Re: [Fink-users] python35-3.5.0-1 fails tests on 10.11/7.2

[Daniel Johnson]

> On Dec 10, 2015, at 5:33 PM, Daniel Johnson  
> wrote:
> 
>> 
>> On Dec 10, 2015, at 5:19 PM, Hanspeter Niederstrasser 
>>  wrote:
>> 
>> 
>> On Thu, December 10, 2015 3:49 pm, Alexander Hansen wrote:
>>> 
 On Dec 10, 2015, at 13:42, Jacques Bloch
  wrote:
 
 Somehow i was able to use DYLD_LIBRARY_PATH with armadillo which i put
 in ~/opt/lib on my system.
 
 Jacques
 
 Sent from my iPhone
>>> 
>>> 
>>> That may be implementation-dependent.  If armadillo just wants _a_
>>> DYLD_LIBRARY_PATH and doesn’t rely on it being an environment variable,
>>> that may still work.
>>> 
>>> Fionna:~ hansen$ DYLD_LIBRARY_PATH="/sw/lib"
>>> Fionna:~ hansen$ export DYLD_LIBRARY_PATH
>>> Fionna:~ hansen$ printenv DYLD_LIBRARY_PATH
>>> Fionna:~ hansen$ echo $DYLD_LIBRARY_PATH
>>> /sw/lib
>> 
>> From what I understand, DYLD_LIBRARY_PATH still works, but it just doesn't
>> get passed to subshells.
>> 
>> So
>> 
>> DYLD_LIBRARY_PATH=/foo/lib /bin/sh BAR.sh
>> 
>> will use /foo/lib successfully, but
>> 
>> DYLD_LIBRARY_PATH=/foo/lib ./BAR.sh
>> 
>> (where BAR.sh starts with #!/bin/sh) will fail.
>> 
> 
> Specifically, all DYLD_* environment variables get removed when running any 
> binary that is considered “protected”. That includes anything in /bin, /sbin, 
> /usr/bin and /usr/sbin. So the problem is that you can’t pass DYLD_* to 
> /bin/sh since it’s protected. If you were to use fink’s bash it WILL work 
> since it isn’t protected but that would mean a dependency on bash. That’s 
> kind of an expensive work around.
> 
> DYLD_FALLBACK_LIBRARY_PATH is the one exception that does get passed but it’s 
> useless for overriding an already installed library.

Ok, the tests on python27, python34 and python35 should be fixed now. What I 
had to do is change the library install_name during the tests and change back 
after. However, that caused _sysconfigdata.py to be regenerated during make 
install with references to %b. This is actually harmless but fink’s validator 
doesn’t like it. To get around it I had to run make a second time during 
InstallScript, patch the file then run make install. Took a while to figure 
that out. It Works For Me but let me know if you encounter any issues.

Daniel



signature.asc
Description: Message signed with OpenPGP using GPGMail
--
___
Fink-users mailing list
Fink-users@lists.sourceforge.net
List archive:
http://news.gmane.org/gmane.os.macosx.fink.user
Subscription management:
https://lists.sourceforge.net/lists/listinfo/fink-users

Re: [Fink-users] python35-3.5.0-1 fails tests on 10.11/7.2

[Jacques Bloch]

Somehow i was able to use DYLD_LIBRARY_PATH with armadillo which i put in 
~/opt/lib on my system.

Jacques

Sent from my iPhone

> On 10 Dec 2015, at 22:14, Alexander Hansen  
> wrote:
> 
> DYLD_LIBRARY_PATH


--
___
Fink-users mailing list
Fink-users@lists.sourceforge.net
List archive:
http://news.gmane.org/gmane.os.macosx.fink.user
Subscription management:
https://lists.sourceforge.net/lists/listinfo/fink-users

Re: [Fink-users] python35-3.5.0-1 fails tests on 10.11/7.2

[Daniel Johnson]

