Consider using ~= as the default instead of .*

[baszalmstra]

Problem description

Currently pixi determines the latest version for all platforms and then inserts

foo = "1.2.3.*"

to the pixi.toml.

The problem with this approach is that we basically pin the dependency to a very narrow version range.

1.2.4 doesn't match 1.2.3.* for instance. Only "subversion" (1.2.3.1) would match it. I think this is a too narrow range.

Proposal

I propose that we use the ~= operator instead of the starts with postfix. ~=1.2.3 is the same as >=1.2.3,<1.3. So we would have:

foo = "~=1.2.3"

[baszalmstra]

We could also do (following the example above):

foo = "1.2.*"

if the latest version is 1.2.3. But that would also select 1.2.1 and 1.2.2 which might not be what you want.

[ruben-arts]

I prefer the ~= over the * on the patch. But the more verbose version(>=1.2.3,<1.3) is a little more readable because the ~= might be unknown to a lot of people.

[YeungOnion]

I concur that the ~= will not be as familiar. But I think the implementation for matching on a "compatible with" operator need not be a logical AND of those comparisons behind the scenes. It doesn't expand to that in rattler.

What are your thoughts on still using ~= and also generates an inline comment that specifies what ~= is logically equivalent to when pixi add is used without a spec?

[dhirschfeld]

But the more verbose version(>=1.2.3,<1.3) is a little more readable because the ~= might be unknown to a lot of people.

Even people who know about ~= might still have to double check the docs to make sure!

Given that, I think the explicit >=1.2.3,<1.3 is preferable. It also has the added benefit of being more easily manually adjusted after the fact if you want to pin to more tightly/loosely.

[SanchitMahajan236]

I want to contribute to this issue but I am not clear what I should do after reading this conversation. Can someone please help ?

[ruben-arts]

He @SanchitMahajan236,

We didn't yet come to a consensus, what the version should look like. So I don't think it needs any work right now.

[baszalmstra]

So let's make a decision here. I think there is more enthusiasm to use >=x.y.z,<x.(y+1). So lets go with that for now! @ruben-arts are you ok with that?

[ruben-arts]

For sure!