5 ms·

The whole PVP debate really boils down to one simple question. Do you believe that a properly functioning dependency solver is important for the community? If

The whole PVP debate really boils down to one simple question. Do you believe that a properly functioning dependency solver is important for the community? If you answer yes, then I think it's pretty straightforward to get from there to the conclusion that version bounds are necessary. And by the contrapositive, if you are arguing against upper bounds, then you are actually arguing against the usefulness of a solver. If you answer no, then we should debate the merits of having a solver.

massysett 10y ago
Would it be possible to do this more like GNU configure does? The software specifies what it needs, and the system looks for a version that works. How about check the Haskell code automatically to see what functions it uses? If I'm using `text`, and all I use from that package is `pack`, `unpack`, and `cons`, I shouldn't have to specify bounds. The Cabal way of doing this--specifying version bounds--creates a big maintenance burden. And even then, it does not always work. So yes, I would question the merits of having a solver that checks some version numbers that developers manually put in. That said, I lack the know-how or resources to come up with anything new, so I'm stuck using curation or a solver. Using someone else's curation winds up being much less work for me than using a solver, so that's what I do.
- hvr_ 10y ago
  > Using someone else's curation winds up being much less work for me than using a solver, so that's what I do. :-( So basically, you value your time more than the time of us Hackage Trustees, who now have to fix the mess you're leaving behind on Hackage. That's like justifying littering because that's less work, since there's public services anyway which pick up the garbage you leave behind. If you have the time to write software, package it up, write tests, package it up, and upload to Hackage, then is it really asked too much to go that extra mile and follow the PVP to not waste our time?
  - massysett 10y ago
    dcoutts said it would not be appropriate to change Hackage to require upper bounds on software. It would be quite simple to change Hackage so that any upload would require upper bounds restricted only to package versions that already exist. Why has no such change been made? If I were in your situation I would not be spending time on menial work when a simple program could prevent that work. If such a change were made I would happily stop uploading to Hackage altogether. That no such change has been made suggests there is insufficient support for your position. I have not wasted any trustee's time, as all my software works and I keep it working or mark it deprecated. That said, I do not want trustees to fix my stuff. I have no desire to waste anyone's time. The only reason I upload to Hackage is to get stuff into Stackage. I would fully support a way to get stuff into Stackage that skips Hackage. I think it would reduce a lot of this friction. However I can understand why Stackage has no such method: Hackage allows no upper bounds. This suggests that people value a large unfragmented Hackage more than they value required upper bounds. Of course I have no role in telling you how to spend your time, but if I were you I would stop fixing software that does not play by your preferred rules.
    - hvr_ 10y ago
      > That said, I do not want trustees to fix my stuff. As soon as you upload stuff to Hackage, it enters the work-queue for Hackage Trustees. That's because the Hackage Trustees' mission is to make sure that Hackage remains in a healthy state. Which, by the way, wouldn't be needed in the first place, if everyone played voluntarily by what you call "your preferred rules", i.e. the PVP. Luckily, in many cases it's just an oversight, and in thoses cases people are actually grateful for the heads-up. Consequently, the only way to keep Hackage Trustees from messing with your stuff is to refrain from uploading to Hackage. > I would fully support a way to get stuff into Stackage that skips Hackage. I think it would reduce a lot of this friction. That's actually an intriguing idea and to be honest I am surprised that Stackage hasn't implemented something to this end already. After all, Stackage curators keep busy fighting against overly strict version bounds, while Hackage trustees are busy fighting against inaccurately lax version bounds... this is quite yin-and-yang-esque ;-)
- hvr_ 10y ago
  > Would it be possible to do this more like GNU configure does? The software specifies what it needs, and the system looks for a version that works. How about check the Haskell code automatically to see what functions it uses? This gets often suggested every time the version bounds topic comes up (to some degree, that's the direction backpack is going btw). But unfortunately this does not work: Because what this argument ignores is that you get a similar problem you have with duck-typing on a different level: You assume that the same name & type-signature provides the same semantics. However, the types usually leave too much degree of freedom to provide a proper contract. And you have to find a way to encode that contract. The best-bang-for-buck way we currently have to encode this contract is... well... the PVP! Version numbers announce the contract, while version bounds define which contract you rely on. Unless we have a better way to solve this, there's no alternative to the PVP. Unfortunately, so far I have not heard any workable solution from PVP complainers. So that's that.

