[arch-general] Discussion on usage of [testing] repo - minimal requirements?
https://wiki.archlinux.org/index.php/Testing#.5Btesting.5D states the following (paraphrased for summary):- [testing] is for experienced users who can deal with broken systems, not for the 'absolute latest' package versions. It may break critical/[core] packages, as such users are strongly encouraged to subscribe to [arch-dev-public] and watch the [testing] sub-forum. Second reference - this forum thread - https://bbs.archlinux.org/viewtopic.php?id=128682 A user basically is using [testing] without fulfilling the above requirements. Should the user be advised not to use [testing] or is this counter-productive to the purpose of [testing]? This discussion is not meant to be commentary on any of the personalities involved in the thread, the thread did spark this but I'd like to hear (esp from devs/TUs) opinions on the general case, not the specific case. The reason for this is that its possible there's quite a bit of difference in opinions, and some clarity would (IMO) help. My personal opinion - If you don't fulfil the above requirement you should not use [testing], and other Arch users should graciously inform you as such. Sorry for the long email, feel free to mute this thread, since I do know its got high potential for unproductive cross-chatter.
[2011-10-24 11:41:47 +0800] Oon-Ee Ng:
[testing] is for experienced users who can deal with broken systems, not for the 'absolute latest' package versions. It may break critical/[core] packages, as such users are strongly encouraged to subscribe to [arch-dev-public] and watch the [testing] sub-forum.
I believe we all agree on that.
https://bbs.archlinux.org/viewtopic.php?id=128682
A user basically is using [testing] without fulfilling the above requirements. Should the user be advised not to use [testing] or is this counter-productive to the purpose of [testing]?
People using [testing] should read arch-dev-public, just as all Arch users should read the news announcements. It is a waste of time for everyone when people ask questions that have already been answered there. -- Gaetan
On Mon, Oct 24, 2011 at 5:41 AM, Oon-Ee Ng <ngoonee.talk@gmail.com> wrote:
A user basically is using [testing] without fulfilling the above requirements. Should the user be advised not to use [testing] or is this counter-productive to the purpose of [testing]?
It is very important that people use testing [0], so we should really encourage _more_ rather than less of that, IMHO. Rather than advice the user to stop using testing, I'd advice them to sign up to the mailinglist :-) Maybe this requirement should be communicated more clearly (e.g. a comment in the standard pacman.conf)? Cheers, Tom [0]: a lot of the non-trivial bugs against my packages are discovered after they move to core, probably because the people with the right hardware/use-cases are not using testing.
On Mon, Oct 24, 2011 at 11:24 AM, Tom Gundersen <teg@jklm.no> wrote: <snip>
Maybe this requirement should be communicated more clearly (e.g. a comment in the standard pacman.conf)?
Great idea. I mean, as a non-[testing] user I get that guinea pig feeling which comes naturally with linux often enough. Don't miss to express that [testing] here is far from what other distros label with "testing" and will hopefully break your system (because we want to know). <snip>
[0]: a lot of the non-trivial bugs against my packages are discovered after they move to core, probably because the people with the right hardware/use-cases are not using testing.
Looks really like I keep missing all the fun. What the hell, I'll add [testing] to my pacman.conf tonight... cheers! mar77i
On Mon, Oct 24, 2011 at 1:40 PM, Martti Kühne <mysatyre@gmail.com> wrote:
On Mon, Oct 24, 2011 at 11:24 AM, Tom Gundersen <teg@jklm.no> wrote: <snip>
Maybe this requirement should be communicated more clearly (e.g. a comment in the standard pacman.conf)?
Great idea. I mean, as a non-[testing] user I get that guinea pig feeling which comes naturally with linux often enough. Don't miss to express that [testing] here is far from what other distros label with "testing" and will hopefully break your system (because we want to know).
