Re: [Fink-beginners] Can't install octave-forge packages in a 32-bit OS X build, but can in a 64-bit one

[Ben Abbott]

On Nov 7, 2011, at 8:59 PM, Alexander Hansen wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> On 11/7/11 5:44 PM, Alexander Hansen wrote:
>> On 11/7/11 5:32 PM, Ben Abbott wrote:
>>> On Nov 6, 2011, at 1:10 PM, Ben Abbott wrote:
>> 
>>>> On Nov 6, 2011, at 12:52 PM, Alexander Hansen wrote:
>>>> 
>>>>> On 11/6/11 11:05 AM, Ben Abbott wrote:
>>>>> 
>>>>>> On Nov 6, 2011, at 8:49 AM, Alexander Hansen wrote:
>>>>>> 
>>>>>>> On 11/5/11 5:08 PM, Ben Abbott wrote:
>>>>>>> 
>>>>>>>> On Nov 5, 2011, at 4:57 PM, c. wrote:
>>>>>>>> 
>>>>>>>>> On 5 Nov 2011, at 21:54, Alexander Hansen wrote:
>>>>>>>>> 
>>>>>>>>>> That might be worth a shot, though on 10.6 fink's 
>>>>>>>>>> flex-devel and the system's /usr/bin/flex are at
>>>>>>>>>> the same version, so that leaves the ancient 2.5.4a
>>>>>>>>>> from our "flex" package.
>>>>>>>>> 
>>>>>>>>> No, they aren't, although the version number is the 
>>>>>>>>> same, the version shiped with the system has been 
>>>>>>>>> patched by apple. c.
>>>>>>>> 
>>>>>>>> Now that you mention that, I also remember we
>>>>>>>> encountered that before. Looking through the mail-list,
>>>>>>>> I think the offending patch was the one below.
>>>>>>>> 
>>>>>>>> http://www.opensource.apple.com/source/flex/flex-24.1/patches/scanEOF.diff
>>>>>>>> 
>>>>>>>> 
>>>>>>>> 
>> 
>>>>>>>> 
> Ben
>>>>>>> 
>>>>>>> In any case, I played around with different the
>>>>>>> available flex options (fink's flex, flex-devel [which
>>>>>>> normally has to be explicitly used because it's out of
>>>>>>> the PATH], or the system's flex) with no change in the
>>>>>>> results.
>>>>>>> 
>>>>>>> Alex
>>>>>> 
>>>>>> I reviewed some old emails.  I expect problems if
>>>>>> autogen.sh is run using Apple's version. When building the
>>>>>> docs, I had encountered errors and warning such as ...
>>>>>> 
>>>>>> warning: invalid character `' (ASCII 0) near line 92,
>>>>>> column 2
>>>>>> 
>>>>>> error: invalid character `' (ASCII 0) near line 164,
>>>>>> column 2
>>>>>> 
>>>>>> In any event, I updated by Fink packages, and am still
>>>>>> able to build octave using my build script. Over the next
>>>>>> few days I'll try modifying my script to reflect the
>>>>>> differences in Fink's info file.
>>>>>> 
>>>>>> Ben
>>>>> 
>>>>> Well and good, but can you install octave-forge packages? 
>>>>> _That's_ the problem I'm having.
>>>>> 
>>>>> Octave _builds_, with minimal warnings, for me.  And the
>>>>> builds look pretty much the same on i386, where pkg won't
>>>>> install anything, and on x86_64, where pkg works.
>>>> 
>>>> I plan to build using a script that closely follows the Fink
>>>> info file, and check that the cs-list problem is absent.
>>>> 
>>>> Once that works, I'll modify the octave343.info file and build 
>>>> and install with that.
>>>> 
>>>> At that point, I'll be able to try to install a package.
>>>> 
>>>> My builds are slow (> 1hr) and can only try in my spare time.
>>>> I expect my progress will be slow.
>>>> 
>>>> Ben
>> 
>>> I've succeeded in building Octave and installing a package.
>> 
>>> I'm running MacOS 10.6.8, my Fink version is ...
>> 
>>> $ fink --version Package manager version: 0.31.4 Distribution 
>>> version: selfupdate-rsync Fri Nov  4 13:20:59 2011, 10.6, i386 
>>> Trees: local/main stable/main stable/crypto unstable/main 
>>> unstable/crypto local/injected
>> 
>>> With the attached octave-3.4.3.info file, I'm able to build
>>> octave and install packages.
>> 
>>> octave-3.4.3:1> ver 
>>> ----------------------------------------------------------------------
>> 
>>> 
>> 
>> GNU Octave Version 3.4.3
>>> GNU Octave License: GNU General Public License Operating System: 
>>> Darwin 10.8.0 Darwin Kernel Version 10.8.0: Tue Jun  7 16:33:36
>>> PDT 2011; root:xnu-1504.15.3~1/RELEASE_I386 i386 
>>> ----------------------------------------------------------------------
>> 
>>> 
>> 
>> Package Name   | Version | Installation directory
>>> ---------------+---------+----------------------- miscellaneous
>>> *| 1.0.11 | /sw/share/octave/3.4.3/packages/miscellaneous-1.0.11
>>> optim *|  1.0.17 | /sw/share/octave/3.4.3/packages/optim-1.0.17
>>> struct *| 1.0.9 | /sw/share/octave/3.4.3/packages/struct-1.0.9
>> 
>>> The only problem I've encountered is when I try to remove 
>>> octave343, I get ...
>> 
>>> Removing octave343 ... dpkg - warning: while removing octave343, 
>>> directory `/sw/var/octave/3.4.3' not empty so not removed. dpkg
>>> - warning: while removing octave343, directory `/sw/var/octave'
>>> not empty so not removed.
>> 
>>> The change I made to the info file is to set the optimization to 
>>> "-O0". I made this change after reviewing my notes. They
>>> indicate that I had to remove "-O2" from the CXXFLAGS in order to
>>> avoid the cs-lists problem.
>> 
>>> I'm now rebuilding with "-O3" now to confirm I can reproduce the 
>>> cs-list problem, and then will try a build with "-O0" in the 
>>> CXXFLAGS and with "-O3" everywhere else.
>> 
>>> Ben
>> 
>> 
>> (fink-beginners accidentally showed up because that's my default 
>> reply-to.  But I'm keeping cc to fink-beginners and -users because 
>> we've had reports of install problems on both lists)
>> 
>> Thanks!
>> 
>> A lot of stuff got carried over blindly in the .info file from
>> 3.0.5 through 3.2.4 to 3.4.3, so I'm not surprised that an
>> optimization flag could change the results.
>> 
>> I'll try builds using both of those changes, too, just for
>> redundancy.
> 
> just having -O0 in CXXFLAGS seems to have done the trick here.  I'll
> do some more tests, but hopefully we'll be back in business for 32 bit
> Intel.
> 
> Thanks again!

I tried with -O0 thru -O4. Both "-O0" and "-O1" resolve the cs-list error. With 
the others, the cs-list error persists.

Ben