ByteString builder causes the optimiser to run out of ticks

I received the following error message from GHC:

<no location info>:
    ghc: panic! (the 'impossible' happened)
  (GHC version 7.10.3 for x86_64-unknown-linux):
	Simplifier ticks exhausted
  When trying UnfoldingDone $fStorableWord21
  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: 8040

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

When I followed the suggestion of adding -fsimpl-tick-factor=200 the compiler proceeds normally (finding a syntax error in a file). Removing this option, strangely, also causes the compiler to proceed normally. I can no longer reproduce this bug, although several compilations in a row (before adding the option) reproduced the bug.

Although I cannot reproduce this bug, the syntax error that is present came from an error of the form:

f :: A -> M B
f a = 
  do
    x <- g a
  let
    c = h a
  in do
    ...

The actual file is attached.

Trac metadata

Trac field	Value
Version	8.0.1
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

Attached file DB.hs ($1226).

The file containing the syntax error that the compiler was eventually able to find.

The error is because GHC was doing too much inlining. The syntax error is completely unrelated. Maybe if you try with -fforce-recomp you will be able to reproduce the problem?

Adding -fforce-recomp did in fact cause the bug to resurface. I've used this to isolate the value I need for -fsimpl-tick-factor to cause the compiler to succeed. In particular, -fsimpl-tick-factor=110 causes the compiler to fail, but -fsimpl-tick-factor=111 causes the compiler to succeed.

Please provide the code which causes the inlining to happen. These kinds of reports are usually because a library author is over zealous adding inline pragmas to everything.

Attached file ghc-bug.zip (download).

A snapshot of the code that can exhibit the bug

I have attached a ZIP file containing a snapshot of the project. The project currently can be compiled via cabal build create-tables. However, if you remove the -fsimpl-tick-factor=131 in the recipes.cabal file (or decrease the number to 130 or below) the bug appears.

From initial inspection, it seems that the reason for the message is that postgresql-simple uses bytestrings to represent the query type. Basically every function in the bytestring library is marked inline so when the String is converted to a Bytestring using the builder function which maps over the whole string, a lot of inlining is then exposed which GHC dutifully performs. The reason this trips things up is that your strings are quite long I think and there are quite a few of them in the file.

So I'm not sure what to suggest here, it seems the best would be to just keep the flag in your cabal file as the compile time didn't seem too long to me.

closed

changed title from GHC panic to ByteString builder causes the optimiser to run out of ticks

Seems like https://github.com/haskell/bytestring/issues/45 is related. so I will close this.

Trac metadata

Trac field	Value
Resolution	Unresolved → ResolvedDuplicate

added compiler crash label

added Pnormal label