Depends on who you compare to, unlike certain other distro's who shall not be named, we actually compile, install and test our packages before pushing to testing. We really don't want any packages in testing to break anyone's system as that will lead to fewer people using it. However, there will obviously be problems from time to time. Personally, I use testing on all my five machines (including for work), and never experienced a big problem (such as loss of data or a failed boot), but your mileage may vary ;-) Cheers, Tom
On Mon, Oct 24, 2011 at 06:53, Tom Gundersen <teg@jklm.no> wrote:
On Mon, Oct 24, 2011 at 1:40 PM, Martti Kühne <mysatyre@gmail.com> wrote:
On Mon, Oct 24, 2011 at 11:24 AM, Tom Gundersen <teg@jklm.no> wrote: <snip>
Maybe this requirement should be communicated more clearly (e.g. a comment in the standard pacman.conf)?
Great idea. I mean, as a non-[testing] user I get that guinea pig feeling which comes naturally with linux often enough. Don't miss to express that [testing] here is far from what other distros label with "testing" and will hopefully break your system (because we want to know).
Depends on who you compare to, unlike certain other distro's who shall not be named, we actually compile, install and test our packages before pushing to testing. We really don't want any packages in testing to break anyone's system as that will lead to fewer people using it. However, there will obviously be problems from time to time.
Personally, I use testing on all my five machines (including for work), and never experienced a big problem (such as loss of data or a failed boot), but your mileage may vary ;-)
Cheers,
Tom
IMHO, one that doesn't count for much, I have to agree with Tom. I also have to agree with those making the point for watching the Arch Dev Public mailing list and reading the news announcements. I moved to Arch because it forces me to learn how to maintain my machines. It also allows me to compile my base and core packages to my machines architecture not a generic configuration. This is my system: Linux gandalf 3.0-pf #1 SMP PREEMPT Mon Oct 24 00:05:45 CDT 2011 x86_64 AMD Phenom(tm) 8450 Triple-Core Processor AuthenticAMD GNU/Linux This is my makepkg configuration: CFLAGS="-march=amdfam10 -m64 -O2 -pipe -fstack-protector --param=ssp-buffer-size=4 -D_FORTIFY_SOURCE=2" CXXFLAGS="-march=amdfam10 -m64 -O2 -pipe -fstack-protector --param=ssp-buffer-size=4 -D_FORTIFY_SOURCE=2" This may not be an extreme, or even close to an edge case, but it might find something that a generic compile doesn't find. I've also discovered whether it boots or not, I can fix it. I will admit, as someone who has already responded on this thread can attest, I can ocassionally as a "I really should have known that" type of question. However, everyone screws up every once in a while, except me I'm perfect. The rest of this may be considered noise/off topic/thread high jacking but I'll try to make a point. Until I became disabled and had to retire in 2009 I was considered one of the best at what I did. I routinely trained people and wrote training manuals. It tooks years of having someone point out to me that I had the same response when training people that some experienced linux users have. If I had to tell someone more than once how to do it they got dressed down, after the third time I had no use for them. That was a hard lesson to learn. My son was an expert, definition of an expert -- a has been little drip, with Windows and worked as a support tech. He had the same opionion when training people, after the third how to do the same thing he had no use for them. The same son who was quick to point out how badly I treated people when trying to train them. I know answering the same questions over and over can be a pain and no one wants to invite "help vampires" but simply saying "don't use testing" just doesn't seem to be the right way to go. I've read the thread linked in the first email and I agree with the point made if, and I point out if and only if, it's done graciously. To many people, no names etc just generic people, jump in when the original poster doesn't get it and start slicing and dicing. Keep the commentary civil. Myra -- Life's fun when your sick and psychotic!