massysett 10y ago

Would it be possible to do this more like GNU configure does? The software specifies what it needs, and the system looks for a version that works. How about check the Haskell code automatically to see what functions it uses? If I'm using `text`, and all I use from that package is `pack`, `unpack`, and `cons`, I shouldn't have to specify bounds. The Cabal way of doing this--specifying version bounds--creates a big maintenance burden. And even then, it does not always work. So yes, I would question the merits of having a solver that checks some version numbers that developers manually put in. That said, I lack the know-how or resources to come up with anything new, so I'm stuck using curation or a solver. Using someone else's curation winds up being much less work for me than using a solver, so that's what I do.

hvr_ 10y ago
> Using someone else's curation winds up being much less work for me than using a solver, so that's what I do. :-( So basically, you value your time more than the time of us Hackage Trustees, who now have to fix the mess you're leaving behind on Hackage. That's like justifying littering because that's less work, since there's public services anyway which pick up the garbage you leave behind. If you have the time to write software, package it up, write tests, package it up, and upload to Hackage, then is it really asked too much to go that extra mile and follow the PVP to not waste our time?
- massysett 10y ago
  dcoutts said it would not be appropriate to change Hackage to require upper bounds on software. It would be quite simple to change Hackage so that any upload would require upper bounds restricted only to package versions that already exist. Why has no such change been made? If I were in your situation I would not be spending time on menial work when a simple program could prevent that work. If such a change were made I would happily stop uploading to Hackage altogether. That no such change has been made suggests there is insufficient support for your position. I have not wasted any trustee's time, as all my software works and I keep it working or mark it deprecated. That said, I do not want trustees to fix my stuff. I have no desire to waste anyone's time. The only reason I upload to Hackage is to get stuff into Stackage. I would fully support a way to get stuff into Stackage that skips Hackage. I think it would reduce a lot of this friction. However I can understand why Stackage has no such method: Hackage allows no upper bounds. This suggests that people value a large unfragmented Hackage more than they value required upper bounds. Of course I have no role in telling you how to spend your time, but if I were you I would stop fixing software that does not play by your preferred rules.
  - hvr_ 10y ago
    > That said, I do not want trustees to fix my stuff. As soon as you upload stuff to Hackage, it enters the work-queue for Hackage Trustees. That's because the Hackage Trustees' mission is to make sure that Hackage remains in a healthy state. Which, by the way, wouldn't be needed in the first place, if everyone played voluntarily by what you call "your preferred rules", i.e. the PVP. Luckily, in many cases it's just an oversight, and in thoses cases people are actually grateful for the heads-up. Consequently, the only way to keep Hackage Trustees from messing with your stuff is to refrain from uploading to Hackage. > I would fully support a way to get stuff into Stackage that skips Hackage. I think it would reduce a lot of this friction. That's actually an intriguing idea and to be honest I am surprised that Stackage hasn't implemented something to this end already. After all, Stackage curators keep busy fighting against overly strict version bounds, while Hackage trustees are busy fighting against inaccurately lax version bounds... this is quite yin-and-yang-esque ;-)
hvr_ 10y ago
> Would it be possible to do this more like GNU configure does? The software specifies what it needs, and the system looks for a version that works. How about check the Haskell code automatically to see what functions it uses? This gets often suggested every time the version bounds topic comes up (to some degree, that's the direction backpack is going btw). But unfortunately this does not work: Because what this argument ignores is that you get a similar problem you have with duck-typing on a different level: You assume that the same name & type-signature provides the same semantics. However, the types usually leave too much degree of freedom to provide a proper contract. And you have to find a way to encode that contract. The best-bang-for-buck way we currently have to encode this contract is... well... the PVP! Version numbers announce the contract, while version bounds define which contract you rely on. Unless we have a better way to solve this, there's no alternative to the PVP. Unfortunately, so far I have not heard any workable solution from PVP complainers. So that's that.