[Scons-users] Determining source of scons crash

Bill Deegan bill at baddogconsulting.com
Sat Apr 25 18:43:12 EDT 2020


Shouldn't crash.
Normally you'd see a warning/error about file generated by two different
builders...

Go ahead and file a github issue with a simple reproducer?

On Fri, Apr 24, 2020 at 3:01 PM Luke Robison <lukerobison at gmail.com> wrote:

> I guess the question is: should this be a valid Sconstruct? If not, should
> Scons issue a error/warning to the user, or should it crash?
>
> env = Environment()
> src = ['test1.f90']
> env.SharedLibrary('testlib-shared',src)
> env.StaticLibrary('testlib-static',src)
>
> On Fri, Apr 24, 2020 at 4:04 PM Bill Deegan <bill at baddogconsulting.com>
> wrote:
>
>> Luke,
>>
>> Not all platforms differentiate the object suffixes between shared and
>> not shared so you could have similar issue with c code..
>> Specify a different target name for one or the other and I think your
>> issue should go away.
>> -Bill
>>
>> On Fri, Apr 24, 2020 at 1:54 PM Luke Robison <lukerobison at gmail.com>
>> wrote:
>>
>>> Correct.  For example, I'm sure with C code you could make a
>>> StaticLibrary and a SharedLibrary, those would require .o and .os files
>>> respectively (compiled from the same source code).  The problem is unique
>>> to the fact that gfortran (and ifort, and most other fortran compilers)
>>> emit a .mod file each time they compile a source file which includes a
>>> "module" statement, and while the .o/.os files deconflict, the .mod files
>>> would be named the same.
>>>
>>> To make matters a little worse: My experience is that ifort produces a
>>> unique .mod file each time you re-compile the same source (I suspect they
>>> have a timestamp in there for some reason -- this forces scons to rebuild
>>> more than necessary!).  Luckily gfortran re-produces the exact same if you
>>> re-compile the same source.
>>>
>>>
>>> On Fri, Apr 24, 2020 at 3:22 PM Bill Deegan <bill at baddogconsulting.com>
>>> wrote:
>>>
>>>> Hmm..
>>>> so .o and.os but they're both generating the same .mod filename?
>>>>
>>>> On Fri, Apr 24, 2020 at 1:21 PM Bill Deegan <bill at baddogconsulting.com>
>>>> wrote:
>>>>
>>>>> So you're building both an object and a shared object from the same
>>>>> source?
>>>>> Without specifying a target to create a different named file?
>>>>>
>>>>>
>>>>> On Fri, Apr 24, 2020 at 12:26 PM Luke Robison <lukerobison at gmail.com>
>>>>> wrote:
>>>>>
>>>>>> Minor Correction: In my case my library was making a StaticLibrary
>>>>>> and I was calling env.SharedObject manually.
>>>>>>
>>>>>> Additionally, on Linux you can remove the Object() builds and replace
>>>>>> them with both SharedLibrary and StaticLibrary, and get the same error.
>>>>>> For some reason windows/MSYS didn't seem to crash on that configuration.
>>>>>>
>>>>>> On Fri, Apr 24, 2020 at 2:20 PM Luke Robison <lukerobison at gmail.com>
>>>>>> wrote:
>>>>>>
>>>>>>> It seems the problem was that I was making my Library as:
>>>>>>> env.SharedLibrary( src_glob ), but that I had also called env.SharedObject(
>>>>>>> src_glob ).  it seems the SharedLibrary was implicitly building .o files
>>>>>>> and I had requested .os files as well, and probably generating those
>>>>>>> compiled headers (.mod files) twice.  When I changed to calling env.Objects
>>>>>>> rather than env.SharedObjects (or, turns out I wasn't using them anyways,
