bustle and hPDB-examples failing to build

[juhp]

bustle and hPDB-examples build okay for Stackage Nightly but for some reason they are failing to build for the LTS 3 (bootstrap).

Here are the respective errors:

Unpacking to bustle-0.5.1/
:
dist/setup/setup.hs:17:18:
    Could not find module ‘Distribution.Simple.I18N.GetText’

Preprocessing executable 'Rg' for hPDB-examples-1.2.0.2...
[1 of 1] Compiling Main             ( examples/Rg.hs,
dist/build/Rg/Rg-tmp/Main.o )

<no location info>:
    ghc: panic! (the 'impossible' happened)
  (GHC version 7.10.2 for x86_64-unknown-linux):
        Simplifier ticks exhausted
  When trying UnfoldingDone c_strlen
  To increase the limit, use -fsimpl-tick-factor=N (default 100)
  If you need to do this, let GHC HQ know, and what factor you needed
  To see detailed counts use -ddump-simpl-stats
  Total ticks: 19564

Please report this as a GHC bug:  http://www.haskell.org/ghc/reportabug

I have seen these failures several times now so it is quite reproducible in the original lts-3 build environment.

Sorry maybe I should have split into 2 reports - since may well be unrelated.
Dunno if @snoyberg is working on recreating the LTS environment.

[snoyberg]

I'm not at all surprised that hPDB-examples is failing, see https://ghc.haskell.org/trac/ghc/ticket/10565. I'm quite surprised that it's succeeding though...

In fact, it probably isn't succeeding, and the previous build success is simply being retained. We should probably delete the entire nightly cache directory each time we bump GHC versions at all. I wouldn't be surprised if we see the same build failure with bustle once we do that too. I'll kick that off now.

[juhp]

Ah right, indeed

Then my suggestion would be to drop them for lts-3 for now anyway - they must both be leaf packages anyway.

Edit: I was also kind of surprised to see some warnings about 7.10.1 .hi or haddock files while building Nightly with ghc-7.10.2 the first times - I guess that was the same phenomon.

[snoyberg]

Just to be on the safe side, I'm wiping everything out and then doing clean builds from scratch for both nightly and LTS. I've already dropped hPDB-examples, and I'm trying to build bustle locally to see if there's an obvious 7.10.2 bug.

[snoyberg]

And for the record, the common denominator between these two packages is that they're both executable-only packages (no library), which is why stackage-curator didn't notice that it needed to rebuild them.

[juhp]

Ah right I see.

[juhp]

bustle seems still failing as above: cc @wjt

[wjt]

I'll take a look, but I don't have time for the next few days I'm afraid. I did see a failure like this before, on Travis – it seemed possibly related to installing hgettext (which depends on Cabal-the-library) and also upgrading Cabal-the-library, maybe there is some order-of-installation problem…

[snoyberg]

That makes a lot of sense. Unfortunately, with the current way of using the cabal executable, I can't think of a good way to solve this kind of a problem. The issue is that there are potentially three different Cabal library versions available (in the global DB, in the user DB, and what cabal-install is compiled against), and we'll run into different problems depending on which of these are out of sync with each other.

[wjt]

The path of least resistance might be to just reimplement the very minimal Cabal hooks provided by hgettext and then not use it at all in Setup.hs. It's all academic anyway since there are no non-English translations …

[snoyberg]

If you could make that change, it would be great for us, thanks!