tests failing for .xs routines

Greetings. Since I moved the innermost routines of Crypt::Tea_JS into
C in an .xs file, it works really well and lots faster. Except that in
some environments (e.g. Cygwin, Freebsd) testers report an error like:

[ERROR] [Sun Feb 20 18:06:28 2005] MAKE TEST failed: No such file or
directory /usr/bin/perl.exe "-Iblib/lib" "-Iblib/arch" test.pl

or

[ERROR] [Tue Mar 15 22:34:06 2005] MAKE TEST failed: No such file or
directory PERL_DL_NONLAZY=1 /usr/local/bin/suidperl "-Iblib/lib"
"-Iblib/arch" test.pl

and then all the tests involving the .xs routines fail :-(
Which is the missing file ? What's invoking it ?
How should I work round this ?

All help gratefully received, Regards, Peter

--

TAS/DPIWE/CIT/Servers hbt/lnd/l8 6233 3061 http://www.pjb.com.au
And how sweet a story it is, when you hear Charley Parker tell it
- Kerouac, Mexico City Blues

Re: tests failing for .xs routines

"Peter Billam" wrote in message
news:slrndk3f4e.s9o.peter@pjb.dpiwe.tas.gov.au...
> Greetings. Since I moved the innermost routines of Crypt::Tea_JS into
> C in an .xs file, it works really well and lots faster. Except that in
> some environments (e.g. Cygwin, Freebsd) testers report an error like:
>
> [ERROR] [Sun Feb 20 18:06:28 2005] MAKE TEST failed: No such file or
> directory /usr/bin/perl.exe "-Iblib/lib" "-Iblib/arch" test.pl
>

I don't know what that message relates to - maybe the particular tester(s)
could help out there. (If test.html didn't get built, perhaps it's
complaining about that.)

I notice that all failures happen on 64-bit int builds, and that the tests
that fail are the four tests that involve negative integers. The binary
(base 2) representation of the positive integers in your test.pl will be
identical on 64-bit and 32-bit ints. But the base 2 representation of any
negative integers on a 32-bit system differs from the base 2 representation
of the same negative integer on a 64-bit system. Could that somehow be
throwing things out ? Seems to me that if you've got code that relies on the
bit structure of integers (and I don't know whether or not that's the case),
then the answer could be "yes".

Cheers,
Rob

Re: tests failing for .xs routines

In article <43425905$0$6924$afc38c87@news.optusnet.com.au>, Sisyphus wrote:
>
> "Peter Billam" wrote:
>> Greetings. Since I moved the innermost routines of Crypt::Tea_JS into
>> C in an .xs file, it works really well and lots faster. Except that in
>> some environments (e.g. Cygwin, Freebsd) testers report an error like:
>> [ERROR] [Sun Feb 20 18:06:28 2005] MAKE TEST failed: No such file or
>> directory /usr/bin/perl.exe "-Iblib/lib" "-Iblib/arch" test.pl
> ...
> I notice that all failures happen on 64-bit int builds,

Ahh, well spotted, thank you :-)

> and that the tests
> that fail are the four tests that involve negative integers. The binary
> (base 2) representation of the positive integers in your test.pl will be
> identical on 64-bit and 32-bit ints. But the base 2 representation of any
> negative integers on a 32-bit system differs from the base 2 representation
> of the same negative integer on a 64-bit system. Could that somehow be
> throwing things out ? Seems to me that if you've got code that relies on the
> bit structure of integers (and I don't know whether or not that's the case),
> then the answer could be "yes".
>
That's probably the issue. It's going to need a detailed look. If
I'm lucky, only some integer constants in test.pl need to be changed.
The fact that asciidigest tested OK gives me reason to hope :-)

Thank you Sisyphus, most helpful! Regards, Peter

--

TAS/DPIWE/CIT/Servers hbt/lnd/l8 6233 3061 http://www.pjb.com.au
And how sweet a story it is, when you hear Charley Parker tell it
- Kerouac, Mexico City Blues

Re: tests failing for .xs routines

Peter Billam wrote:
> Greetings. Since I moved the innermost routines of Crypt::Tea_JS into
> C in an .xs file, it works really well and lots faster. Except that in
> some environments (e.g. Cygwin, Freebsd) testers report an error like:
> [ERROR] [Sun Feb 20 18:06:28 2005] MAKE TEST failed: No such file or
> directory /usr/bin/perl.exe "-Iblib/lib" "-Iblib/arch" test.pl

It's a strange error message, though, isn't it, this "No such file or
directory" ?

In article <43425905$0$6924$afc38c87@news.optusnet.com.au>, Sisyphus wrote:
> I notice that all failures happen on 64-bit int builds, ...
> the tests that fail are the four tests that involve negative integers.

Peter Billam wrote:
> If I'm lucky, only some integer constants in test.pl need to be changed.
> The fact that asciidigest tested OK gives me reason to hope :-)

Yup, the 64-bit testers descended in droves on 2.14, which merely
has the -ve integers in test.pl re-expressed in octal, and

2.14 (6 PASSes)
* 256942 PASS 5.8.5 on Solaris 2.9 (sun4-solaris-thread-multi)
* 256758 PASS 5.8.7 on Cygwin 1.5.18(0.13242) (cygwin-thread-multi-64int)
* 256718 PASS 5.8.7 on Linux 2.4.30 (i686-linux-thread-multi-64int-ld)
* 256691 PASS 5.8.7 on Freebsd 5.4-stable (i386-freebsd-64int)
* 256444 PASS 5.8.6 on Freebsd 5.4-release (i386-freebsd-64int)
* 256443 PASS 5.8.7 on Linux 2.4.21-37.elsmp (i686-linux)

That looks heaps better :-) Thanks again, Regards, Peter

Regards, Peter

--

TAS/DPIWE/CIT/Servers hbt/lnd/l8 6233 3061 http://www.pjb.com.au
And how sweet a story it is, when you hear Charley Parker tell it
- Kerouac, Mexico City Blues

Re: tests failing for .xs routines

"Peter Billam" wrote in message
news:slrndkb803.k6.peter@pjb.dpiwe.tas.gov.au...
> Peter Billam wrote:
> > Greetings. Since I moved the innermost routines of Crypt::Tea_JS into
> > C in an .xs file, it works really well and lots faster. Except that in
> > some environments (e.g. Cygwin, Freebsd) testers report an error like:
> > [ERROR] [Sun Feb 20 18:06:28 2005] MAKE TEST failed: No such file or
> > directory /usr/bin/perl.exe "-Iblib/lib" "-Iblib/arch" test.pl
>
> It's a strange error message, though, isn't it, this "No such file or
> directory" ?
>

Yes, indeed - it puzzles me as to where that comes from - and I don't think
I'd ever work it out without access to one of those machines. On the face of
it, the expectation appears to be that the entire string (/usr/bin/perl.exe
\"-Iblib/lib\" \"-Iblib/arch\" test.pl) is the name of a file/directory -
and it comes as no surprise that such a file/directory does not exist. How
that expectation comes about, however, is a mystery.

Google was of no help.

(Nice of you to post this follow-up, btw - usually we're just "left to
wonder" :-)

Cheers,
Rob