Upgrading 4.5.3 -> 4.6.2.alpha4 fails

[jdemeyer]

This is a problem I discovered when upgrading from sage-4.5.3 to a candidate 4.6.2.alpha4. Probably, the problem has existed for a while, it just "surfaced" or got noticed now.

The problem is that the file sage/algebras/quatalg/quaternion_algebra_element.cpp is not recreated from the corresponding .pyx file when upgrading. This then causes a compile error.

The file sage/algebras/quatalg/quaternion_algebra_element.pyx includes (using Cython's include statement) sage/ext/gmp.pxi and the latter file was changed some Sage versions ago.

The following happens:

$ ./sage -b

----------------------------------------------------------
sage: Building and installing modified Sage library files.


Installing c_lib
scons: `install' is up to date.
=================== setup.py ========================
Updating Cython code....
Time to execute 0 commands: 1.31130218506e-05 seconds
Finished compiling Cython code (time = 0.0120220184326 seconds)
running install
running build
running build_py
running build_ext
building 'sage.algebras.quatalg.quaternion_algebra_element' extension
building 'sage.calculus.riemann' extension
[...]
building 'sage.symbolic.pynac' extension
building 'sage.symbolic.ring' extension
gcc -fno-strict-aliasing -g -O2 -DNDEBUG -g -O3 -Wall
-Wstrict-prototypes -fPIC
-I/mnt/usb1/scratch/jdemeyer/sage-4.5.3-4.6.2.alpha4/local/include/FLINT/ -I/mnt/usb1/scratch/jdemeyer/sage-4.5.3-4.6.2.alpha4/local/include
-I/mnt/usb1/scratch/jdemeyer/sage-4.5.3-4.6.2.alpha4/local/include/csage
-I/mnt/usb1/scratch/jdemeyer/sage-4.5.3-4.6.2.alpha4/devel/sage/sage/ext
-I/mnt/usb1/scratch/jdemeyer/sage-4.5.3-4.6.2.alpha4/local/include/python2.6
-c sage/algebras/quatalg/quaternion_algebra_element.cpp -o
build/temp.linux-x86_64-2.6/sage/algebras/quatalg/quaternion_algebra_element.o
-w
[trouble...]

Depends on #10233

Apply trac_10751_fix.patch

That is, first apply trac_10233_fix_cython_include_path.patch from the dependency and then trac_10751_fix.patch.

[vbraun]

Initial patch

[vbraun]

Here is the initial version of my rewrite. The basic idea is to split the dependency tracking off into the sage library (sage/misc/dependencies.py) so it can be reused, documented, and doctested. And instead of stuffing filenames as strings into dictionaries, classes encapsulating source file names are introduced. Let me know if you have any comments on these design decisions.

I still want to polish the code a bit more, but the current patch fixes Jeroen's bug:

[vbraun@volker-two hg]$ touch sage/ext/gmp.pxi
[vbraun@volker-two hg]$ sage -b

----------------------------------------------------------
sage: Building and installing modified Sage library files.


Installing c_lib
scons: `install' is up to date.
Updating Cython code....
Building sage/algebras/quatalg/quaternion_algebra_element.pyx because it depends on sage/ext/gmp.pxi.
Building sage/combinat/expnums.pyx because it depends on sage/ext/gmp.pxi.
Building sage/ext/multi_modular.pyx because it depends on sage/ext/gmp.pxi.
Building sage/matrix/matrix_cyclo_dense.pyx because it depends on sage/ext/gmp.pxi.
Building sage/matrix/matrix_integer_dense.pyx because it depends on sage/ext/gmp.pxi.
Building sage/matrix/matrix_rational_dense.pyx because it depends on sage/ext/gmp.pxi.
Building sage/matrix/matrix_rational_sparse.pyx because it depends on sage/ext/gmp.pxi.
Building sage/matrix/misc.pyx because it depends on sage/ext/gmp.pxi.
Building sage/modular/modform/eis_series_cython.pyx because it depends on sage/ext/gmp.pxi.
Building sage/quadratic_forms/count_local_2.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/integer.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/integer_ring.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/fast_arith.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/fraction_field_FpT.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/rational.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/padics/padic_base_coercion.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/padics/padic_capped_absolute_element.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/padics/padic_capped_relative_element.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/padics/padic_fixed_mod_element.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/padics/padic_generic_element.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/padics/padic_printing.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/padics/pow_computer.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/padics/pow_computer_ext.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/polynomial/polynomial_integer_dense_flint.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/polynomial/polynomial_rational_flint.pyx because it depends on sage/ext/gmp.pxi.
Building sage/rings/polynomial/real_roots.pyx because it depends on sage/ext/gmp.pxi.
Execute 26 commands (using 4 threads)

[fbissey]

It's big. I guess the split in depency.py may make thing more manageable in the long term. We how to discuss the potential removal of debian stuff on the the list, #5903 doesn't say anything about setup.py and module_list.py (which I was thinking of cleaning).

The only mistake I can see from a first look is on line 529 of dependency.py

sage: gmp = SourceFile('/home/vbraun/opt/sage-4.6.2.alpha4/local/include/gmp.h')

I don't think we want to hardcode your setup :)

[robertwb]

We should switch Sage to use ​http://wiki.cython.org/enhancements/distutils_preprocessing . This would let us get rid of most of the manual module_list.py declarations as well.

[vbraun]

Robert's distutils-preprocessing seems to be the best way to do things, and we should investigate that once the requisite Cython-14.1 is included.

In the meantime, I've made a small patch that fixes the dependency discovery bug (tested by touching sage/ext/gmp.pxi as above). This leaves the setup.py code in its current somewhat messy state, but I think we should go with the minimal bugfix until we can give distutils-preprocessing a try.

Also, enables parallel compile by default. I didn't even know we supported that. You can still disable it via setting the MAKE='make -j1' environment variable. Rebuilding the Sage library dropped from half an hour to 5 minutes ;-)

[vbraun]

Hmm patch bot still lists both patches erroneously.

Apply trac_10751_fix.patch

[jdemeyer]

make -j without argument to -j means: use unlimited number of threads. Therefore, the code

pos = MAKE.rfind('-j') 
if pos>=0: 
    nthreads = int(MAKE[pos+2:].split()[0])

should be changed to

pos = MAKE.rfind(' -j')
if pos >= 0:
    try: 
        nthreads = int(MAKE[pos+3:].split()[0])
    except ValueError:
        # If there is no argument to -j, use maximum number of threads
        nthreads = 2*cpu_count

In any case, the ValueError must be handled somehow.

Note also that I have added a space before -j.

[jdemeyer]

also IndexError should be caught in the same except clause.

[vbraun]

Ok, fixed. Also handling the make -j=N case now.

[jdemeyer]

I think you need to add a try/except as follows:

try:
    if MAKE[pos+3] == '=':   # make -j=N is the same as make -jN 
        pos += 1
except IndexError:
    pass 

[vbraun]

Yep, that if clause needs to be moved into the existing try/except block. