RFE: news item for new build dependencies

[a.m]

I may not be naive enough, but this is definitely noticeable:

The naive user would not have set PYTHON_TARGETS locally. As with many things, if you deviate from the default, you may sometimes need to work to stay with your deviation. In this case, the package manager did exactly what it should have done. The bad result would be if the package started to build, then died due to not having a supported Python version. Instead, you got an error message explaining what you need to change.

Fair enough. Straightening out bends like this is the cost of using Gentoo,
and I can deal with that, when I have to, but what struck a nerve in this case
was that the effects of apparently "minor tweaks" (based on all those *proto
packages getting only -rX bumps) was comparable to switching major versions,
and that without any notice.

If you really run a python2 only system at this point I would say you are part of a very tiny minority.

And, I'd posit that you have no idea what you're talking about.
Can you provide a credible source or are you pulling this from a hat?

As a very straightforward credible source, the Gentoo default is not to do Python2 only. Users are free to change this locally, but most probably won't.

I guess this is becoming hair-splitting, but while the "defaults" argument makes
it very plausible (although does not prove) that python2-only systems/users are
a minority (and that I wouldn't dispute), it says nothing about how "very tiny"
that minority happens to be, and that's the point of contention.

Right... that's what new users might use (if there were any...)

Not only new users. I have been a Gentoo user for a long time. Some of my systems explicitly leave PYTHON_TARGETS unset, so I get whatever is default in the tree of the day. When the Gentoo maintainers change that default, it applies to my system unless I decide to set PYTHON_TARGETS to override it. (I might do that if I want to migrate on a different schedule than the maintainers intend.) Based on your distrowatch argument, you seem to think that new users copy the Gentoo defaults into their local configuration and freeze at those values. Some people might do that, but it's not a good idea, and many seem not to do that.

Actually, what I was assuming is decidedly not that users would *copy* and freeze
configurations rather that the proverbial Gentoo user would tweak the hell out of their
setup in a relatively short "break-in" period and then leave that mostly unchanged,
unless major changes occur for the relevant components they use. And, that's where
having news items about such changes becomes crucial.

Now, switching build system for a package may not be considered a major change
in general, but when I have to hunt for and exclude over two dozen packages to
get a routine update to pass, I'm thinking that's somewhat out of the ordinary...
Telling me that this is all my fault for using python2 exclusively is not exactly the
kind of answer I'd be looking for; at least, not until an advance notice regarding
plans to phase out python2 has been communicated. Until then python2-only is
a valid configuration and pulling in python3 just to prematurely switch build systems
at a whim *without any notice* is predictably an unwelcome change.

BTW, since we're talking about Gentoo, I'd be *very* surprised if those who don't
change default settings were not a minority (although I'm not sure I'd go as far as
calling them a *very tiny* minority ;^P).

I guess, I'm with the thread starter on this one. If you're going to break
my system, I'd like to know why; preferably, communicated in a news item.

I agree. Now, where is your system broken? All you've shown so far is a routine USE dependency conflict.

OK, sure. See above regarding the "cost of using Gentoo".
And, you're also right that the *system* itself was not broken,
only the update process, which needed human attention due to
what I'd label an inconsiderate mistake in the tree.

The reason PYTHON_TARGETS exists is exactly to -following long-standing
gentoo tradition- let people set it as they see fit... (-;

You can set it as you see fit. In this case, you have seen fit to set it to a value that prohibits installing Meson. If this is a problem for you, modify Meson to support Python2, convince someone else to do that, or find a way to avoid Meson.

Done, as evidenced by my post listing the contents of the relevant package.mask...
And, truth be told, I got here looking for that kind of information, but I only found
bickering, which I couldn't resist getting sucked into... BUT, I'm moving on now.
I hope...

[a.m]

I just love how you twist and bend reality. (-;

Why do you feel the need to emphasize your arguments with foolery like that? Maybe because you stand on loose ground?

Is it the smiley you're calling "foolery"? In what way does that emphasize anything?
And those loose grounds you can keep for yourself...

You believe therefore you don't know.

Words of wisdom. Might actually work in the opposite direction...

If all you want to argue about is the real tininess of the minority then we can stop wasting our time right there.

