On Wed, Oct 09, 2019 at 05:56:18PM +0100, Ramsay Jones wrote:
> On 09/10/2019 12:22, Luc Van Oostenryck wrote:
> > Before the release, I would like to add the patch for the problem
> > you had with Cygwin. Would you be fine with the:
> > LLC_ARCH_OPTS="$LLC_ARCH_OPTS -mtriple=$(llvm-config --host-target)"
> > we had discussed before, with a note saying it's needed for recent
> > cygwin64 but may be inadequate for cygwin32?
>
> Ha, yes, I should have mentioned that, on cygwin, I actually
> tested:
>
> $ sparse --version
> v0.6.1-rc2-dirty
> $
>
> ... where the 'dirty' was:
>
> $ git diff
> diff --git a/sparsec b/sparsec
> index bafe2da..8a056e7 100755
> --- a/sparsec
> +++ b/sparsec
> @@ -39,6 +39,7 @@ case "$(uname -s)" in
> *CYGWIN*)
> # cygwin uses the sjlj (setjmp-longjmp) exception model
> LLC_ARCH_OPTS="-exception-model=sjlj"
> + LLC_ARCH_OPTS="$LLC_ARCH_OPTS -mtriple=x86_64-unknown-windows-cygnus"
> ;;
> *)
> ;;
> $
>
> Given that:
>
> $ llvm-config --host-target
> x86_64-unknown-windows-cygnus
> $
>
> ... the above should be just fine. (Do you want me to send
> a tested patch?).
No, thank you, it's not needed, I consider it as tested :)
I'll send the patch later.
> Note that I have been messing around with 'sparsec' and, in
> particular the invocation of llc, which can also output the
> object file directly. ie there is no need to generate assembler
> output and then assemble with the platform assembler. (which
> allows us to sidestep the incorrect assembler directives used
> for the wineh exception model!).
>
> However, 'sparsec' seems to have several problems (not just
> on cygwin). For example, on Linux, we see:
>
> $ ./sparsec -o hello hello.c
> /usr/bin/ld: /tmp/tmp.F57wOC: relocation R_X86_64_32 against `.rodata' can not be used when making a PIE object; recompile with -fPIC
> /usr/bin/ld: final link failed: Nonrepresentable section on output
> collect2: error: ld returned 1 exit status
> $
Yes, I'm aware of this. As far as I understand it, this is because
sparse-llvm.c doesn't yet correctly handle literal strings (and, in
general, top-level objects). If you try instead something like:
#include <stdio.h>
int main(int argc, char *argv[])
{
puts(argv[1]);
return 0;
}
then sparsec produces a functioning executable. It should also work
if you use simple global variables.
> whereas, on cygwin:
>
> $ ./sparsec -o hello hello.c
> $ ./hello.exe
> hello world
> $
>
> ... which was somewhat surprising! ;-)
That is indeed very surprising.
> Note that the testsuite is happy if 'sparcec' produces an
> object file without error - it never tests the creation of
> an executable (let alone that it actually runs and produces
> correct output).
>
> So, I don't know if we want to bust a gut to fix this up.
Yes ... Saying that sparse-llvm is not much used would not be
an overstatement. Having a good, multi-arch support for it
would need more time that I can afford and willing to spend on.
OTOH, it's not very far from being functional.
Cheers,
-- Luc
