Via-C unregisterised build fails for integer-simple

[singpolyma]

The cross-compiling toolchain I used can be had from  https://developer.blackberry.com/develop/platform_choice/ndk.html

./configure --target=arm-unknown-nto-qnx8.0.0eabi --enable-unregistered

I will attach my build.mk to this ticket, but I did *not* set it to use LLVM for this attempt.

I have also attached two make.log files. One that succeeds up to #7490, and another quite short one that fails very fast with:

when making flags consistent: Warning:
    Compiler unregisterised, so compiling via C
/tmp/ghc25531_0/ghc25531_0.hc: In function 'c2pA_entry':

/tmp/ghc25531_0/ghc25531_0.hc:3691:1:
     warning: this decimal constant is unsigned only in ISO C90 [enabled by default]

/tmp/ghc25531_0/ghc25531_0.hc:3691:17:
     error: expected ')' before numeric constant

Can current integer-simple not be compiled via-C for unregistered?

[singpolyma]

Log of make up until expected failure

[singpolyma]

Failure on running make again

[singpolyma]

Output of the configure command

[singpolyma]

A useful output of the configure script

[psycotica0]

Running

"inplace/bin/ghc-stage1" -static  -H64m -O0 -fasm    -package-name integer-simple-0.1.1.0 \
-hide-all-packages -i -ilibraries/integer-simple/. -ilibraries/integer-simple/dist-install/build \
-ilibraries/integer-simple/dist-install/build/autogen -Ilibraries/integer-simple/dist-install/build \
-Ilibraries/integer-simple/dist-install/build/autogen -Ilibraries/integer-simple/.    -optP-include \
-optPlibraries/integer-simple/dist-install/build/autogen/cabal_macros.h -package ghc-prim-0.3.1.0  \
-package-name integer-simple -Wall -XHaskell98 -XCPP -XMagicHash -XBangPatterns -XUnboxedTuples \
-XForeignFunctionInterface -XUnliftedFFITypes -XNoImplicitPrelude -O -fasm  -no-user-package-db \
-rtsopts      -odir libraries/integer-simple/dist-install/build -hidir libraries/integer-simple/dist-install/build \
-stubdir libraries/integer-simple/dist-install/build -hisuf hi -osuf  o -hcsuf hc \
-c libraries/integer-simple/./GHC/Integer/Type.hs \
-o libraries/integer-simple/dist-install/build/GHC/Integer/Type.o

Which is what make runs, gives me the same error. Adding -keep-hc-file allows one to inspect the generated C-Code.

The problem line looks like:

_s1ny = (_s1nj--2147483648) + (_s1nk + _s1ns);

Which is not valid.

A few of the lines around use 0x80000000U, which happens to be the same binary value on 32-bit systems. It also corresponds to "half_bound_up" in the Type.hs that generated this bad C.

So, I don't know if this issue is that the code generator should spit out (_s1nj-0x80000000U) like it does in other places, or if the issue is that it should spit out (_s1nj-(-2147483648)), but I don't know enough about the Code Generator to find it myself right now.

[psycotica0]

Oh, also, this was just on arm-linux-gnueabi unregisterised, so QNX is not to blame, it just doesn't generate valid C right now.

[psycotica0]

Ok, so, as I expected, I just confirmed that on my computer, even without ARM, just with unregisterised and my local i386, it fails with the same error.

The code generator just spits out bad C.

[psycotica0]

Patch To Fix Code Gen Issue

[igloo]

The patch looks right to me, but I think there might be a larger problem here.

halfBoundUp and fullBound are of type Digit, which is a synonym for Word#, so why do we end up representing them with Int?

[simonmar]

The patch is a bit heavy-handed - the special case for negative offsets is not an optimisation, it's there to make the generated C easier to read, and it's pretty important (IMHO).

It would be good to look into @igloo's concerns too while we're here. I'm also wondering why my unregisterised builds using integer-simple didn't run into this...

[simonmar]

I attached another patch, if someone experiencing the problem could verify that it works, that would be great.

[psycotica0]

This patch seems to work as well as mine.

For the line in question it provides the same output, but I can understand not wanting to have every negative number be represented as _blah+-number.

[marlowsd@…]

commit b91f3d260f3ccb57e8fcd61f943c6fadf26391c8

Author: Simon Marlow <marlowsd@gmail.com>
Date:   Mon Jan 28 10:04:34 2013 +0000

    hopefully fix #7620

 compiler/cmm/PprC.hs |    7 +++++--
 1 files changed, 5 insertions(+), 2 deletions(-)