18:06:31 mitr, +1 18:06:38 which they did previously, but the problem was the way they achieved it, by overriding the TMPDIR variable which incidentally then also override it in the env passed to other programs 18:06:48 so do we have some distro wide policy or proposal here? 18:07:04 Proposal: design a sane system-wide API with a single implementation. 18:07:09 There, that’s going to help, isn’t it? 18:07:09 kalev, they want to stop that and use /tmp from now 18:07:12 * mitr shuts up 18:07:23 mitr: sure... but then the questions come. ;) 18:07:26 mitr, heh 18:07:40 mitr, is there actually a way tho get a sane API? 18:08:05 t8m: All it would take is FESCo saying Thou Shalt Use $library and “I don’t like it” doesn’ 18:08:09 t count as a reason not to. 18:08:17 Are we willing to do it and would we want to? 18:08:18 wait, what? 18:08:35 we have to write a library to have a system-wide tmpdir location? 18:08:38 I really do not think we can mandate anything that will fix this 18:08:41 mitr: well you could avoid that by adding it to libc 18:08:52 t8m: Technically this is easy enough, but it would be a very noticeable change to FESCo’s typical role. 18:09:03 drago01, no you can't. 18:09:07 it would need to be something worked on by all upstreams and all other distros also 18:09:19 mitr, I am afraid that this change of FESCo's role is not something that would pass 18:09:22 jwb: so people complain about having to use libc? 18:09:25 drago01: Having it in a library doesn’t yet imply usage. 18:09:35 what mitr said 18:10:14 mitr: surely that's mkstemp()/mkostemp() and we should mandate using it anyway for the obvious reasons? 18:10:45 mitr: oh well misread your comment then ... I though you meant people would say I don't want to use $lib ... which wouldn't happen if $lib is libc but anyway ot 18:10:57 jwb: Well, it’s either define a hard-coded convention for a path to use (in some way that promises the /tmp-on-tmpfs app-breaking redefinition will not reoccur), or define a common configuration mechanism to use to determine the path, something different than env variables because they don’t work so well. 18:11:05 I think the ticket reporter is looking for something like a "you should use /tmp by default for tmp files and you should honor $TMPDIR to override that" ? 18:11:12 or do nothing 18:11:17 which is what i think we should do here 18:11:31 Note that #1365 has an _underlying assumption_ that the path should be configurable, not hard-coded; I don’t think we have really discussed this. And there is yet one more underlying assumption that claws-mail is using $TMPDIR for its intended purpose, which I’ 18:11:35 m not entirely sure about. 18:12:04 pjones: AFAICS mkstemp() doesn’ 18:12:18 nirik, unfortunately that is broken by the tmp on tmpfs 18:12:19 t define the directory, and random names wouldn’t be usable for the claws-mail usage of IPC 18:12:31 t8m, no it isn't 18:12:36 yes, it is 18:12:38 t8m: well, 'broken' isn't what I would say... 18:12:45 really it means it's more nuanced... 18:12:47 mitr: if its use is ipc it should use /run 18:12:48 mitr: okay but they can be fixed at one place for the former case. for the latter, .... 18:12:58 no, it isn't. you may have less space, but you still are bounded by however much space is available to /tmp 18:13:00 tmp on tmpfs creates artificial limits on /tmp usage 18:13:11 that's not "broken" 18:13:16 that is broken 18:13:25 Let’s forget about /tmp as a _solution_ and instead talk about _requirements_? 18:13:38 t8m, we've now reached a pointless point in our conversation so we'll just stop. 18:13:42 yes 18:13:49 "you should use /tmp for tmpfiles you don't care about keeping accross reboots that are smaller than memory + swap, you should use /var/tmp for tmp files you don't care about keeping accorss reboots that are smaller than your /var/tmp, you should use $HOME/.local/tmp for files you want to keep accross reboots that are smaller than your homedir...etc. etc. 18:14:21 ie, there is no answer to "what should tmp files always use all the time" 18:14:38 nirik: This is actually a good start, separating the use cases. 