syphilis has asked for the wisdom of the Perl Monks concerning the following question:

Hi,
This post relates to my attempts to build Math-Pari-2.010709 (against pari-2.1.7) on Windows.

Using VC7 (with a VC7-built perl) on Windows 2000 there's no problem. (I fully expect that the same holds wrt VC6.) And using MinGW's gcc compiler on Vista 64 (with a 32-bit build of perl) there's still no problem.

The fun begins when I try to build it on my Vista 64 box with ActiveState's 64-bit build of perl-5.8.8, using the very same Platform SDK that was used to build that perl. Running 'nmake' culminates in: 
C:\_64\Platform_SDK\Include\WinUser.h(9659) : error C2146: syntax erro
+r : missing ')' before identifier 'gpsi'
C:\_64\Platform_SDK\Include\WinUser.h(9659) : error C2061: syntax erro
+r : identifier 'gpsi'
C:\_64\Platform_SDK\Include\WinUser.h(9659) : error C2059: syntax erro
+r : ';'
C:\_64\Platform_SDK\Include\WinUser.h(9659) : error C2059: syntax erro
+r : ','
C:\_64\Platform_SDK\Include\WinUser.h(9660) : error C2059: syntax erro
+r : ')'
C:\_64\Platform_SDK\Include\WinUser.h(9668) : error C2146: syntax erro
+r : missing ')' before identifier 'gpsi'
C:\_64\Platform_SDK\Include\WinUser.h(9668) : error C2061: syntax erro
+r : identifier 'gpsi'
C:\_64\Platform_SDK\Include\WinUser.h(9668) : error C2059: syntax erro
+r : ';'
C:\_64\Platform_SDK\Include\WinUser.h(9668) : error C2059: syntax erro
+r : ')'
../pari-2.1.7\src\gp\highlvl.c(201) : warning C4311: 'type cast' : poi
+nter truncation from 'GEN' to 'long'
../pari-2.1.7\src\gp\highlvl.c(204) : warning C4311: 'type cast' : poi
+nter truncation from 'GEN' to 'long'
../pari-2.1.7\src\gp\highlvl.c(218) : warning C4311: 'type cast' : poi
+nter truncation from 'GEN' to 'long'
NMAKE : fatal error U1077: 'cl' : return code '0x2'
Stop.
NMAKE : fatal error U1077: 'cd' : return code '0x2'
Stop.

[download]
The obvious thing to do is to then take a look at the offending code - even though one knows that there's not going to be any syntax errors in Platform SDK's WinUser.h. Anyway, when I run my little app that retrieves lines from a file I get: 
Enter the line number
9659
Enter (full path to) file
C:\_64\Platform_SDK\Include\WinUser.h
9655: WINAPI
9656: SetScrollInfo(
9657:     __in HWND hwnd,
9658:     __in int nBar,
9659:     __in LPCSCROLLINFO lpsi,
9660:     __in BOOL redraw);
9661:
9662: WINUSERAPI
9663: BOOL
9664: WINAPI
9665: GetScrollInfo(
9666:     __in HWND hwnd,
9667:     __in int nBar,
9668:     __inout LPSCROLLINFO lpsi);
9669:

[download]
It puzzles me that there's not a 'gpsi' to be seen. (Is it possible that computers can suffer hallucinations from passive smoking ??)

I get exactly the same error if I use VC8 instead of Platform SDK, but since my VC8 is using the exact same header files, I guess it's not very surprising.

Ummm ... any advice on how to proceed ? I know it's not actually a perl problem - which is the reason that I marked it "OT".

Cheers,
Rob

Replies are listed 'Best First'.
Re: [OT] Building Math::Pari on M$ Windows
by almut (Canon) on Apr 03, 2007 at 13:44 UTC

    Hi Rob, you probably want to look at the preprocessed sources. Option /E dumps the output of the C preprocesser to stdout (i.e. with macros expanded, etc.), which might give you some clue as to what's going wrong...

    Just rerun the failing cl command with option /E added (use mouse to cut-n-paste a lengthy command line), and redirect stdout to a file for further inspection.  Good luck.

[reply]
[d/l]
[select]
Re: [OT] Building Math::Pari on M$ Windows
by cdarke (Prior) on Apr 03, 2007 at 14:11 UTC
    I have had problems like this in the past (not specifically this one) where path names have had embedded spaces. Yes, use /E and check the header file included immediately before WinUser.h.
[reply]
      Thanks almut, cdarke for the /E tip. It revealed that there's a function called 'gpsi' in the pari source, and I guess that was the root of the problem the compiler was having with the 'gpsi' identifier. I couldn't find out (didn't look too hard) what was changing the name of the identifier from 'lpsi' to 'gpsi' - in the end I just changed the 2 occurrences of 'lpsi' in WinUser.h to 'lxpsi' and that got me passed that hurdle.

      Next problem was: 
      kernel1.c(223) : error C4235: nonstandard extension used : '__asm' key
+word not supported on this architecture

      [download]
      Is that telling me that inlining of assembly code is not supported on this architecture ? Or does it just mean that one needs to use a different keyword and/or technique to inline assembly code ?

      Finally, I tried a straight C build (no asm). That compiled ok, but resulted in a number of segfaults and test failures during the running of the test suite.

      I think I'll just settle for the 32-bit build - ftm, anyway.

      Cheers,
      Rob
      Update:"ftm" eq "for the moment". (I think there are some who might have thought it meant something else :-)
[reply]
[d/l]
        Is that telling me that inlining of assembly code is not supported on this architecture ?

        Yes, probably. __asm{} was previously, only supported on x86. It would appear to indicate that they haven't yet, or maybe, are not intending to, ungrade the inline assembler to support iapx-64?

        This post on the MS forums might suggest a way around this, via the use of another non-standard extension __emit{}. But then you'd have to translate the opcode to hex yourself, which seems more than a little retro. Shades of Spectrum Duck Shoot :)

        Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
        "Science is about questioning the status quo. Questioning authority".
        In the absence of evidence, opinion is indistinguishable from prejudice.
[reply]
[d/l]
[select]

Back to Seekers of Perl Wisdom