> On Dec 10, 2015, at 5:19 PM, Hanspeter Niederstrasser 
>  wrote:
> 
> 
> On Thu, December 10, 2015 3:49 pm, Alexander Hansen wrote:
>> 
>>> On Dec 10, 2015, at 13:42, Jacques Bloch
>>>  wrote:
>>> 
>>> Somehow i was able to use DYLD_LIBRARY_PATH with armadillo which i put
>>> in ~/opt/lib on my system.
>>> 
>>> Jacques
>>> 
>>> Sent from my iPhone
>> 
>> 
>> That may be implementation-dependent.  If armadillo just wants _a_
>> DYLD_LIBRARY_PATH and doesn’t rely on it being an environment variable,
>> that may still work.
>> 
>> Fionna:~ hansen$ DYLD_LIBRARY_PATH="/sw/lib"
>> Fionna:~ hansen$ export DYLD_LIBRARY_PATH
>> Fionna:~ hansen$ printenv DYLD_LIBRARY_PATH
>> Fionna:~ hansen$ echo $DYLD_LIBRARY_PATH
>> /sw/lib
> 
> From what I understand, DYLD_LIBRARY_PATH still works, but it just doesn't
> get passed to subshells.
> 
> So
> 
> DYLD_LIBRARY_PATH=/foo/lib /bin/sh BAR.sh
> 
> will use /foo/lib successfully, but
> 
> DYLD_LIBRARY_PATH=/foo/lib ./BAR.sh
> 
> (where BAR.sh starts with #!/bin/sh) will fail.
> 

Specifically, all DYLD_* environment variables get removed when running any 
binary that is considered “protected”. That includes anything in /bin, /sbin, 
/usr/bin and /usr/sbin. So the problem is that you can’t pass DYLD_* to /bin/sh 
since it’s protected. If you were to use fink’s bash it WILL work since it 
isn’t protected but that would mean a dependency on bash. That’s kind of an 
expensive work around.

DYLD_FALLBACK_LIBRARY_PATH is the one exception that does get passed but it’s 
useless for overriding an already installed library.

Daniel




signature.asc
Description: Message signed with OpenPGP using GPGMail
--
___
Fink-users mailing list
Fink-users@lists.sourceforge.net
List archive:
http://news.gmane.org/gmane.os.macosx.fink.user
Subscription management:
https://lists.sourceforge.net/lists/listinfo/fink-users

Re: [Fink-users] python35-3.5.0-1 fails tests on 10.11/7.2

[Hanspeter Niederstrasser]

> Yep, that’s exactly what’s going on. Python uses DYLD_LIBRARY_PATH to
> find the just-built library and that doesn’t work anymore. In theory, I
> could use install_name_tool to change the library path before tests and
> reset it after, but that would require changing a LOT of files since all
> the C-based modules point to libpython. Maybe just changing python.exe
> would suffice but I’ll need to play with it a while. It isn’t going to
> be fixed right away but I will work on it. Fortunately this only effects
> tests.

Would using rpath work to get around the DYLD_* blockage?  Easiest would
be to just add -rpath  to the linking step of the
executable(s). But do the libraries needto have an install_name that
starts with @rpath in order for the rpath linker option to work?

Hanspeter

-- 
More agile than a turtle, stronger than a mouse, nobler than a lettuce


--
___
Fink-users mailing list
Fink-users@lists.sourceforge.net
List archive:
http://news.gmane.org/gmane.os.macosx.fink.user
Subscription management:
https://lists.sourceforge.net/lists/listinfo/fink-users

Re: [Fink-users] python35-3.5.0-1 fails tests on 10.11/7.2

[Alexander Hansen]

> On Dec 10, 2015, at 13:42, Jacques Bloch 
>  wrote:
> 
> Somehow i was able to use DYLD_LIBRARY_PATH with armadillo which i put in 
> ~/opt/lib on my system.
> 
> Jacques
> 
> Sent from my iPhone


That may be implementation-dependent.  If armadillo just wants _a_ 
DYLD_LIBRARY_PATH and doesn’t rely on it being an environment variable, that 
may still work.

Fionna:~ hansen$ DYLD_LIBRARY_PATH="/sw/lib"
Fionna:~ hansen$ export DYLD_LIBRARY_PATH
Fionna:~ hansen$ printenv DYLD_LIBRARY_PATH
Fionna:~ hansen$ echo $DYLD_LIBRARY_PATH
/sw/lib

—akh
--
___
Fink-users mailing list
Fink-users@lists.sourceforge.net
List archive:
http://news.gmane.org/gmane.os.macosx.fink.user
Subscription management:
https://lists.sourceforge.net/lists/listinfo/fink-users

Re: [Fink-users] python35-3.5.0-1 fails tests on 10.11/7.2

[Jacques Bloch]

why can we not put libraries in /usr/local/lib which is not locked by SIP?