>>>>>>> get rid of the call entirely) then the build is happy.
>>>>>>>
>>>>>>> Here is minimum to reproduce:
>>>>>>>
>>>>>>> test1.f90
>>>>>>> module mod1
>>>>>>>     integer :: mod1_int
>>>>>>> end module
>>>>>>>
>>>>>>> Sconstruct
>>>>>>> env = Environment()
>>>>>>> src = ['test1.f90']
>>>>>>> env.Object(src)
>>>>>>> env.SharedObject(src)
>>>>>>>
>>>>>>> Execution:
>>>>>>> $ scons
>>>>>>> scons: Reading SConscript files ...
>>>>>>> scons: done reading SConscript files.
>>>>>>> scons: Building targets ...
>>>>>>> gfortran -o test1.os -c test1.f90
>>>>>>> gfortran -o test1.o -c test1.f90
>>>>>>> scons: done building targets.
>>>>>>> AttributeError: 'NoneType' object has no attribute 'get_build_env':
>>>>>>>   File "/usr/lib/python3.7/site-packages/SCons/Script/Main.py", line
>>>>>>> 1381:
>>>>>>>     _exec_main(parser, values)
>>>>>>>   File "/usr/lib/python3.7/site-packages/SCons/Script/Main.py", line
>>>>>>> 1344:
>>>>>>>     _main(parser)
>>>>>>>   File "/usr/lib/python3.7/site-packages/SCons/Script/Main.py", line
>>>>>>> 1119:
>>>>>>>     nodes = _build_targets(fs, options, targets, target_top)
>>>>>>>   File "/usr/lib/python3.7/site-packages/SCons/Script/Main.py", line
>>>>>>> 1318:
>>>>>>>     jobs.run(postfunc = jobs_postfunc)
>>>>>>>   File "/usr/lib/python3.7/site-packages/SCons/Job.py", line 111:
>>>>>>>     self.job.start()
>>>>>>>   File "/usr/lib/python3.7/site-packages/SCons/Job.py", line 216:
>>>>>>>     task.executed()
>>>>>>>   File "/usr/lib/python3.7/site-packages/SCons/Script/Main.py", line
>>>>>>> 256:
>>>>>>>     SCons.Taskmaster.OutOfDateTask.executed(self)
>>>>>>>   File "/usr/lib/python3.7/site-packages/SCons/Taskmaster.py", line
>>>>>>> 312:
>>>>>>>     t.push_to_cache()
>>>>>>>   File "/usr/lib/python3.7/site-packages/SCons/Node/FS.py", line
>>>>>>> 2967:
>>>>>>>     self.get_build_env().get_CacheDir().push(self)
>>>>>>>   File "/usr/lib/python3.7/site-packages/SCons/Node/__init__.py",
>>>>>>> line 654:
>>>>>>>     result = self.get_executor().get_build_env()
>>>>>>>
>>>>>>> On Fri, Apr 24, 2020 at 11:59 AM Bill Deegan <
>>>>>>> bill at baddogconsulting.com> wrote:
>>>>>>>
>>>>>>>> Can you repro with a small example?
>>>>>>>> --taskmastertrace=trace.log can provide some useful information.
>>>>>>>>
>>>>>>>> You could wrap the line 624 in a try except and have it pop into
>>>>>>>> the debugger in the except.
>>>>>>>>
>>>>>>>> On Fri, Apr 24, 2020 at 9:31 AM Luke Robison <lukerobison at gmail.com>
>>>>>>>> wrote:
>>>>>>>>
>>>>>>>>> I'm trying to track down a crash I have triggered in scons (both
>>>>>>>>> 3.0.1 and 3.1.2).  This crash only happens when I ask scons to build my
>>>>>>>>> entire output directory (it will happily compile if my command-line target
>>>>>>>>> is build-variant/my_prog.exe, but giving it the target build-variant causes
>>>>>>>>> the crash).
>>>>>>>>>
>>>>>>>>> AttributeError: 'NoneType' object has no attribute 'get_build_env':
>>>>>>>>>   File "/usr/lib/scons/SCons/Script/Main.py", line 1376:
>>>>>>>>>     _exec_main(parser, values)
>>>>>>>>>   File "/usr/lib/scons/SCons/Script/Main.py", line 1339:
>>>>>>>>>     _main(parser)
>>>>>>>>>   File "/usr/lib/scons/SCons/Script/Main.py", line 1103:
>>>>>>>>>     nodes = _build_targets(fs, options, targets, target_top)
>>>>>>>>>   File "/usr/lib/scons/SCons/Script/Main.py", line 1313:
>>>>>>>>>     jobs.run(postfunc = jobs_postfunc)