18:14:44 Proposal: Close the issue as Reject 18:14:58 But then, how will a poor newbie programmer (coming from Windows or Android) ever find the convention; hence my talk about an API. 18:15:00 right, but then everything gets more complex. ;) and there's likely more use cases. 18:15:17 mitr, a poor newbie programmer is not going to be reading Fedora conventions to figure that out. 18:15:35 Why do we want a bazillion of tmp directories for god's sake 18:15:41 jwb: No, but they _may_ be using devhelp for the library they are already using to open files. 18:15:47 t8m: yes, let’; the IPC use case is clearly not the primary use of $TMPDIR, so 18:16:12 i'm still of the opinion that we should do nothing here 18:16:18 not sure we need to pass a guideline covering the general case 18:16:21 I think we can at least agree on that redefining environment variable is not a good approach, right? 18:16:25 There should be some directory for IPC and some directory for rest of the TMP usages 18:16:25 t8m: +1. claws-mail can store the fixed-path IPC-related files in a hard-coded path (whether /tmp or other is immaterial). 18:16:28 well even if fesco decides on something it wont make all developers out there go change their apps 18:16:33 so the point is... ? 18:16:45 thozza: Yet that is the existing mechanism. 18:16:52 right, it would need someone driving it to check and fix apps. 18:17:02 and upstream the changes. 18:17:12 drago01, exactly 18:17:15 jwb: We _have_ a deficiency in the platform, but, realistically, yeah, we are not fixing it here. 18:17:35 mitr: ok, so we have to come up with something else as an alternative.... I get it now 18:17:54 can we just say something like "FESCo asks Firefox to avoid overriding TMPDIR and to leave it as user configuration" and just close the issue? 18:18:03 and that's already implemented too 18:18:08 kalev: Well that is not _quite_ the question asked. 18:18:29 kalev, as it's already implemented it would be superficial to ask for that 18:18:47 even if someone asks us to solve the world hunger, that doesn't mean we can do that 18:18:59 mitr: the answer to the question asked needs to go somewhere higher than FESCo 18:19:08 dgilmore, uh, what? 18:19:09 we can possibly pass a general guideline that says where which kind of temporary files could be saved 18:19:10 no 18:19:12 come on 18:19:14 Yes, as we will not come to solution for a single TMP directory we should reject the ticket 18:19:14 it needs to involve all upstreams and distros 18:19:15 Proposal: FESCo does not currently want to mandate $TMPDIR being reliably consistent throughout an user session. If claws-mail requires a consistent place to use, it should use a different mecahnism. 18:19:30 mitr, +1 18:19:34 mitr: +1 18:19:38 mitr, +1 18:19:43 mitr: +1 18:19:56 dgilmore: No, this _is_ something we could actually solve within Fedora. However, on the list of priorities I think it is too low (without an interested leader at least). 18:20:04 mitr: +1 18:20:18 mitr: +1 18:20:53 sorry, someone at the door here... 18:20:56 mitr: perhaps. it would take someone interested and commited regardless, and I agree that person does not seem to exist 18:21:26 #agreed FESCo does not currently want to mandate $TMPDIR being reliably consistent throughout an user session. If claws-mail requires a consistent place to use, it should use a different mecahnism. (+6,0,0) 18:21:47 #topic #1359 Automatic OpenH264 download by Firefox 18:21:48 .fesco 1359 18:21:48 https://fedorahosted.org/fesco/ticket/1359 18:21:50 nirik: #1359 (Automatic OpenH264 download by Firefox) – FESCo - https://fedorahosted.org/fesco/ticket/1359 18:21:58 * nirik added his proposal in the ticket. 18:23:16 comments? discussion? or should we just vote? :) 18:23:42 i'm concerned the "accepted upstream" part won't happen 18:23:45 Hmmm this is even better example of FESCo's inability to mandate anything :( 18:23:53 jwb, +1 18:23:54 nirik: I think that (disabling auto-download and asking users to do manual work) would be reasonable for WebRTC only; but this will return the moment Firefox will start using OpenH264 for