Re: Build status for debian pspp package 0.8.3-1.2

[Ben Pfaff]

On Sun, Aug 17, 2014 at 12:52 PM, John Darrington
<address@hidden> wrote:
> On Sun, Aug 17, 2014 at 12:25:20PM -0700, Ben Pfaff wrote:
>      On Sun, Aug 17, 2014 at 11:48:02AM -0700, Ben Pfaff wrote:
>      > On Sun, Aug 17, 2014 at 04:59:24PM +0200, Friedrich Beckmann wrote:
>      > > here is the build status for the pspp 0.8.3-1.2 package:
>      > >
>      > > https://buildd.debian.org/status/package.php?p=pspp
>      > >
>      > > Still the armhf architecture build fails but due to Bens package 
> change the testsuite.log is visible. All four
>      > > failing tests fail due to a segmentation fault. The failing tests 
> are:
>      > >
>      > > 297: split-file.at:61   SPLIT FILE  - vs procedures
>      > > 485: means.at:247       MEANS harmonic and geometric means
>      > > 487: means.at:349       MEANS missing=table
>      > > 488: means.at:459       MEANS user missing values
>      >
>      > I did a build on the Debian machine "harris".  All of these tests fail
>      > with the same backtrace:
>      >
>      > #0  variable_matches_case (address@hidden, c=0x0) at 
> src/data/case.c:475
>      > #1  0xb6e35cda in case_data (address@hidden, v=0x1b15a80) at 
> src/data/case.c:267
>      > #2  0xb6f1be5e in update_n (aux1=0xbe9482d8, aux2=0x1b15a38,
>      >     user_data=0x1b189d8, c=<optimized out>, weight=1)
>      >     at src/language/stats/means.c:917
>      > #3  0xb6f56ed0 in categoricals_update (cat=0x1b18270, address@hidden)
>      >     at src/math/categoricals.c:450
>      > #4  0xb6f1c438 in run_means (ds=0x1b052e8, input=0x1b18258, 
> cmd=0xbe9482d8)
>      >     at src/language/stats/means.c:1021
>      > #5  cmd_means (lexer=<optimized out>, ds=0x1b052e8)
>      >     at src/language/stats/means.c:784
>      > #6  0xb6ef9198 in do_parse_command (state=CMD_STATE_DATA, 
> ds=<optimized out>,
>      >     lexer=0x1b05958) at src/language/command.c:233
>      > #7  cmd_parse_in_state (lexer=0x1b05958, ds=0x1b052e8, 
> state=CMD_STATE_DATA)
>      >     at src/language/command.c:148
>      > #8  0x00012702 in main (argc=<optimized out>, argv=<optimized out>)
>      >     at src/ui/terminal/main.c:138
>      >
>      > I'll take a further look.
>
>      Looking further, this looks like a compiler bug to me.  In the
>      update_n() function in src/language/stats/means.c, sometimes the
>      parameter 'c' changes from its original value to NULL between loop
>      iterations.  When I turn off compiler optimization, all of these tests
>      pass.
>
>      I don't see much value in spending much time on this.  I think that I
>      will upload a package that turns off compiler optimization on armhf.
>
> What if you make that parameter volatile?

That might avoid it too but I'm just as happy to compile without optimization
on a platform that I don't really expect anyone to use PSPP on.  Also the
Debian test machine is pretty slow so I'm not having too much fun trying
to debug there.