Using a binary 2.0.0 I've compiled 2.0.4 without problems, running on Debian 4.0 "Etch" on an Ultra-10.

Using this 2.0.4 I'm trying to compile 2.1.4 using recent sources. First problem is a missing line as fixed in the patch below:

--- fpcsrc/rtl/linux/sparc/stat.inc~ 2007-03-29 07:26:34.000000000 +0000 +++ fpcsrc/rtl/linux/sparc/stat.inc 2007-07-04 13:07:52.000000000 +0000 @@ -15,7 +15,8 @@

  {$IFNDEF FPC_USE_LIBC}

    Stat = packed record  // No unix typing because of differences
-    0:
+    case byte of
+     0:
        (dev   : culonglong;
        ino    : culonglong;
        mode   : cuint;

Hopefully I've got that right. However with that applied a build all fails with:

..
/usr/src/fpc/fpcbuild-2.1.4/fpcsrc/compiler/ppc3 -Ur -Xs -O2 -n -Fusparc -Fusystems -Fu/usr/src/fpc/fpcbuild-2.1.4/fpcsrc/rtl/units/sparc-linux -Fisparc -FE. -FUsparc/units/sparc-linux -dRELEASE -dsparc -dGDB -dBROWSERLOG pp.pas
/bin/mv -f ./pp ppcsparc
make[5]: Leaving directory `/usr/src/fpc/fpcbuild-2.1.4/fpcsrc/compiler' make echotime
make[5]: Entering directory `/usr/src/fpc/fpcbuild-2.1.4/fpcsrc/compiler' Start now 14:04:23
make[5]: Leaving directory `/usr/src/fpc/fpcbuild-2.1.4/fpcsrc/compiler' make[4]: Leaving directory `/usr/src/fpc/fpcbuild-2.1.4/fpcsrc/compiler' /usr/bin/
Binary files ppc3 and ppcsparc differ
make[3]: *** [cycle] Error 2
make[3]: Leaving directory `/usr/src/fpc/fpcbuild-2.1.4/fpcsrc/compiler' make[2]: *** [compiler_cycle] Error 2
make[2]: Leaving directory `/usr/src/fpc/fpcbuild-2.1.4/fpcsrc' make[1]: *** [build-stamp.sparc-linux] Error 2
make[1]: Leaving directory `/usr/src/fpc/fpcbuild-2.1.4/fpcsrc' make: *** [fpcsrc/build-stamp.sparc-linux] Error 2

Anybody got any suggestions please?

Op Wed, 4 Jul 2007, schreef Mark Morgan Lloyd:

This one is correct.

Hopefully I've got that right. However with that applied a build all fails with: Anybody got any suggestions please?

This could mean mean incorrect code is generated which cause the resulting compiler not to crash but generate even more incorrect code. The Sparc code generator needs to be checked for bugs.

This is unfortunate because as we are really close to freezing 2.2.0 it'll be hard to get a fix in time.

Daniël

There are none for this particular problem. I've done a lot of sparc fixes after 2.1.4 was released and this problem is now rarer, but still occurs every now and then. It's caused by indeterminism in the register allocator. See
http://www.freepascal.org/mantis/view.php?id=8121

I haven't seen it happen at all lately in 2.3.1, but I have no idea why that is (and whether this is pure coincidence, whether the bug is still there but somehow hidden by other code, or whether it is actually fixed there.

Jonas

One thing I didn't mention is that I followed as near as I could manage to an identical build procedure on i386 and then SPARC, and it's only the latter that errored. It's mainly SPARC I'm interested in rather than i386, largely for the sake of Lazarus.

Which takes me to the interesting observation that with 2.0.4 Lazarus 0.9.22 bombs during startup but that this disappears if the suspect code is preceded by a DebugLn(). I'd concluded reluctantly that that might be a code-generation problem and wanted to progress to a more recent FPC version before raising it.

What's the best way to move forwards- try to disable the check, investigate the unofficial 2.1.4 .deb, or jump straight to a later development version?