>>>>>>>>>   File "/usr/lib/scons/SCons/Job.py", line 111:
>>>>>>>>>     self.job.start()
>>>>>>>>>   File "/usr/lib/scons/SCons/Job.py", line 229:
>>>>>>>>>     task.executed()
>>>>>>>>>   File "/usr/lib/scons/SCons/Script/Main.py", line 237:
>>>>>>>>>     SCons.Taskmaster.OutOfDateTask.executed(self)
>>>>>>>>>   File "/usr/lib/scons/SCons/Taskmaster.py", line 312:
>>>>>>>>>     t.push_to_cache()
>>>>>>>>>   File "/usr/lib/scons/SCons/Node/FS.py", line 2910:
>>>>>>>>>     self.get_build_env().get_CacheDir().push(self)
>>>>>>>>>   File "/usr/lib/scons/SCons/Node/__init__.py", line 624:
>>>>>>>>>     result = self.get_executor().get_build_env()
>>>>>>>>>
>>>>>>>>> I have two questions:
>>>>>>>>> 1) How can I go about debugging this?  I found --debug=pdb to be
>>>>>>>>> usless here, as I end up in Main.py calling _build_targets rather than
>>>>>>>>> where the error occurs.  Similarly, I can't do --tree, since python crashes
>>>>>>>>> before it gets to that point.  Modifying the scons source with print
>>>>>>>>> statements has been my way forward for now.
>>>>>>>>> 2) Is it considered a bug to have scons crash like this (even if I
>>>>>>>>> may have fed it something bad?)
>>>>>>>>>
>>>>>>>>> After adding some prints to the scons code, It seems one of my
>>>>>>>>> compiled header outputs (code is Fortran, so the compiled header is
>>>>>>>>> foo.mod) is the "self" here, and self.get_executor() returns "None", and
>>>>>>>>> then python fails on trying to call None.get_build_env().  This crash
>>>>>>>>> happens immediately after building the relevant SharedObject file (foo.os)
>>>>>>>>> from the source (foo.f90).  It doesn't happen for all files, but it happens
>>>>>>>>> predictably on some, and those tend to have no dependencies themselves.
>>>>>>>>> Re-running the build enough times will eventually let it complete.
>>>>>>>>>
>>>>>>>>> Any suggestions greatly appreciated.
>>>>>>>>>
>>>>>>>>> Luke
>>>>>>>>> _______________________________________________
>>>>>>>>> Scons-users mailing list
>>>>>>>>> Scons-users at scons.org
>>>>>>>>> https://pairlist4.pair.net/mailman/listinfo/scons-users
>>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> Scons-users mailing list
>>>>>>>> Scons-users at scons.org
>>>>>>>> https://pairlist4.pair.net/mailman/listinfo/scons-users
>>>>>>>>
>>>>>>> _______________________________________________
>>>>>> Scons-users mailing list
>>>>>> Scons-users at scons.org
>>>>>> https://pairlist4.pair.net/mailman/listinfo/scons-users
>>>>>>
>>>>> _______________________________________________
>>>> Scons-users mailing list
>>>> Scons-users at scons.org
>>>> https://pairlist4.pair.net/mailman/listinfo/scons-users
>>>>
>>> _______________________________________________
>>> Scons-users mailing list
>>> Scons-users at scons.org
>>> https://pairlist4.pair.net/mailman/listinfo/scons-users
>>>
>> _______________________________________________
>> Scons-users mailing list
>> Scons-users at scons.org
>> https://pairlist4.pair.net/mailman/listinfo/scons-users
>>
> _______________________________________________
> Scons-users mailing list
> Scons-users at scons.org
> https://pairlist4.pair.net/mailman/listinfo/scons-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://pairlist4.pair.net/pipermail/scons-users/attachments/20200425/5ceac240/attachment-0001.html>


More information about the Scons-users mailing list