Is epoch increment optional on version scheme change in -git packages?

dreieck · 2024-07-16 13:48:45

Dear Arch people,

regarding this discussion about the AUR-package gr-lora_sdr-git I want to ask about the `$epoch` variable in `-git` packages in the AUR.

The maintainer argues that `$epoch` may be used, but not has to, when the versioning scheme changes so that a newer version of the `PKGBUILD` would result in a lower version according to `pacman`'s version compare logic.

In this case, the old `$pkgver` (before adoption and fixing of the AUR package) was `8133965`, the new one now is `0.5.r4.g8133965`.

The maintainer argues as following to not increment `$epoch`:

xiota wrote:

AUR is still build on demand. AUR helper and Chaotic AUR users will receive notification and can force the update if they want.
Users unwilling to manually force the upgrade can keep the old version. It's ultimately their choice.

I was reasoning that `$epoch` _must_ be incremented in such a case:

dreirund wrote:

xiota wrote:
Also, I don't use epoch on AUR packages because they're build on demand.
But for the packages in the AUR you still need to set `$epoch`. Since the AUR is independently usable from `chaotic-aur`.
The version scheme change you did in `gr-lora_sdr-git` requires an increment of `$epoch` in the AUR package, since the old version was `8133965`, the new one is `0.5.r4.g8133965`, which is lower by version compare logic.
So setting `epoch=1` in the AUR package actually is required.

The wiki seems not to be clear about it, it may be read in both ways:

One standpoint:

dreirund wrote:

Regardless what opinions anyone has, the guidelines are clear about that:
https://wiki.archlinux.org/title/PKGBUILD#epoch
https://wiki.archlinux.org/title/PKGBUILD#epoch wrote:
It is used when the version numbering scheme of a package changes (or is alphanumeric), breaking normal version comparison logic.

and answer to that:

xiota wrote:

From your own reference:
quote=https://wiki.archlinux.org/title/PKGBUILD#epoch wrote:
Warning: epoch should only be used when absolutely required to do so.
There is no statement requiring epoch to be used. Only requirements to meet when it is used.

So, is there any clear official rule for this case?

---

EDIT: After a discussion over there, the maintainer has moved the discussion thread "to a repository you don't have access to.".

Regards!

Last edited by dreieck (2024-07-16 15:07:38)

stanczew · 2024-07-16 14:35:59

vercmp <old> <new>

If the result is > 0, increment $epoch.

(Disclaimer: not an official statement.)

Last edited by stanczew (2024-07-16 14:36:58)

Lone_Wolf · 2024-07-16 15:08:52

The discussion is on chaotic-aur github pages and chaotic-aur has no relation with Archlinux or Archlinux User Repository.
They just use AUR as one of the sources to retrieve build files from.

I am not aware of an official rule about this, but if there is/was it would need to come from https://wiki.archlinux.org/title/Arch_P … _Standards .
(better known as Arch package guidelines nowadays) .
You can try starting a discussion there.

My personal POV :

If you look up the aur packages I maintain you'll notice several do use epoch (one even has epoch 3 ) , including vcs packages.

The argument "Also, I don't use epoch on AUR packages because they're build on demand." strikes me as simply wrong as I know many users of aur packages that build & update them regularly .
This is especially true for vcs packages tracking fast moving upstreams like kernel, mesa and llvm/clang to name a few .

Trilby · 2024-07-16 15:19:23

There's a clearly silly word-nitpicking argument there that 1) Epoch should only be used when "required", and 2) the guidelines say when it should be used, but they don't explicitly use the word "required", therefore there's the conclusion that it should not be used??

Following that logic, Epoch would never ever be used and should not exist at all. But it does exist. It exists for a purpose. That purpose is precisely the conditions which you are facing. The version numbering scheme has changed. Increment or add an epoch. I don't see how this could be even remotely controversial aside from someone deliberately trolling the language used in the wiki.

Last edited by Trilby (2024-07-16 15:20:56)

dreieck · 2024-07-16 17:38:43

Lone_Wolf wrote:

The discussion is on chaotic-aur github pages and chaotic-aur has no relation with Archlinux or Archlinux User Repository.
They just use AUR as one of the sources to retrieve build files from.

That discussion is relevant here because it is the same person who also maintains the AUR package. And that AUR package has to follow the AUR rules.

Actually, the discussion on the github site was moved to a repository without read access (not accessible anymore), and I was removed from the (co) maintainer list afterwards (to cut off the possibility to increment `$epoch`?)

Anyway, if we come to a conclusion here that `$epoch` should be incremented in the given case, I think AUR maintainers somehow should take action.