J.

---
Dr. Jacques C.R. Bloch
Institute for Theoretical Physics
University of Regensburg
93040 Regensburg, Germany
tel: +49 941 943-2018
fax: +49 941 943-3887
email: jacques.bl...@ur.de
web: http://homepages.uni-regensburg.de/~blj05290



> On 10 Dec 2015, at 18:15, Alexander Hansen  
> wrote:
> 
> 
> 
>> On Dec 10, 2015, at 08:51, Daniel Johnson > > wrote:
>> 
>>> 
>>> On Dec 10, 2015, at 6:52 AM, Hanspeter Niederstrasser 
>>> > wrote:
>>> 
>>> [370/396/3] test_platform
>>> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>>> Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>>> Reason: image not found
>>> test test_platform failed -- Traceback (most recent call last):
>>> File 
>>> "/sw/build.build/python35-3.5.0-1/Python-3.5.0/Lib/test/test_platform.py", 
>>> line 287, in test_popen
>>>   self.assertEqual(hello, "Hello")
>>> AssertionError: '' != 'Hello'
>>> + Hello
>>> ...
>>> [339/396/2] test_popen
>>> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>>> Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>>> Reason: image not found
>>> test test_popen failed -- Traceback (most recent call last):
>>> File 
>>> "/sw/build.build/python35-3.5.0-1/Python-3.5.0/Lib/test/test_popen.py", 
>>> line 33, in test_popen
>>>   ["foo", "bar"]
>>> File 
>>> "/sw/build.build/python35-3.5.0-1/Python-3.5.0/Lib/test/test_popen.py", 
>>> line 26, in _do_test_commandline
>>>   got = eval(data)[1:] # strip off argv[0]
>>> File "", line 0
>>> 
>>>   ^
>>> SyntaxError: unexpected EOF while parsing
>>> ...
>>> [183/396/1] test_subprocess
>>> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>>> Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>>> Reason: image not found
>>> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>>> Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>>> Reason: image not found
>>> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>>> Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>>> Reason: image not found
>>> ...
>>> FAILED (failures=1, skipped=1)
>>> test test_platform failed
>>> 3 tests failed again:
>>>   test_platform test_popen test_subprocess
>>> 
>>> If I install python35 w/out tests and then retest, they all pass.  So 
>>> 10.11's 'protection' seems to strike again.
>>> 
>>> Hanspeter
>> 
>> Yep, that’s exactly what’s going on. Python uses DYLD_LIBRARY_PATH to find 
>> the just-built library and that doesn’t work anymore. In theory, I could use 
>> install_name_tool to change the library path before tests and reset it 
>> after, but that would require changing a LOT of files since all the C-based 
>> modules point to libpython. Maybe just changing python.exe would suffice but 
>> I’ll need to play with it a while. It isn’t going to be fixed right away but 
>> I will work on it. Fortunately this only effects tests.
>> 
>> Daniel
>> 
> 
> That seems pretty nasty, yeah.  I’m assuming that DYLD_FALLBACK_LIBRARY_PATH 
> won’t do what you need, since it wouldn’t avoid a library from the previously 
> installed version?
> 
> Maybe we need a new InfoTest field like “SipTestUnderSIP: true” which will 
> skip the test suite for build bots and display a message for power users like
> 
> WARNING:  Skipping test suite because it is not compatible with SIP (thanks 
> for that, Apple).  If you want to run the test suite, you will need to 
> disable SIP by using .
> 
> -- 
> Alexander Hansen, Ph.D.
> Fink User Liaison
> 
> --
> ___
> Fink-users mailing list
> Fink-users@lists.sourceforge.net
> List archive:
> http://news.gmane.org/gmane.os.macosx.fink.user
> Subscription management:
> https://lists.sourceforge.net/lists/listinfo/fink-users

--
___
Fink-users mailing list
Fink-users@lists.sourceforge.net
List archive:
http://news.gmane.org/gmane.os.macosx.fink.user
Subscription management:
https://lists.sourceforge.net/lists/listinfo/fink-users

Re: [Fink-users] python35-3.5.0-1 fails tests on 10.11/7.2

[Alexander Hansen]

Fink doesn’t put things in /usr/local to avoid having files overwritten by 
third-party installs.  

