Ideas for Portage:/profiles [ANSWERED]

[pingtoo]

This is to some extents continue of this topic thread. I open this new because what I want to discuses is not direct related to that topic subject.

I am hoping ideas gathered here one day can be include into Gentoo.

The reason I think we need this is because current ::gentoo:/profiles area is not a well known. other than ebuild developer not to many user care of understand what profile can do or cause.

So let me introduce my first idea, It is more of concept than specific way of doing thing. I am thinking "constrain". What I mean is that current Gentoo is defined as rolling release and things change every days/hours. This is great and we have been enjoy this freedom for long time. However in the event that related to the qt6/wayland changes, I start wonder if it could be good to have way to keep static for a defined time frame so user have choice. Should one wish to move in slower pace than current Gentoo speed they have way to do so.

I need to step out for a bit. I just want to throw it out for your thoughts for now. I will come back later to complete/explain what I have in mind.

Thank you for your attention!

Edit: subject line. (I realize ::gentoo is too specific for Gentoo developer team. Switch to Portage for the software in general)

[pingtoo]

Continue from my previous post.

I understood that /etc/profile can serve the purpose (constrain) to some degree. However my idea is to come up a mechanism like ::gentoo:/profiles such that is can be distributed, not limit to just on per machine basis.

In my wild thinking using profile constrain concept one can create a debian profile that mimic the real debian distributions or redhat profile for Redhat derivative.

Another idea of using profile naming is mark a "milestone"? (if "version" is not a good description). I imagine if the qt6/wayland disagreement portrait in the other thread would have landed in a "milestone" it would have reduce (or eliminate) the entire "discussion".

What else can we use profile for?

Do you think profile should carry a overlay indicator? like the way ebuild have version/slot/overlay.

[pjp]

In the referenced thread, asturm posted a link to the Gentoo wiki Profile document. You really should read it, carefully.
https://wiki.gentoo.org/wiki/Profile_%28Portage%29

Once you understand the basics, regarding your ideas, pay attention to these sections of that document:

4 Profiles for developers and power users
4.2 Combining profiles
4.3 Creating custom profiles

In case that document doesn't mention it, the normal profiles typically used are within the ::gentoo ebuild repository. Specifically (by default) at /var/db/repos/gentoo/profiles/.

If you create an overlay, it may also have profiles. For example, I have the pentoo overlay available (the "default" profiles are from ::gentoo:

[bstaletic]

@pingtoo
I'm unsure if your insistence on qt6 coming with a new profile should be taken as an example, or if "a qt6 profile" is a specific point to be addressed.
Either way, profile migration from 17.1 to 23.0 was done because the compilers were reconfigured to use RELRO and BIND_NOW by default.
From 13.0 to 17.0 it was Position Independent Code.
Those are changes that affect almost everyone. At the very least everyone not already on a hardened profile.
By contrast, I couldn't care less what is happening with Qt libraries/applications - I have none on my system.

My point is that new profiles are introduced very sparingly and carefully.
A new profile every time Qt has a major release would be a lot more busywork for users.
Where do we draw the line? Do new python versions deserve new profiles? Do new perl or ruby versions deserve new profiles?
The sys-kernel/gentoo-kernel has recently started depending on perl during build. Is that worthy of a new profile?

Personally, I like the conservative approach that is employed currently.

[pingtoo]

To bstaletic, pjp

Thank you very much for respond to my idea. I really appreciate both of you got my point about the concept of constraint and a good profile evolution history.

To some degree I understand the concept of Combining profiles. May that minimal understanding lead to thinking create additional profile is simple procedure (and now I understand is NOT EASY task in term of promotion)

I have seen a few 3rd party overlays (I currently don't use any) I am not aware anybody actually create their own profiles in that 3rd party overlay.

So it is relative speaking easy to distribute a private (or public) constrained profile using overlay method.

I see from @pjp's profile evolution history that profile changes were on mostly compiler changes which affect most packages (except script language based) as well on the base-layout decision.

So as far as my concern current Portage Implementation can satisfy my idea of create constrained profile.

However I wish to hear more about what other thing deserve a profile.Because I still think profile gave choice (not just features) and Gentoo is about having freedom to choose.

Too bad at this point I am no longer can make this topic/thread a poll. I wish to have hypothetically question for every one.

If I start an overlay that will create quatally (or seasonal) for ::gentoo will you joint me to maintain the overlay?

[pjp]

[pingtoo]

pjp,

Thank you very much for you points.

English is not my first language, so my understanding of the word "profile" may be wrong. I am thinking "profile" refer to a "type" or "kind" in a sense that umbrella can have many colours, like red, black and so on. So I would say red coloured umbrella have red profile. because there could be different shape red coloured umbrella.

That is why I wish to know what other "thing" deserve a profile, for example we know currently Gentoo have "desktop" as profile. So I wonder should there be a "router" profile? A "NAS" profile?

So I guess what I mean "profile" as in the targeted usage? But can there be other kind of target? for example "time" is it a good target?

I guess asking other to "join" is wrong way to express my desire. I think what I want is too see if any one else share my idea of using "profile" as way to control/identify is good way to do something. I think may be "brainstorm" is right word. I like hear some one can point out may be this is wrong idea or this is an idea deserved more discussion.