If we come to the conclusion that no `$epoch` increment is necessary here, we should leave it as it is.

Regards!

Trilby · 2024-07-16 17:56:11

If the maintainer thinks he's not supposed to add / increment epoch, he's mistaken.

If the maintainer removed you from it just over this, he's an asshat.

But on the flip side, if TUs had to take action against every mistaken asshat on the AUR, they'd never sleep.

WorMzy · 2024-07-16 18:05:10

Please raise this on the aur-general mailing list, very few Package Maintainers (AUR custodians) frequent the forums, so you are unlikely to get an authoritative response.

dreieck · 2024-07-16 19:59:35

WorMzy wrote:

Please raise this on the aur-general mailing list

Done.

EDIT: Not done, I could not send the email:

aur-general-owner(at)lists.archlinux.org wrote:

Your message to the Aur-general mailing-list was rejected for the following
reasons:
The message is not from a list member

Can anyone who is subscribed to the list report?

I do not want to subscribe because I do not want to receive the emails.

For reference, this is what I tried to send:

Subject: Is epoch increment optional on version scheme change in -git packages?
Dear AUR-maintainers,
in this post [¹] I was told to raise this issue on this mailing list,
so I do so.
Note that I am not part of the mailing list; if there is an authorative
answer maybe the best is to post it at [¹] so everyone can read.
The issue is described in full in [²] and some short discussion
followed.
I also copypaste the issue here (from the Forum -- so forum-syntax):
---
[... copypaste of the opeining post of this thread ...]
---
Regards!,
@dreieck.

[¹] https://bbs.archlinux.org/viewtopic.php … 6#p2184346
[²] https://bbs.archlinux.org/viewtopic.php?pid=2184286

Regards!

Last edited by dreieck (2024-07-17 08:21:35)

dreieck · 2024-07-23 08:55:50

WorMzy wrote:

Please raise this on the aur-general mailing list

Now here.

Lone_Wolf · 2024-07-23 11:03:08

Thanks.

Note that the list content is public and you can follow it through aur general archives .

dreieck · 2024-07-23 15:55:15

P.S.:

Since I also maintain packages, I am also personally interested in this issue.

It happened that your package, @xiota, did "trigger" this issue, but I am generally interested in the authorative answer, also for my own packages (for now, I did increment `$epoch` when the version scheme changed so that new version <= old version, but if it turns out that I should not I will not do anymore and maybe revert my changes in the packages I maintain).

So do not take it personally at this point that I ask in the forum and now in the mailing list (where I was referred to), and if you have opinions on the discussion or my "campaign", best to discuss publically I think.

Regards!

Last edited by dreieck (2024-07-23 16:00:45)

Trilby · 2024-07-23 16:02:49

dreieck wrote:

... if it turns out that I should not...

I don't see that as even possible. The question seems to be whether it is an absolute requirement to do so - I'd argue it is (or used to be) though it could be argued that the relevant documentation doesn't make this 100% clear. However, there is absolutely no basis on which to say that the epoch must not be incremented in these cases.

In other words, drawing on RFC 2119 terms, it may be unclear whether the guideline to increment epoch when the version numbering scheme changes falls into either a SHOULD/RECOMMENDED category or a MUST category - but there's absolutely no way it would fall into a SHOULD NOT / MUST NOT category.

Arch Linux

#1 2024-07-16 13:48:45

Is epoch increment optional on version scheme change in -git packages?

#2 2024-07-16 14:35:59

Re: Is epoch increment optional on version scheme change in -git packages?

#3 2024-07-16 15:08:52

Re: Is epoch increment optional on version scheme change in -git packages?

#4 2024-07-16 15:19:23

Re: Is epoch increment optional on version scheme change in -git packages?

#5 2024-07-16 17:38:43

Re: Is epoch increment optional on version scheme change in -git packages?

#6 2024-07-16 17:56:11

Re: Is epoch increment optional on version scheme change in -git packages?

#7 2024-07-16 18:05:10

Re: Is epoch increment optional on version scheme change in -git packages?

#8 2024-07-16 19:59:35

Re: Is epoch increment optional on version scheme change in -git packages?

#9 2024-07-23 08:55:50

Re: Is epoch increment optional on version scheme change in -git packages?

#10 2024-07-23 11:03:08

Re: Is epoch increment optional on version scheme change in -git packages?

#11 2024-07-23 15:55:15

Re: Is epoch increment optional on version scheme change in -git packages?

#12 2024-07-23 16:02:49

Re: Is epoch increment optional on version scheme change in -git packages?

Board footer