On 25.07.2020 18:47, Rick McGuire wrote:
> Something to try. Switch the order of the two ::REQUIRES to see if the
> problem goes away. It
> really does look like there's something that BSF4ooRexx is doing that causes
> this behavior, but it
> is puzzling that it doesn't happen on other platforms.
Once both (BSF.CLS, rxmath) are required, no matter in which sequence, will
cause the wrong result
to be returned by rxmath:
#!/usr/bin/env rexx
say "digits():" pp(digits())
say "Before BSF is called:" rxcalcsqrt(129)
say "now invoking BSF"
-- call BSF.CLS
say "digits():" pp(digits())
say "After BSF is called:" rxcalcsqrt(129)
say "digits():" pp(digits())
say "---"
parse version v
say "Rexx version :" pp(v)
say "bsfVersion() :" pp(bsfVersion())
say ".bsf4rexx~version :" pp(.bsf4rexx~version)
say ".bsf4rexx~java.version;" pp(.bsf4rexx~java.version)
::requires BSF.CLS
::requires 'rxmath' LIBRARY
::routine pp
return "["arg(1)"]"
will yield:
rony@rony-linux:~/Dropbox/xfer/bsf4oorexx/tmp/ruurd$ rexx test4.rex
digits(): [9]
Before BSF is called: 1.10000000E+11
now invoking BSF
digits(): [9]
After BSF is called: 1.10000000E+11
digits(): [9]
---
Rexx version : [REXX-ooRexx_5.0.0(MT)_64-bit 6.05 3 May 2020]
bsfVersion() : [600.20181002 org/rexxla/bsf/engines/rexx 64-bit]
.bsf4rexx~version : [641.20200130]
.bsf4rexx~java.version; [14.0.1]
rony@rony-linux:~/Dropbox/xfer/bsf4oorexx/tmp/ruurd$
whereas requiring "rxmath" yields: 11.3578167, then immediately followed by a
"call bsf.cls" will
yield the wrong value:1.10000000E+11.
---
Ruurd suggested using "readelf -a -W libjvm.so | grep sqrt".
Both, libjvm.so and librxmath.so, are using sqrt from glibc ("GLIBC_2.2.5").
However there are small
differences between the two:
rony@rony-linux:~/Dropbox/xfer/bsf4oorexx/tmp/ruurd$ readelf -a -W
/usr/lib/jvm/adoptopenjdk-14-hotspot-amd64/lib/server/libjvm.so | grep sqrt |
grep -i glib
00000000014f0288 0000008e00000007 R_X86_64_JUMP_SLOT 0000000000000000
sqrt@GLIBC_2.2.5 + 0
00000000014f0638 0000016e00000007 R_X86_64_JUMP_SLOT 0000000000000000
sqrtf@GLIBC_2.2.5 + 0
142: 0000000000000000 0 FUNC GLOBAL DEFAULT UND sqrt@GLIBC_2.2.5
(6)
366: 0000000000000000 0 FUNC GLOBAL DEFAULT UND
sqrtf@GLIBC_2.2.5 (6)
56008: 0000000000000000 0 FUNC GLOBAL DEFAULT UNDsqrt@@GLIBC_2.2.5
56232: 0000000000000000 0 FUNC GLOBAL DEFAULT UND
sqrtf@@GLIBC_2.2.5
rony@rony-linux:~/Dropbox/xfer/bsf4oorexx/tmp/ruurd$ readelf -a -W
/usr/lib/librxmath.so | grep sqrt | grep -i glib
0000000000205128 0000001600000007 R_X86_64_JUMP_SLOT 0000000000000000
sqrt@GLIBC_2.2.5 + 0
22: 0000000000000000 0 FUNC GLOBAL DEFAULT UND sqrt@GLIBC_2.2.5
(3)
130: 0000000000000000 0 FUNC GLOBAL DEFAULT UND sqrt@@GLIBC_2.2.5
rony@rony-linux:~/Dropbox/xfer/bsf4oorexx/tmp/ruurd$
---rony
>
> On Sat, Jul 25, 2020 at 12:29 PM Rony G. Flatscher <rony.flatsc...@wu.ac.at
> <mailto:rony.flatsc...@wu.ac.at>> wrote:
>
> To clarify: to the best of my knowledge BSF4ooRexx does nothing that
> would explain this weird
> behaviour on Linux. (It works as expected on Windows and MacOS.)
>
> ---rony
>
> P.S.: Ruurd came up with the idea that maybe the Java virtual machine
> that gets loaded may cause
> this effect on Linux (both, librxmath.so and libjvm.so, use sqrt from the
> runtime). Even if
> so, this
> awkward behaviour in Rexx should not be a side-effect of using any other
> library together with
> Rexx.
>
>
> On 25.07.2020 12:12, Ruurd Idenburg wrote:
> > Running the attached script "rxmath.rex" will yield a weird result on
> my Linux (Mint 19.3,
> which
> > is based on Ubuntu 18.04) when the RxCalcSqrt function is called from
> the RXBSF.REX for a
> > non-perfect square (129).
> >
> > The code in RXBSF.REX does only ::REQUIRE BSF.CLS without using any of
> its possibilities
> (unless
> > some lines are uncommented) but in turn requires the Java library
> "libjvm.so"
> >
> > It looks like the ::REQUIRE causes the sqrt function to produce a
> different outcome of the
> > resulting floating point number.
> >
> > When I run the code on a Windows 10 machine, the results from RXBSF.REX
> are correct. Rony
> Flatcher
> > ran the scripts on Darwin (MacOS) and there the results were correct
> too.
> >
> > I included the pure Rexx ::ROUTINE my_sqrt (more or less copied from
> Walter Pachl's rxm.cls on
> > Rosetta.org) to verify that that works OK.
> >
> > Also before calling RXBSF.REX from rxmath.rex and after returning from
> RXBSF.REX the results of
> > the RxMath rxCalcSqrt function are OK again.
> >
> > Seems to me that RxMath should not be impacted by another ::REQUIRE (if
> that is the case??).
> >
> > Anybody with more system experience has some thoughts on how to tackle
> this?
> >
> > Cheers,
> >
> > Ruurd
> >
> > P.S.
> >
> > I have now switched to the use of Walter Pachl's very nice package
> (thanks Walter) that you can
> > find here:
> >
> > <http://rosettacode.org/wiki/Trigonometric_functions#ooRexx>
>
_______________________________________________
Oorexx-devel mailing list
Oorexx-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/oorexx-devel