Edit.

After I finish above and thought about it, I think I ask "join" also because I do wish someone genuinely want, not just simple talk. I am assume because one willing to put in the effort it mean they truly believe the idea is good enough for them.

For now I will settle for talk. because only discussion will help define right/wrong and/or boundaries.

[pjp]

For the general use of the word profile, a "red umbrella" is fine.

However, within Portage, the use of profile has a very specific meaning.

For a router or NAS, a Portage "set" may be the better solution. But perhaps there are USE flags that could be modified from the base profile to create your ideal for those solutions. Create them and find out.
_________________
Quis separabit? Quo animo?

[GDH-gentoo]

[pingtoo]

[eschwartz]

https://www.funtoo.org/Funtoo_Profiles

Funtoo has an interesting take on this entire profiles topic, which if I understand correctly, amounts to:

Funtoo doesn't.

Have them, that is.

Instead, Funtoo has "mixins", and you configure your system via an interactive tool to select which feature mixins you want to use, such as "X / Wayland / gnome / KDE / desktop / server / hardened / print".

Actually, under the hood this is still all just portage profiles, but it's their equivalent of "eselect profile set" that actually builds a new profile on the fly for you. It simply provides a nicer interface for them as opposed to the current one in the ::gentoo tree. It makes use of a fundamental feature of portage profiles, which is that they can inherit each other -- but as of today, Gentoo mainly uses this for internal organization reasons and code sharing, and "eselect profile list" only shows a list of pre-defined final profiles. You can build your own profiles, including making use of inheritance from the internal mixins that Gentoo provides, but it's a bit of "do it at your own risk" advanced knowledge.

Perhaps if it were easier for users to define their own ad-hoc profiles, they could have their personal profiles just inherit from "desktop & X".

[szatox]

[pingtoo]

szatox, eschwartz,

Thank you for join this conversation.

I heard about Funtoo but never really learn any thing from them. I will surely check the link to learn more.

The "mixins" idea seems to be something I would interested.

I should try to investigate the different approach of USE flags. It is always nice to learn from a different angle.

[Leonardo.b]

Portage profiles does the very same thing of /etc/portage/* files.
They include the same files, where they set defaults and so on.
The only complication is inheritance, as profiles inherits each other and get a bit messy, but overall it's not so complicated.

[pingtoo]

Leonardo.b,

Thank you for join and sharing.

From another thread I found someone did use a "packages" file in /etc/portage/profile/. So yes it is very much a private profile on the node.

However when I start out for this idea is that it may be good to have some way to share. In a sense if a group of people share same desire for a fixed "profile" and in such they accept the defined scope and limit of the "profile". My idea of such "profile" have defined scope and changes when scope change accepted by the group.

Back then I am not sure if this is possible. And now it looks like it is very possible from technical point of view.

Funtoo's "mixins" and the tool "ego" seems to match the need.

I plan later to try if "ego" is straight drop in to Gentoo Portage or some thing need to modify.

[asturm]

[pingtoo]

[Hu]

If you don't want changes, then you could just not run emerge --sync or not run emerge --update --deep options @world. For systems that are kept frozen in time, this is a very simple and effective way to avoid changes. If you want some changes, you may be forced to take others, due to limitations in compatibility between what you want to change and what you want to leave unchanged.

[pingtoo]

[szatox]

[pingtoo]

I agree. The distribution method most likely will be using overlay.

I imagine however the overlay initially will only have profile(s) + use + mask + unmask. additional ebuild(s) get populate into overlay when current Portage tree ebuild moved out of scope of the defined profile scope.

For example a profile scope include semi-annually range and scope also define accept month change. then may be for example January past and February come, the minor profile update will happen. (Note, this is just conceptual, I don't know if this is doable)

As far as my current understanding, overlay priority is a choice of user. So ideally a user believe/agree my idea of limited scope profile concept should set this profile with higher priority. And if the profile is setup correctly will have mask to prevent unintentionally upgrade.

[szatox]

[pingtoo]

[szatox]

So, is the idea to keep a profile from ::gentoo linked as make.profile and link in a patch from your overlay as local customization via portage/profile?
Well, if you make all package.{use,mask,accept_keywords} etc directories, it could even allow linking in multiple snippets, possibly from various overlays, effectively achieving something like mixins. Portage does support this directory structure, so making it usable would just require a simple script to create and delete all necessary links in several directories with a single command. An eselect module would be great for users' convenience, but a shell loop is enough of a prototype.

Portage has sets too, which could be used along with those profile patches to actually install the software affected by the profile patch. using those 2 together might make a complete solution for adding functionality to the system in a form of ready-to-use blocks rather than IKEA boxes.
_________________
Make Computing Fun Again

[pingtoo]

szatox,

When I initially start out was doing stage4 build with catalyst. I raise a question about using /etc/portage/profile/packages but did not get any answer. However later I saw another post that indicate that is how the OP got it work so I learn that is possible case. But in between I have settle my workflow use the "sets" concept.

as the concept of "mixins" I rather like it. although I don't know if it will or it is the better idea for doing what I proposed here I don't know yet. I am very happy that you too think it is possible way to go.

I will continue develop this idea, will update if I got something more concreted.