That wouldn’t help here, anyway, because the issue is that SIP prevents the use 
of DYLD_LIBRARY_PATH at all.  Even to point to even an *arbitrary* unprotected 
library directory like 
"/sw/src/fink.build/mypackage-1.2.3-1/mypackage-1.2.3/src/.libs” as might be 
expected for a test suite to allow linking to a library in a build directory 
rather than in its final installed location.  

You’d experience the same thing even if building a package by hand to install 
in /usr/local, since the test suite is run in the build directory.

—akh

> On Dec 10, 2015, at 12:59, Jacques Bloch  wrote:
> 
> why can we not put libraries in /usr/local/lib which is not locked by SIP?
> 
> J.
> 
> ---
> Dr. Jacques C.R. Bloch
> Institute for Theoretical Physics
> University of Regensburg
> 93040 Regensburg, Germany
> tel: +49 941 943-2018
> fax: +49 941 943-3887
> email: jacques.bl...@ur.de 
> web: http://homepages.uni-regensburg.de/~blj05290 
> 
> 
> 
>> On 10 Dec 2015, at 18:15, Alexander Hansen > > wrote:
>> 
>> 
>> 
>>> On Dec 10, 2015, at 08:51, Daniel Johnson >> > wrote:
>>> 
 
 On Dec 10, 2015, at 6:52 AM, Hanspeter Niederstrasser 
 > wrote:
 
 [370/396/3] test_platform
 dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
 Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
 Reason: image not found
 test test_platform failed -- Traceback (most recent call last):
 File 
 "/sw/build.build/python35-3.5.0-1/Python-3.5.0/Lib/test/test_platform.py", 
 line 287, in test_popen
   self.assertEqual(hello, "Hello")
 AssertionError: '' != 'Hello'
 + Hello
 ...
 [339/396/2] test_popen
 dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
 Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
 Reason: image not found
 test test_popen failed -- Traceback (most recent call last):
 File 
 "/sw/build.build/python35-3.5.0-1/Python-3.5.0/Lib/test/test_popen.py", 
 line 33, in test_popen
   ["foo", "bar"]
 File 
 "/sw/build.build/python35-3.5.0-1/Python-3.5.0/Lib/test/test_popen.py", 
 line 26, in _do_test_commandline
   got = eval(data)[1:] # strip off argv[0]
 File "", line 0
 
   ^
 SyntaxError: unexpected EOF while parsing
 ...
 [183/396/1] test_subprocess
 dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
 Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
 Reason: image not found
 dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
 Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
 Reason: image not found
 dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
 Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
 Reason: image not found
 ...
 FAILED (failures=1, skipped=1)
 test test_platform failed
 3 tests failed again:
   test_platform test_popen test_subprocess
 
 If I install python35 w/out tests and then retest, they all pass.  So 
 10.11's 'protection' seems to strike again.
 
 Hanspeter
>>> 
>>> Yep, that’s exactly what’s going on. Python uses DYLD_LIBRARY_PATH to find 
>>> the just-built library and that doesn’t work anymore. In theory, I could 
>>> use install_name_tool to change the library path before tests and reset it 
>>> after, but that would require changing a LOT of files since all the C-based 
>>> modules point to libpython. Maybe just changing python.exe would suffice 
>>> but I’ll need to play with it a while. It isn’t going to be fixed right 
>>> away but I will work on it. Fortunately this only effects tests.
>>> 
>>> Daniel
>>> 
>> 
>> That seems pretty nasty, yeah.  I’m assuming that DYLD_FALLBACK_LIBRARY_PATH 
>> won’t do what you need, since it wouldn’t avoid a library from the 
>> previously installed version?
>> 
>> Maybe we need a new InfoTest field like “SipTestUnderSIP: true” which will 
>> skip the test suite for build bots and display a message for power users like
>> 
>> WARNING:  Skipping test suite because it is not compatible with SIP (thanks 
>> for that, Apple).  If you want to run the test suite, you will need to 
>> disable SIP by using .
>> 
>> -- 
>> Alexander Hansen, Ph.D.
>> Fink User Liaison
>> 

--
___
Fink-users mailing list

Re: [Fink-users] python35-3.5.0-1 fails tests on 10.11/7.2

[Alexander Hansen]

