irc.log

irc.log (text/plain), 10.29 KB, created by Paul Bredbury on 2006-08-06 12:45:57 UTC

Aug 05 23:40:18 brebs	Can a Portage dev please give a second opinion on bug #139842 (and then whack SpanKY over the head with a dictionary opened at the page defining a "bug")...
Aug 05 23:41:06 antarus	bug 139842
Aug 05 23:41:07 jeeves	antarus: http://bugs.gentoo.org/139842 nor, P2, All, [e-mail]->[e-mail], RESOLVED, WONTFIX, has_version and built_with_use ignore package.provided
Aug 05 23:41:16 *	antarus giggles
Aug 05 23:41:18 *	antarus looks
Aug 05 23:42:26 nichoj	how can you check the use flags of a package that isn't installed...?
Aug 05 23:42:32 antarus	you can't
Aug 05 23:42:37 antarus	so I see what his point i there
Aug 05 23:42:43 antarus	although he made it badly
Aug 05 23:42:53 antarus	open the second bug about has_version if you haven't alreadyt
Aug 05 23:42:55 antarus	since that is a bug
Aug 05 23:43:11 Caster	SpanKY's point is probably "it must be installed because obviously you depend on it"
Aug 05 23:43:20 SpanKY	cookie for Caster
Aug 05 23:43:21 antarus	built_with_use <pkg> <flag> will rightly fail if pkg is injected or provided
Aug 05 23:43:27 spb	"if the package is not installed, the return code is undefined"
Aug 05 23:43:32 spb	sounds reasonable to me
Aug 05 23:43:46 SpanKY	antarus: it returns true actually
Aug 05 23:43:56 antarus	SpanKY: either way :P
Aug 05 23:43:57 SpanKY	antarus: the point being that the value is meaningless when the package is not installed
Aug 05 23:44:11 antarus	perhaps fix it to return something specific
Aug 05 23:44:18 SpanKY	heh
Aug 05 23:44:25 SpanKY	i want to know what world you live in
Aug 05 23:44:27 Caster	SpanKY: unless you don't depend on it but if it was installed and had that useflag, it would be bad for you.... but is there such case? :)
Aug 05 23:44:30 SpanKY	where you have tristate boolean logic
Aug 05 23:44:35 SpanKY	{true,false,antarus}
Aug 05 23:44:46 antarus	SpanKY: you don't real dailywtf do you  :P
Aug 05 23:44:47 brebs	built_with_use should return a sensible answer. That's what my patch does. It's crazy to reject it.
Aug 05 23:44:48 antarus	heh
Aug 05 23:44:59 nichoj	truth is only as you define it
Aug 05 23:45:03 SpanKY	brebs: you dont get it, there is no sensible answer
Aug 05 23:45:10 antarus	SpanKY: the correct tri-state is {true,false,filenotfound}
Aug 05 23:45:17 SpanKY	{true,false,die}
Aug 05 23:45:22 SpanKY	i'm not about to add 'die' to the function
Aug 05 23:45:24 brebs	Yes there is. It's boolean, not tri-state.
Aug 05 23:45:27 antarus	{true,false,yourmominacan}
Aug 05 23:45:33 antarus	anyway
Aug 05 23:45:53 brebs	The logic goes as described in the bug. If the package isn't installed, then it can't possibly be "built_with_use" anything.
Aug 05 23:46:23 antarus	brebs: see but it could be
Aug 05 23:46:23 nichoj	well, unless you had way of 'providing' it as being built with some use flags...
Aug 05 23:46:25 antarus	we jsut can't tell ;)
Aug 05 23:46:28 SpanKY	read what Caster wrote brebs
Aug 05 23:46:44 SpanKY	he explained where as i said "youresodumb"
Aug 05 23:46:49 antarus	nichoj: that would be a good bug :P
Aug 05 23:47:31 Caster	die would at least put some QA check for missing dep
Aug 05 23:48:10 *	antarus blinks
Aug 05 23:48:14 antarus	there is no missing dep?
Aug 05 23:48:15 brebs	SpanKY: Eh? "it must be installed because obviously you depend on it" - WRONG.
