Compile-time regression in 8.2 when compiling bloodhound's test suite

GHC 8.2 takes much longer to compile bloodhound's test suite than GHC 8.0.2. With 8.0.2, compilation takes approximately a minute on my machine, 8.2 takes approximately four minutes and uses much more memory.

How to reproduce:

$ git clone -b ghc-8.2 https://github.com/23Skidoo/bloodhound.git
$ cd bloodhound
$ cabal new-build -w ghc-8.0.1 --enable-tests
$ cabal new-build -w ghc-8.2.1 --enable-tests 

GHC version:

$ ghc-8.2.1 --version
The Glorious Glasgow Haskell Compilation System, version 8.2.0.20170505

These warnings (fixed in my ghc-8.2 branch of bloodhound) may provide a clue to what's happening:

tests/V1/tests.hs:1876:17: warning: [-Wsimplifiable-class-constraints]
    • The constraint ‘SOP.All SOP.SListI (SOP.GCode a)’
        matches an instance declaration
      instance forall k (f :: k -> Constraint) (xs :: [k]).
               (Generics.SOP.Constraint.AllF f xs, SOP.SListI xs) =>
               SOP.All f xs
        -- Defined in ‘Generics.SOP.Constraint’
      This makes type inference for inner bindings fragile;
        either use MonoLocalBinds, or simplify it using the instance
    • In the type signature:
        sopArbitrary :: forall a.
                        (Generic a,
                         SOP.GTo a,
                         SOP.All SOP.SListI (SOP.GCode a),
                         SOP.All2 Arbitrary (SOP.GCode a)) =>
                        Gen a

Trac metadata

Trac field	Value
Version	8.2.1-rc2
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture

changed weight to 5

added Tbug Trac import labels

I don't think the warning, per se, indicates much. However, there are definitely some strange things going on with these types. Lots of type-level recursion. I had hoped that the fix for the specializer generating loopy code might have done something good here, but as far as I can tell this is still an issue in the released 8.2.1.

added compiler perf label

added Pnormal label

Unfortunately the original regression has become difficult to reproduce:

$ git clone -b ghc-8.2 https://github.com/23Skidoo/bloodhound.git
Cloning into 'bloodhound'...
fatal: Remote branch ghc-8.2 not found in upstream origin

I've been able to build bloodhound's testsuite on the current master with GHC 8.2.2 to 8.10.4, but didn't find anything remarkable. Compile times were around 30–40s, peak allocations around 600MB.

I've also attempted building the testsuites in bloodhound 0.12.0.0, 0.13.0.0 and 0.14.0.0 but didn't manage to build them with GHC 8.2 for various reasons.

@trac-refold Do you think you could provide a different way to reproduce the original issue?

Thanks @sjakobi, going to close this as fixed.

closed

FWIW, I don't know whether the regression was fixed in GHC. AFAIK the testsuite has also changed considerably since this issue was opened, so maybe this took care of the compile time issue. Building it with GHC 8.0.2 fails, so I can't reproduce the regression.