> On Dec 10, 2015, at 13:14, Alexander Hansen  
> wrote:
> 
> 
> Fink doesn’t put things in /usr/local to avoid having files overwritten by 
> third-party installs.  
> 
> That wouldn’t help here, anyway, because the issue is that SIP prevents the 
> use of DYLD_LIBRARY_PATH at all.  Even to point to even an *arbitrary* 
> unprotected library directory like 
> "/sw/src/fink.build/mypackage-1.2.3-1/mypackage-1.2.3/src/.libs” as might be 
> expected for a test suite to allow linking to a library in a build directory 
> rather than in its final installed location.  
> 
> You’d experience the same thing even if building a package by hand to install 
> in /usr/local, since the test suite is run in the build directory.
> 
> —akh
> 

The above should have said “the test suite is run against libraries in the 
build directory”.  




--
___
Fink-users mailing list
Fink-users@lists.sourceforge.net
List archive:
http://news.gmane.org/gmane.os.macosx.fink.user
Subscription management:
https://lists.sourceforge.net/lists/listinfo/fink-users

Re: [Fink-users] python35-3.5.0-1 fails tests on 10.11/7.2

[Hanspeter Niederstrasser]

On Thu, December 10, 2015 3:49 pm, Alexander Hansen wrote:
>
>> On Dec 10, 2015, at 13:42, Jacques Bloch
>>  wrote:
>>
>> Somehow i was able to use DYLD_LIBRARY_PATH with armadillo which i put
>> in ~/opt/lib on my system.
>>
>> Jacques
>>
>> Sent from my iPhone
>
>
> That may be implementation-dependent.  If armadillo just wants _a_
> DYLD_LIBRARY_PATH and doesn’t rely on it being an environment variable,
> that may still work.
>
> Fionna:~ hansen$ DYLD_LIBRARY_PATH="/sw/lib"
> Fionna:~ hansen$ export DYLD_LIBRARY_PATH
> Fionna:~ hansen$ printenv DYLD_LIBRARY_PATH
> Fionna:~ hansen$ echo $DYLD_LIBRARY_PATH
> /sw/lib

>From what I understand, DYLD_LIBRARY_PATH still works, but it just doesn't
get passed to subshells.

So

DYLD_LIBRARY_PATH=/foo/lib /bin/sh BAR.sh

will use /foo/lib successfully, but

DYLD_LIBRARY_PATH=/foo/lib ./BAR.sh

(where BAR.sh starts with #!/bin/sh) will fail.

Hanspeter


-- 
More agile than a turtle, stronger than a mouse, nobler than a lettuce


--
___
Fink-users mailing list
Fink-users@lists.sourceforge.net
List archive:
http://news.gmane.org/gmane.os.macosx.fink.user
Subscription management:
https://lists.sourceforge.net/lists/listinfo/fink-users

Re: [Fink-users] python35-3.5.0-1 fails tests on 10.11/7.2

[Alexander Hansen]

> On Dec 10, 2015, at 08:51, Daniel Johnson  wrote:
> 
>> 
>> On Dec 10, 2015, at 6:52 AM, Hanspeter Niederstrasser 
>>  wrote:
>> 
>> [370/396/3] test_platform
>> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>> Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>> Reason: image not found
>> test test_platform failed -- Traceback (most recent call last):
>> File 
>> "/sw/build.build/python35-3.5.0-1/Python-3.5.0/Lib/test/test_platform.py", 
>> line 287, in test_popen
>>   self.assertEqual(hello, "Hello")
>> AssertionError: '' != 'Hello'
>> + Hello
>> ...
>> [339/396/2] test_popen
>> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>> Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>> Reason: image not found
>> test test_popen failed -- Traceback (most recent call last):
>> File "/sw/build.build/python35-3.5.0-1/Python-3.5.0/Lib/test/test_popen.py", 
>> line 33, in test_popen
>>   ["foo", "bar"]
>> File "/sw/build.build/python35-3.5.0-1/Python-3.5.0/Lib/test/test_popen.py", 
>> line 26, in _do_test_commandline
>>   got = eval(data)[1:] # strip off argv[0]
>> File "", line 0
>> 
>>   ^
>> SyntaxError: unexpected EOF while parsing
>> ...
>> [183/396/1] test_subprocess
>> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>> Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>> Reason: image not found
>> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>> Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>> Reason: image not found
>> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>> Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>> Reason: image not found
>> ...
>> FAILED (failures=1, skipped=1)
>> test test_platform failed
>> 3 tests failed again:
>>   test_platform test_popen test_subprocess
>> 
>> If I install python35 w/out tests and then retest, they all pass.  So 
>> 10.11's 'protection' seems to strike again.
>> 
>> Hanspeter
> 
> Yep, that’s exactly what’s going on. Python uses DYLD_LIBRARY_PATH to find 
> the just-built library and that doesn’t work anymore. In theory, I could use 
> install_name_tool to change the library path before tests and reset it after, 
> but that would require changing a LOT of files since all the C-based modules 
> point to libpython. Maybe just changing python.exe would suffice but I’ll 
> need to play with it a while. It isn’t going to be fixed right away but I 
> will work on it. Fortunately this only effects tests.
> 
> Daniel
> 

