7.10.1RC2 compile problem? ghc: internal error: PAP object entered!

Maybe this is something I shouldn't be doing, but I thought it was worth mentioning in case I have found a compiler bug. Should I file a bug for this? cabal install *--allow-newer=base* accelerate ... [10 of 10] Compiling Data.Label.Base ( src/Data/Label/Base.hs, dist/build/Data/Label/Base.o ) ghc: internal error: PAP object entered! (GHC version 7.10.0.20150123 for x86_64_apple_darwin) Please report this as a GHC bug: http://www.haskell.org/ghc/reportabug

On Sat, Jan 31, 2015 at 8:34 AM, George Colpitts
cabal install *--allow-newer=base* accelerate
Never safe, because base contains the runtime and the runtime and the compiler are very tightly tied together. Crashes are not surprising. -- brandon s allbery kf8nh sine nomine associates allbery.b@gmail.com ballbery@sinenomine.net unix, openafs, kerberos, infrastructure, xmonad http://sinenomine.net

On Sat, Jan 31, 2015 at 10:16 AM, Brandon Allbery
On Sat, Jan 31, 2015 at 8:34 AM, George Colpitts < george.colpitts@gmail.com> wrote:
cabal install *--allow-newer=base* accelerate
Never safe, because base contains the runtime and the runtime and the compiler are very tightly tied together. Crashes are not surprising.
Actually it should always be safe: --allow-newer=base is essentially the equivalent of removing the upper bound on base from the .cabal file (of every package that was installed during that run). However, I'm quite confused about something, namely that as far as I can tell, neither accelerate nor any of its dependencies contain a module Data.Label.Base. What package was GHC trying to build when it crashed? Regards, Reid Barton

On Sat, Jan 31, 2015 at 10:47 AM, Reid Barton
On Sat, Jan 31, 2015 at 10:16 AM, Brandon Allbery
wrote: On Sat, Jan 31, 2015 at 8:34 AM, George Colpitts < george.colpitts@gmail.com> wrote:
cabal install *--allow-newer=base* accelerate
Never safe, because base contains the runtime and the runtime and the compiler are very tightly tied together. Crashes are not surprising.
Actually it should always be safe: --allow-newer=base is essentially the equivalent of removing the upper bound on base from the .cabal file (of every package that was installed during that run).
However, I'm quite confused about something, namely that as far as I can tell, neither accelerate nor any of its dependencies contain a module Data.Label.Base. What package was GHC trying to build when it crashed?
Oops, I was running the wrong command: it's in fclabels. Please file a bug report and attach the output of `cabal install --ghc-options=-v fclabels`, thanks! Regards, Reid Barton

On 2015-01-31 at 14:34:35 +0100, George Colpitts wrote:
Maybe this is something I shouldn't be doing, but I thought it was worth mentioning in case I have found a compiler bug. Should I file a bug for this?
cabal install *--allow-newer=base* accelerate ... [10 of 10] Compiling Data.Label.Base ( src/Data/Label/Base.hs, dist/build/Data/Label/Base.o ) ghc: internal error: PAP object entered! (GHC version 7.10.0.20150123 for x86_64_apple_darwin) Please report this as a GHC bug: http://www.haskell.org/ghc/reportabug
I suspect this has to do with `--allow-newer=base` allowing template-haskell-2.9.0.0 to be re-installed (which then becomes a build-dependency of `fclabels`). GHC 7.10, however, comes with template-haskell-2.10.0.0 e.g. try the following in an empty sandbox: cabal install fclabels --allow-newer=base --constraint 'template-haskell == 2.9.*' (wheares a plain `cabal install fclabels` in a fresh sandbox should succeed w/o panics with GHC 7.10)

On Sat, Jan 31, 2015 at 11:09 AM, Herbert Valerio Riedel wrote: On 2015-01-31 at 14:34:35 +0100, George Colpitts wrote: Maybe this is something I shouldn't be doing, but I thought it was worth
mentioning in case I have found a compiler bug.
Should I file a bug for this? cabal install *--allow-newer=base* accelerate
...
[10 of 10] Compiling Data.Label.Base ( src/Data/Label/Base.hs,
dist/build/Data/Label/Base.o )
ghc: internal error: PAP object entered!
(GHC version 7.10.0.20150123 for x86_64_apple_darwin)
Please report this as a GHC bug:
http://www.haskell.org/ghc/reportabug I suspect this has to do with `--allow-newer=base` allowing
template-haskell-2.9.0.0 to be re-installed (which then becomes a
build-dependency of `fclabels`). GHC 7.10, however, comes with
template-haskell-2.10.0.0 Ah yes, you're exactly right. I didn't encounter this when I tried to
reproduce the issue because I have a bunch of lines like "constraint:
template-haskell installed" in my .cabal/config file.
Regards,
Reid Barton
participants (4)
-
Brandon Allbery
-
George Colpitts
-
Herbert Valerio Riedel
-
Reid Barton