On (10/24/11 09:37), Myra Nelson wrote: -~> On Mon, Oct 24, 2011 at 06:53, Tom Gundersen <teg@jklm.no> wrote: -~> > On Mon, Oct 24, 2011 at 1:40 PM, Martti Kühne <mysatyre@gmail.com> wrote: -~> >> On Mon, Oct 24, 2011 at 11:24 AM, Tom Gundersen <teg@jklm.no> wrote: -~> >> <snip> -~> >>> Maybe this requirement should be communicated more clearly (e.g. a -~> >>> comment in the standard pacman.conf)? -~> >> -~> <snip> -~> doesn't get it and start slicing and dicing. Keep the commentary -~> civil. -~> -~> Myra -~> -~> -~> -- -~> Life's fun when your sick and psychotic! It is simple: if you don't use testing you never learn. Telling others "RTFM and don't ask questions" is ridiculous, because following this logic >50% of forum posts is just noise. And just because you subscribe to ML doesn't mean that you'll remember 1 relevant message out of 100 (personally I learned more from http://allanmcrae.com/2011/08/pacman-package-signing-3-pacman/ about pacman package signing than from all of [{arch,pacman}-dev*]. This is of course not to say that ML are not important, beacuse they are. Besides, one really doesn't have to enable testing in pacman.conf -- individual pacman -U will do, imho. Regarding your compile flags, I would use -match=native (instead of your -march and -m) and -fstack-protector-all (instead of -fstack-protector) if you don't mind increasing the size of binaries a little. -- Leonid Isaev GnuPG key ID: 164B5A6D Key fingerprint: C0DF 20D0 C075 C3F1 E1BE 775A A7AE F6CB 164B 5A6D
On Mon, Oct 24, 2011 at 5:52 PM, Leonid Isaev <lisaev@umail.iu.edu> wrote:
Besides, one really doesn't have to enable testing in pacman.conf -- individual pacman -U will do, imho.
I've read that [testing] is all or nothing and you shouldn't cherrypick packages because you might break something. Somewhat relevant https://bbs.archlinux.org/viewtopic.php?id=127144
On (10/24/11 18:00), Karol Blazewicz wrote: -~> On Mon, Oct 24, 2011 at 5:52 PM, Leonid Isaev <lisaev@umail.iu.edu> wrote: -~> > Besides, one really doesn't have to enable testing in pacman.conf -- individual -~> > pacman -U will do, imho. -~> -~> I've read that [testing] is all or nothing and you shouldn't -~> cherrypick packages because you might break something. -~> Somewhat relevant https://bbs.archlinux.org/viewtopic.php?id=127144 That's where brain comes in handy :) -- Leonid Isaev GnuPG key ID: 164B5A6D Key fingerprint: C0DF 20D0 C075 C3F1 E1BE 775A A7AE F6CB 164B 5A6D
On Tue, Oct 25, 2011 at 12:19 AM, Leonid Isaev <lisaev@umail.iu.edu> wrote:
On (10/24/11 18:00), Karol Blazewicz wrote: -~> On Mon, Oct 24, 2011 at 5:52 PM, Leonid Isaev <lisaev@umail.iu.edu> wrote: -~> > Besides, one really doesn't have to enable testing in pacman.conf -- individual -~> > pacman -U will do, imho. -~> -~> I've read that [testing] is all or nothing and you shouldn't -~> cherrypick packages because you might break something. -~> Somewhat relevant https://bbs.archlinux.org/viewtopic.php?id=127144
That's where brain comes in handy :)
Yes, its a REALLY good idea to state that its okay to pacman -U individual [testing] packages on a public mailing list with at least some users who really don't know any better than to do just that.
On Mon, Oct 24, 2011 at 19:38, Oon-Ee Ng <ngoonee.talk@gmail.com> wrote:
On Tue, Oct 25, 2011 at 12:19 AM, Leonid Isaev <lisaev@umail.iu.edu> wrote:
On (10/24/11 18:00), Karol Blazewicz wrote: -~> On Mon, Oct 24, 2011 at 5:52 PM, Leonid Isaev <lisaev@umail.iu.edu> wrote: -~> > Besides, one really doesn't have to enable testing in pacman.conf -- individual -~> > pacman -U will do, imho. -~> -~> I've read that [testing] is all or nothing and you shouldn't -~> cherrypick packages because you might break something. -~> Somewhat relevant https://bbs.archlinux.org/viewtopic.php?id=127144
That's where brain comes in handy :)
Yes, its a REALLY good idea to state that its okay to pacman -U individual [testing] packages on a public mailing list with at least some users who really don't know any better than to do just that.
