Correct semigroups conditional version bounds

[RyanGlScott]

Data.Semigroup was introduced in base-4.9 (GHC 8.0), but impl flag was causing vector to depend on the semigroups library even when GHC 8.0 was being used. This fixes it so that vector doesn't depend on semigroups for GHC 8.0.

[hvr]

@RyanGlScott

I don't agree with this commit because I can't reproduce the problem this is supposed to fix, and also because

if !impl(ghc > 8.0)

means, "if GHC is not used, or GHC is used and its version is less than or equal to 8.0"

however, there is no official GHC that has the version number "8.0"; it's always a 3-part version, i.e. "8.0.x".

whereas

if impl(ghc < 8.0)

means, "if GHC is used, and its version is smaller than 8.0", which is quite different semantically.

PS/EDIT: That being said, it's always suspicious if > is used rather than >= for lower version bounds, unless it's used to exclude single specific versions.

[RyanGlScott]

@hvr, I apologize, it appears I didn't understand how ! interacts with impl. I'll revert this commit.

That being said, that behavior is quite unintuitive to me—I would expect if !impl(ghc > 8.0) to be logically equivalent to impl(ghc <= 8.0), and indeed the Cabal documentation doesn't seem to mention anything about this.

Your point about makeVersion [8,0] being less than makeVersion [8,0,x] is something I didn't consider, admittedly.

[cartazio]

You can write intersections of upper and lower bounds of impl versions , or
at least cabal accepts them :)

On Wednesday, August 17, 2016, Ryan Scott notifications@github.com wrote:

@hvr https://github.com/hvr, I apologize, it appears I didn't
understand how ! interacts with impl. I'll revert this commit.

That being said, that behavior is quite unintuitive to me—I would expect if
!impl(ghc > 8.0) to be logically equivalent to impl(ghc <= 8.0), and
indeed the Cabal documentation
https://github.com/haskell/cabal/blob/9b4712583ca390d0f034cc66eca4de51c00cc8d0/Cabal/doc/developing-packages.markdown#L1717
doesn't seem to mention anything about this.

Your point about makeVersion [8,0] being less than makeVersion [8,0,x] is
something I didn't consider, admittedly.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#134 (comment), or mute
the thread
https://github.com/notifications/unsubscribe-auth/AAAQwtT3e2Jm9hOYX0v-DVLJz52ho0Nyks5qgwhHgaJpZM4JlYle
.

[RyanGlScott]

Right, that's what's interesting about !'s interaction with impl. At first glance, I would expect something like if !impl(ghc >= 7.10 && < 8.0) to be exactly the same as impl(ghc < 7.10 || ghc >= 8.0), de Morgan–style. But that's not completely true, apparently—there's an additional stipulation that "If GHC isn't used, or impl(ghc < 7.10 || ghc >= 8.0)". That part wasn't clear at all to me, since I've never seen that written down anywhere.

[cartazio]

Oh. I do the and outside the impl()

On Wednesday, August 17, 2016, Ryan Scott notifications@github.com wrote:

Right, that's what's interesting about !'s interaction with impl. At
first glance, I would expect something like if !impl(ghc >= 7.10 && < 8.0)
to be exactly the same as impl(ghc < 7.10 || ghc >= 8.0), de
Morgan–style. But that's not completely true, apparently—there's an
additional stipulation that "If GHC isn't used, or impl(ghc < 7.10 ||
ghc >= 8.0)". That part wasn't clear at all to me, since I've never seen
that written down anywhere.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#134 (comment), or mute
the thread
https://github.com/notifications/unsubscribe-auth/AAAQwh0VmbDMIZvi-CprnaQJae1PaiaJks5qgwwOgaJpZM4JlYle
.

[hvr]

@RyanGlScott it's not so obvious, and it may be a good idea to write that one down in the cabal docs (you're not the first one to be surprised about it)

IMO: impl(ghc >= x.y) is trivially equivalent to impl(ghc) && impl(ghc >= x.y), and if you negate that, you end up with !impl(ghc) || !impl(ghc >= x.y), i.e. you get that additional !impl(ghc) term.

PS: no need to apologize!