That seems pretty nasty, yeah.  I’m assuming that DYLD_FALLBACK_LIBRARY_PATH 
won’t do what you need, since it wouldn’t avoid a library from the previously 
installed version?

Maybe we need a new InfoTest field like “SipTestUnderSIP: true” which will skip 
the test suite for build bots and display a message for power users like

WARNING:  Skipping test suite because it is not compatible with SIP (thanks for 
that, Apple).  If you want to run the test suite, you will need to disable SIP 
by using .

-- 
Alexander Hansen, Ph.D.
Fink User Liaison

--
___
Fink-users mailing list
Fink-users@lists.sourceforge.net
List archive:
http://news.gmane.org/gmane.os.macosx.fink.user
Subscription management:
https://lists.sourceforge.net/lists/listinfo/fink-users

Re: [Fink-users] python35-3.5.0-1 fails tests on 10.11/7.2

[Daniel Johnson]

> On Dec 10, 2015, at 6:52 AM, Hanspeter Niederstrasser 
>  wrote:
> 
> [370/396/3] test_platform
> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>  Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>  Reason: image not found
> test test_platform failed -- Traceback (most recent call last):
>  File 
> "/sw/build.build/python35-3.5.0-1/Python-3.5.0/Lib/test/test_platform.py", 
> line 287, in test_popen
>self.assertEqual(hello, "Hello")
> AssertionError: '' != 'Hello'
> + Hello
> ...
> [339/396/2] test_popen
> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>  Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>  Reason: image not found
> test test_popen failed -- Traceback (most recent call last):
>  File "/sw/build.build/python35-3.5.0-1/Python-3.5.0/Lib/test/test_popen.py", 
> line 33, in test_popen
>["foo", "bar"]
>  File "/sw/build.build/python35-3.5.0-1/Python-3.5.0/Lib/test/test_popen.py", 
> line 26, in _do_test_commandline
>got = eval(data)[1:] # strip off argv[0]
>  File "", line 0
> 
>^
> SyntaxError: unexpected EOF while parsing
> ...
> [183/396/1] test_subprocess
> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>  Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>  Reason: image not found
> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>  Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>  Reason: image not found
> dyld: Library not loaded: /sw/lib/python3.5/config-3.5m/libpython3.5m.dylib
>  Referenced from: /sw/build.build/python35-3.5.0-1/Python-3.5.0/python.exe
>  Reason: image not found
> ...
> FAILED (failures=1, skipped=1)
> test test_platform failed
> 3 tests failed again:
>test_platform test_popen test_subprocess
> 
> If I install python35 w/out tests and then retest, they all pass.  So 10.11's 
> 'protection' seems to strike again.
> 
> Hanspeter

Yep, that’s exactly what’s going on. Python uses DYLD_LIBRARY_PATH to find the 
just-built library and that doesn’t work anymore. In theory, I could use 
install_name_tool to change the library path before tests and reset it after, 
but that would require changing a LOT of files since all the C-based modules 
point to libpython. Maybe just changing python.exe would suffice but I’ll need 
to play with it a while. It isn’t going to be fixed right away but I will work 
on it. Fortunately this only effects tests.

Daniel




signature.asc
Description: Message signed with OpenPGP using GPGMail
--
___
Fink-users mailing list
Fink-users@lists.sourceforge.net
List archive:
http://news.gmane.org/gmane.os.macosx.fink.user
Subscription management:
https://lists.sourceforge.net/lists/listinfo/fink-users