This is the reason these discussions become useless, there are some opinions that will never change. All the warnings are in place and as I agreed with earlier "point that out to graciously" if need be. But you never learn anything if you take the safest route all the time. Learning is about experimentation. Without experimentation a person becomes stagnant. You can't tell someone you can't do that, or at least don't tell me that or I'll bust my ass and break my neck trying to do what I've been told not to do. I'm not sure but it may be that being born in the USA, working in the oil field for 30+ years where the main incentive was the line "can't get it can't stay", or maybe because I live in Texas and am one of those obstinate know it all Texans. Better put was a joke years ago by the comedian Red Skelton. It used to reside on a bill board along I10 in South Texas. When asked how you can tell a Texan, his reply was "Yep you can tell a Texan but you can't tell him much". The attitude about not telling some users who really don't know any better than to do just that grates on my nerves. It might be a better idea to put a better explanation on the wiki about what exactly might happen to those who chose to use testing. Of course I'm one of those who build there own packages from testing and trunk, keeps their own repo for base and core, and uses pacman -U to install my packages. I wouldn't have learned how to do that without making some mistakes, doing a lot of reading, and asking a few "I really should have known that" type questions. I wouldn't have learn how to fix my box when it's broken. I've now broken my own rule and made this personal on a list where that shouldn't be done. I apologize to all those whom I didn't mean to offend and did and for the excess noise on the list. Myra Nelson -- Life's fun when your sick and psychotic!
On Mon, Oct 24, 2011 at 19:38, Oon-Ee Ng <ngoonee.talk@gmail.com> wrote: -~> > On Tue, Oct 25, 2011 at 12:19 AM, Leonid Isaev <lisaev@umail.iu.edu> wrote: -~> >> On (10/24/11 18:00), Karol Blazewicz wrote: -~> >> -~> On Mon, Oct 24, 2011 at 5:52 PM, Leonid Isaev <lisaev@umail.iu.edu> wrote: -~> >> -~> > Besides, one really doesn't have to enable testing in pacman.conf -- individual -~> >> -~> > pacman -U will do, imho. -~> >> -~> -~> >> -~> I've read that [testing] is all or nothing and you shouldn't -~> >> -~> cherrypick packages because you might break something. -~> >> -~> Somewhat relevant https://bbs.archlinux.org/viewtopic.php?id=127144 -~> >> -~> >> That's where brain comes in handy :) -~> >> -~> > Yes, its a REALLY good idea to state that its okay to pacman -U -~> > individual [testing] packages on a public mailing list with at least -~> > some users who really don't know any better than to do just that. -~> > Say I want to try package X, but instead I download pkgs X, Y and Z from testing. Now my scripts which rely on /proc/.../BAT0/* fail because pkg Y is a new kernel, /dev/cdrom is gone since pkg Z is udev. And all I wanted is to try out new qemu-kvm... IMHO saying that testing is for experienced people is misleading since "experienced" is a vague term; such statements only repell users. A useful guideline would be "think three times before you type and understand how package management works". -- Leonid Isaev GnuPG key ID: 164B5A6D Key fingerprint: C0DF 20D0 C075 C3F1 E1BE 775A A7AE F6CB 164B 5A6D
On Tue, Oct 25, 2011 at 4:58 PM, Leonid Isaev <lisaev@umail.iu.edu> wrote:
Say I want to try package X, but instead I download pkgs X, Y and Z from testing. Now my scripts which rely on /proc/.../BAT0/* fail because pkg Y is a new kernel, /dev/cdrom is gone since pkg Z is udev. And all I wanted is to try out new qemu-kvm...
IMHO saying that testing is for experienced people is misleading since "experienced" is a vague term; such statements only repell users. A useful guideline would be "think three times before you type and understand how package management works".
Cherry-picking might work, but we are not even trying to make sure it does, so it would be by coincidence. -t
On Tue, Oct 25, 2011 at 10:56, Tom Gundersen <teg@jklm.no> wrote:
On Tue, Oct 25, 2011 at 4:58 PM, Leonid Isaev <lisaev@umail.iu.edu> wrote:
Say I want to try package X, but instead I download pkgs X, Y and Z from testing. Now my scripts which rely on /proc/.../BAT0/* fail because pkg Y is a new kernel, /dev/cdrom is gone since pkg Z is udev. And all I wanted is to try out new qemu-kvm...