Since your dismissal of "entitlement to a news item" seemed to be based on the size of
the minority, I am very interested in learning just how small (or big) a minority Gentoo
developers like yourself (assuming you're not alone in this) think is OK to inconvenience.

I'm also very interested in understanding how you'd estimate that size, since you seem
to be making assumptions about what is or isn't possible with python2-only in deployments.
None of the 400-something packages that require python3 are essential in most embedded
environments. Are you aware of any at least partially informed estimates on what segments
use/deploy Gentoo most widely? Chances are it's not the desktop...

don't assume that everyone else making that decision in 2012 is doing the same thing today.

I don't. Starting from scratch today, I might opt for python3, although as long as python2
works and I can reuse my established well-working configurations, I won't invest in a switch.

And don't assume that everyone else back then was equally naive to expect this to work forever,

Apropos, twisting&bending and loose grounds, who said anyone expected python2 to work forever?

let alone be entitled to a news item at the moment exactly their system would happen to require python3.

Perhaps you disagree, but the change we're discussing affects a major system component
and makes python3 a requirement solely based on an (arguably premature) internal decision.
I tend to think this would be a good time to post a news blurb.

Expect a news item by the time python2 support is dropped altogether.

I certainly do; and, until then, I also expect news items when *major system components* need
extra attention due to python3 becoming an unnecessary requirement.

[asturm]

Apropos, twisting&bending and loose grounds, who said anyone expected python2 to work forever?

Your bickering certainly makes it appear you did.

It is surprisingly hard to get the fact into your head that python2 being phased out is an ongoing process that started years ago. Running python2-only has always depended on the set of packages you chose to install, and has naturally become increasingly harder. You say X is a *major system component* while a lot of users not running desktops would heavily disagree, yet the existence of X more often than not implies a bigger @world set that already has a >=python3-only package somewhere else. Depending on your priorities many packages (or their revdeps) depending on >=python-3 would easily fit that *major system component* label as well. Which means that users running python2-only were/are exposed to the PYTHON_TARGETS conflict you've seen at wildly different occasions, and they did not/do not complain about it either.

I don't. Starting from scratch today, I might opt for python3, although as long as python2
works and I can reuse my established well-working configurations, I won't invest in a switch.

PYTHON_TARGETS is exactly the tool that enables you to make that gradual *transition* from your well-working configurations. Maintaining python purity just makes your *switch* harder, but if that's how you roll...

[a.m]

It is surprisingly hard to get the fact into your head that python2 being phased out is an ongoing process that started years ago.

That's not the difficult part, actually.

My difficulty is in understanding why moving an important package
(or, any package, for that matter) from working with BOTH v2 and v3
to v3-only is not news-worthy; especially, since said package has no
direct dependency on python to start with.

Or, more precisely, I'd understand if the answer was "Oops, sorry!
That didn't occur to anyone.", but all the nonsense about entitlement
feels more than a little bit odd to me. Although this may have simply
been provoked by the truly poor tone of the first post.

BTW, there aren't that many packages (~30) that made a transition to v3-only,
and most of them I would not consider an important run-time dependency,
but -then again-the latter is entirely subjective - obviously.
In any case, that might explain the lack of complaints before...

[Hu]

It's not newsworthy because users don't need help to make it happen. In the case of non-suid Xorg, user systems are expected to break if the user upgrades to non-suid Xorg and does not change the system configuration accordingly. Compensating for this in the ebuild is hard, maybe impossible. Thus, a news item was appropriate. In the case you cite, Portage identified the problem and reported it to you. You were able to solve it from that output, possibly with a bit of extra help from the forum. If this were still the dark days where Portage could neither express nor solve USE constraints, a Python-2-only user probably would have gotten a hard build break when Meson fell over. That might have warranted a news item, if only to head off bogus bug reports. Here, the ebuild language allows conforming package managers to communicate the problem to you without a news item. As a bonus, this way, no messages are shown to the people who do not need to take action.

As a slightly off-topic question, why are your paragraphs formatted so strangely? It looks like you're manually line-wrapping your text at 70 columns. Please don't do that. Modern browsers are quite capable of line-wrapping your text at an appropriate width for the viewer's screen.