Aug 05 23:49:02 Caster	it would be best if you provide a real-world example proving that wrong
Aug 05 23:49:04 SpanKY	great use of capslock tard
Aug 05 23:49:22 brebs	Caster: I have:  games-fps/quake2-data
Aug 05 23:49:40 antarus	SpanKY: I alwasys wondered why the KY in your name was capitaliz..oh wait that makes perfect sense
Aug 05 23:50:04 nichoj	gross
Aug 05 23:50:11 antarus	nichoj:  :)
Aug 05 23:50:18 Caster	brebs: elaborate? :)
Aug 05 23:51:32 brebs	Caster: In bug 139693 the solution would be to call has_version first, to protect against built_with_use returning a completely wrong answer.
Aug 05 23:51:32 jeeves	brebs: http://bugs.gentoo.org/139693 nor, P2, All, [e-mail]->[e-mail], NEW, pending, quake2-data does not merge (dies because of quake2-demodata)
Aug 05 23:51:56 SpanKY	it's a wrong answer regardless
Aug 05 23:52:04 Caster	ok I looked in the ebuild and see your point
Aug 05 23:52:08 SpanKY	true or false, it's wrong
Aug 05 23:52:17 SpanKY	i'll fix the ebuild to has_version first
Aug 05 23:52:35 antarus	the bug would be in the provided stuff, that you need to provide with a set of use flags
Aug 05 23:52:37 Caster	so it really is the "you don't depend on it but if it was installed and had that useflag, it would be bad for you" case
Aug 05 23:52:39 antarus	and then patch build_with_use
Aug 05 23:52:48 antarus	s/build/built
Aug 05 23:52:55 brebs	The point here is simply:  built_with_use and has_version ignore package.provided, and I've provided patches to fix the broken commands.
Aug 05 23:52:57 antarus	Caster: or good, depending on how built_with_use was being used
Aug 05 23:52:59 SpanKY	if the package is provided, then you have to assume the user didnt fuck it up
Aug 05 23:53:09 SpanKY	otherwise it'd be impossible for user to install
Aug 05 23:53:14 SpanKY	they'd have to edit the ebuild
Aug 05 23:54:11 brebs	SpanKY: You are missing the point that those 2 ebuilds commands are BROKEN. Jeez.
Aug 05 23:54:17 *	antarus sighs
Aug 05 23:54:22 antarus	brebs: chill out
Aug 05 23:54:49 antarus	built_with_use is not broken, as we've stated like a dozen times, because there is no data to know either way
Aug 05 23:55:04 antarus	and there is nowhere to look up the data
Aug 05 23:55:11 brebs	There is. That's how my patch fixes it.
Aug 05 23:55:20 SpanKY	you read the user's mind ?
Aug 05 23:55:21 SpanKY	awesome
Aug 05 23:55:39 brebs	In the circumstance described in the bug, I mean.
Aug 05 23:55:40 *	antarus sighs
Aug 05 23:56:01 SpanKY	well while you focus on the small picture, we'll fix the bigger picture
Aug 05 23:56:19 brebs	SpanKY: You're talking about fixing 1 ebuild, I'm talking about fixing portage.
Aug 05 23:57:22 *	antarus sighs and returns to being pissed at git
Aug 05 23:58:38 brebs	If the two commands weren't broken, that what's the point in my 2 patches to fix them? They alter portage's behaviour so that the ebuild (which contains correct logic) actually works.
Aug 06 00:04:10 SpanKY	and we're telling you that your patch doesnt fix anything
Aug 06 00:04:27 SpanKY	it merely changes the behavior so it works in this one case
Aug 06 00:05:06 brebs	SpanKY: Yes it does. I tested it. It fixes the broken logic in the circumstance described on the bug. That's a bug, and I submitted patches to fix it.