IMHO saying that testing is for experienced people is misleading since "experienced" is a vague term; such statements only repell users. A useful guideline would be "think three times before you type and understand how package management works".
Cherry-picking might work, but we are not even trying to make sure it does, so it would be by coincidence.
-t
To All: To clarify what I do, I build my base and core packages from testing compiled to my architecture. If anything else is broken I have to find the error and fix it, wait for the broken package to be updated, or find an alternative that does work. The working alternative is sometimes better than the original. By that I mean a lighter weight app, one that forces the user to think about what they're doing, or actually works better just doens't have all the bells and whistles. That's my version of knowing how to use and take care of a computer. That's one reason I like Linux. I don't think it's for everyone, but neither do I think a blanket statement about users that don't know any better might just hinder the learning process. I once had a manager, who after I'd made a fairly serious and costly error to the tune of about $60000 circa 1978, made the statement "If you never screw up your not out there doing anything." I agree with Tom about not cherry picking just for the sake of doing so, but don't turn people off to using testing. IMHO it helps find edge cases and bugs that might not be found for years. This is turning into a rant from me about knowing how to use a computer besides turning it on and clicking the mouse. Seems like I'm highjacking an earnest discussion but can't seem to see the side that says don't do that. Maybe I should put my money where my mouth is and try to write a wiki article on the plusses and minuses of using testing and let people critique that. Myra -- Life's fun when your sick and psychotic!
Am Mon, 24 Oct 2011 11:24:26 +0200 schrieb Tom Gundersen <teg@jklm.no>:
On Mon, Oct 24, 2011 at 5:41 AM, Oon-Ee Ng <ngoonee.talk@gmail.com> wrote:
A user basically is using [testing] without fulfilling the above requirements. Should the user be advised not to use [testing] or is this counter-productive to the purpose of [testing]?
It is very important that people use testing [0], so we should really encourage _more_ rather than less of that, IMHO. Rather than advice the user to stop using testing, I'd advice them to sign up to the mailinglist :-)
Maybe this requirement should be communicated more clearly (e.g. a comment in the standard pacman.conf)?
Cheers,
Tom
[0]: a lot of the non-trivial bugs against my packages are discovered after they move to core, probably because the people with the right hardware/use-cases are not using testing.
I agree with the Wiki. From the developer's point of view it's of course better to have more users using [testing]. But when I switched to Arch Linux some years ago I was used to using the "testing" tree from Gentoo because without "testing" you wouldn't get a lot of packages there. It's because a lot of packages have never been moved to the stable tree. But when I switched to Arch Linux and used the [testing] repo, too, I once had a serious issue, which has broken my system. As a result, I had to reinstall Arch Linux completely. Maybe the reinstall wouldn't have been necessary if I wasn't new to Arch Linux. After several years I can't tell you anymore what has happened exactly. But since then I know that Arch's [testing] repo is completely different from Gentoo's "testing" tree. On the other hand I never had any serious issues with packages from the stable repos. And if there was a new package which didn't run anymore it was only this single package. And until the bug is fixed I can still use the old version. So [testing] is definitely not for users who need a stable production system or for new Arch users. Those users should stay with the stable repos. That's what they are meant for. People should use [testing] only if they know what they are doing, if they don't rely on a stable system, and if they want or are asked to help testing the packages. [testing] is not meant for having a bleeding edge system. Packages usually only stay in [testing] for a few days. So people still have a bleeding edge system if they are using the stable repos. The same question regularly appears on AUR at packages which depend or are based on another package in the repos and a new version of this dependency is put to [testing]. So, please, don't encourage every user to using [testing], and stay with what the Wiki says. Heiko
participants (8)
-
Gaetan Bisson
-
Heiko Baums
-
Karol Blazewicz
-
Leonid Isaev
-
Martti Kühne
-
Myra Nelson
-
Oon-Ee Ng
-
Tom Gundersen