[Anon-E-moose]

It is surprisingly hard to get the fact into your head that python2 being phased out is an ongoing process that started years ago.

That's not the difficult part, actually.

It's seemingly just as difficult to make some devs realize that some people prefer python2 because it's still useful
and there's a reason that it's still around and it's more than just a few posters at the gentoo forum

[Anon-E-moose]

It's not newsworthy because users don't need help to make it happen.

I've seen other "news" in the past that was just info, and that's what my original musing was about, info or in this case lack of info.
Why is it so difficult to write and post a 100 word or so fyi news that may be just info?
Are you telling me that a dev can figure out the gentoo ebuild system but can't manage to create a small info post?


Anyway, y'all have fun.

[asturm]

It's seemingly just as difficult to make some devs realize that some people prefer python2 because it's still useful
and there's a reason that it's still around and it's more than just a few posters at the gentoo forum

It is around because not all python packages have migrated so far. It does not imply at all that it is feasible to run pure-python2 at this point. But if you don't depend on certain packages, sure.

I've seen other "news" in the past that was just info, and that's what my original musing was about, info or in this case lack of info.
Why is it so difficult to write and post a 100 word or so fyi news that may be just info?

By that standard we would have already cluttered our news report system with just the same 'python-3-warning' news items about e.g. systemd, udev, fuse that had converted to meson before that, and probably a fair number of the 400+ other existing python3-only python ebuilds existing today in portage.

The perceived 'system brokenness' would have been the same with any of these preceding packages.

[a.m]

It is around because not all python packages have migrated so far.

That's an understatement...

I briefly considered switching to v3-only,
but that's _still_ not possible...

Gentoo has close to 3000 v2-only ebuilds in the tree,
although that's really only 1500+ packages,
some of which also have newer v2+v3 variants,
but v2-only is still more the norm than the exception
(compared to v3-only, that is).

Here's what keeps one of my projects squarely in v2-land today:

Code: Select all

dev-python/pygtk
dev-python/pygobject
app-emulation/virtualbox
gnome-base/libglade
net-misc/wicd
media-fonts/liberation-fonts
net-wireless/crda
sys-devel/llvm
media-gfx/graphite2
media-libs/harfbuzz
media-libs/mesa
media-libs/alsa-lib
sys-apps/usbutils
dev-libs/glib
dev-libs/libxslt

BTW, in case anyone cares or wonders,
one reason for "python-purity" is to minimize size,
which translates directly into boot time,
since the whole system lives in kernel-embedded initrd.

In those minimalist builds,
I simply cannot afford two python versions installed.

Naturally, on my dev system I do have both.

By that standard we would have already cluttered our news report system with just the same 'python-3-warning' news items about e.g. systemd, udev, fuse that had converted to meson before that, and probably a fair number of the 400+ other existing python3-only python ebuilds existing today in portage.

The perceived 'system brokenness' would have been the same with any of these preceding packages.

Sort of... except that those who run v2-only are not very likely to care about systemd,
and portage figures out all by itself that eudev can satisfy virtual/udev avoiding any issues.
Fuse doesn't sound like necessary infrastructure to me, but -again- that's likely subjective.

Also, ebuilds that start life as v3-only are inconsequential.
What matters is when a transition from v2-only or v2+v3 to v3-only is made.

[a.m]

It's not newsworthy because users don't need help to make it happen.
[...]
In the case you cite, Portage identified the problem and reported it to you.
[...]
As a bonus, this way, no messages are shown to the people who do not need to take action.

Fine. That works. I can certainly sympathize with the last point.

As a slightly off-topic question, why are your paragraphs formatted so strangely? It looks like you're manually line-wrapping your text at 70 columns. Please don't do that. Modern browsers are quite capable of line-wrapping your text at an appropriate width for the viewer's screen.

I'm well aware of what modern browsers are capable of.
That's, in fact, why I prefer to break my lines manually;
exactly where I want.

(Of course, my lines could still suffer unintended breaks due to wrap-around
if/when the viewer's device is too narrow, but I can live with that)

Does that inconvenience you in any way?

[Ant P.]

"This post is best viewed at 1024x768 in Netscape Navigator 4.0"