This issue tracker has been migrated to GitHub, and is currently read-only.
For more information, see the GitHub FAQs in the Python's Developer Guide.

Author Joshua.J.Cogliati
Recipients Alexander.Sulfrian, Joshua.J.Cogliati, Xuefer.x, dstufft, eric.araujo, jrincayc, ned.deily, ronaldoussoren, tarek
Date 2015-04-02.17:49:28
SpamBayes Score -1.0
Marked as misclassified Yes
Message-id <1427996969.51.0.0201883208751.issue8027@psf.upfronthosting.co.za>
In-reply-to
Content
Let me try and explain what is trying to be done in the original code, 
what the fix was, and then discuss some possible better solutions.

Original code:
if target_lang == "c++" and self.compiler_cxx:
    linker[0] = self.compiler_cxx[0]

Current code:
if target_lang == "c++" and self.compiler_cxx:
    i = 0
    if os.path.basename(linker[0]) == "env":
        i = 1
        while '=' in linker[i]:
            i += 1
    linker[i] = self.compiler_cxx[i]

Basically, what we have, is two variables, linker, and self.compiler_cxx.

By default on my linux system these are:
linker=['gcc', '-pthread', '-shared']
self.compiler_cxx=['g++', '-pthread']

So under the current code, since self.compiler_cxx[0] != "env"
i=0 so
linker[0] = self.compiler_cxx[0]
so linker=['g++', '-pthread', '-shared']

So the goal is to switch the linker to something that works better with c++.  

In the original code:
linker[0] = self.compiler_cxx[0]
improves things if the first element in the linker variable is the linker executable, and the first element in compiler_cxx is a compiler executable, and the compiler executable works better at linking c++ code than the c linker executable.

Next we switch to the current code.  

I am guessing that Ronald had something like this:
linker=['env','BAR=FOO','gcc', '-pthread', '-shared']
self.compiler_cxx=['env','BAR=FOO','g++', '-pthread']
and so with the current code we end up with i=2, and the linker variable ends up with:
linker=['env','BAR=FOO','g++', '-pthread', '-shared']

Now, what is the problem we are encountering with the current code?

Basically, people want to use things like CXX="ccache g++"
So if 
linker=['gcc', '-pthread', '-shared']
self.compiler_cxx=['ccache', 'g++']
we have i=0, 
linker[0]=self.compiler_cxx[0] = 'ccache'
resulting in
linker=['ccache', '-pthread', '-shared']
which will fail, because ccache expects the first argument to be the compiler executable (that is 'g++' not '-pthread')

So, how can we fix this?

If the linker can link c++ code, then the optimal solution is to do nothing, 
as in remove the entire 
if target_lang == "c++" and self.compiler_cxx:
   ...
(The fix-distutils-* patches I have created do this solution)

We could special case ccache:
if target_lang == "c++" and self.compiler_cxx:
    if os.path.basename(self.compiler_cxx[0]) == "ccache":
        linker[0:1] = self.compiler_cxx[0:2]
    elif os.path.basename(linker[0]) == "env":
    ...

We could hope that what we actually want is the entire compiler_cxx:
if target_lang == "c++" and self.compiler_cxx:
    linker[0:1] = self.compiler_cxx[:]


Maybe we could special case c++ a little earlier (since linker_exe is just cc by default):
if target_desc == CCompiler.EXECUTABLE and target_lang == "c++" and self.compiler_cxx:
    linker = self.compiler_cxx[:]
elif target_desc == CCompiler.EXECUTABLE:
    linker = self.linker_exe[:]
else:
    linker = self.linker_so[:]


None of these solutions make me feel awesome.  

The real problem is that we have no way to set the c++ linker.  I don't see any variable listed like this Make's implicit variables section.

A secondary problem is that we string concatenate the LDSHARED and LDFLAGS 
(in the line: ldshared = ldshared + ' ' + os.environ['LDFLAGS'] in sysconfig.py)

Another secondary problem is that we don't look for a CXXFLAGS varible, and so people stuff parameters into CXX instead.

If the two secondary issues did not exist we could reasonably reliably do something like
linker = cxx + ldflags
which might actually work for all the cases that are currently covered and work where cxx=['ccache','g++']

The current code is broken, how do we want to fix it?
History
Date User Action Args
2015-04-02 17:49:29Joshua.J.Cogliatisetrecipients: + Joshua.J.Cogliati, ronaldoussoren, tarek, ned.deily, eric.araujo, jrincayc, Xuefer.x, Alexander.Sulfrian, dstufft
2015-04-02 17:49:29Joshua.J.Cogliatisetmessageid: <1427996969.51.0.0201883208751.issue8027@psf.upfronthosting.co.za>
2015-04-02 17:49:29Joshua.J.Cogliatilinkissue8027 messages
2015-04-02 17:49:28Joshua.J.Cogliaticreate