Aug 06 00:05:09 SpanKY	the has_version change you posted is correct, the built_with_use is wrong
Aug 06 00:05:19 jakub	IMO there's no correct logic in pretending that something's been installed with use flags you can't verify, as the whole point of those checks is that something really does need the flags to work
Aug 06 00:05:55 SpanKY	brebs: you dont get it, it's broken regardless of returning true or false
Aug 06 00:05:58 SpanKY	it's always going to be wrong
Aug 06 00:06:56 brebs	SpanKY: Not in the circumstance described on the bug. If a package is not installed, then it cannot be built_with_use *anything*. This is Boolean logic. Portage is wrong, without my patch.
Aug 06 00:07:28 SpanKY	that's the entire point
Aug 06 00:07:34 SpanKY	"not in the circumstance described"
Aug 06 00:07:53 SpanKY	i can come up with a circumstance going the opposite direction where yours breaks but mine works
Aug 06 00:08:04 brebs	Do tell.
Aug 06 00:09:56 SpanKY	your example works just as well
Aug 06 00:10:15 SpanKY	if you package.provide the demo ebuild, the symlink check will return true
Aug 06 00:10:42 brebs	Correct. So it *should*.
Aug 06 00:10:58 SpanKY	and so it does
Aug 06 00:11:19 brebs	So where is my patch broken?
Aug 06 00:12:11 SpanKY	i'm saying it's flawed fundamentally, no amount of changing the code will fix that
Aug 06 00:13:06 brebs	Not true. It's a boolean result, so it should return the correct answer where it can. And it can in the situation described on the bug, which the patches fix.
Aug 06 00:15:00 SpanKY	exactly, you're asking about the state of the system in a corner case that has more than boolean logic ... you're truncating it down to two values
Aug 06 00:16:56 brebs	When a package isn't installed, it cannot be built_with_use anything. Because it's not installed. So the answer is quite clearly False. It's not unknown, or maybe, or Schroedinger's cat. It's False.
Aug 06 00:19:09 Caster	I'm getting lost, what it does now? if it's not installed/provide/injected it returns false? if it is provided it returns true?
Aug 06 00:19:23 antarus	Caster: I believe it returns true
Aug 06 00:19:41 Caster	hm
Aug 06 00:19:45 antarus	which IMHO is correct behavior, otherwise building a system with provided packages would be impossible
Aug 06 00:20:02 antarus	as SpanKY noted, you need to assume the user providing the package knows what they are doing
Aug 06 00:20:13 Caster	well brebs you should instead fill enhancement bug about specifying USE in package.provided
Aug 06 00:20:20 antarus	and in most of the corner cases the package fails to build properly if the build_with_use is wrong
Aug 06 00:20:27 antarus	so they will find out anyway
Aug 06 00:20:39 nichoj	so is ${FEATURES} considered internal state, and shouldn't be poked from eclasses / ebuilds?
Aug 06 00:21:13 antarus	nichoj: you can't do any kind of DEPEND stuff based on features
Aug 06 00:21:46 nichoj	antarus: not what I was getting at
Aug 06 00:21:56 nichoj	ie, for checking for FEATURES=strict
Aug 06 00:22:30 antarus	generally no
Aug 06 00:22:41 antarus	what kind of use case are you shooting for
Aug 06 00:23:03 nichoj	antarus: mm, well I'm already doing it in java-utils-2.eclass... :)
Aug 06 00:23:20 nichoj	antarus: like is-java-strict()
Aug 06 00:23:25 nichoj	to do qa type stuffs
Aug 06 00:23:44 *	antarus shivers
Aug 06 00:23:50 nichoj	what?
Aug 06 00:23:52 antarus	I shall stab you now :P
Aug 06 00:24:00 brebs	Caster: I provided 2 patches, to fix both has_version and built_with_use so that they don't blindly ignore package.provided. They shouldn't have undesirable side-effects.
Aug 06 00:24:04 *	nichoj gets his knife proof vest
Aug 06 00:24:43 antarus	brebs: I suggest you make your case on gentoo-portage-dev and not here
Aug 06 00:25:36 brebs	Oh, wrong channel? OK.
Aug 06 00:25:39 antarus	wtf is java-strict
Aug 06 00:25:43 antarus	brebs: no, the mailing list..
Aug 06 00:25:52 *	antarus stabs nichoj 
Aug 06 00:25:56 brebs	Ah.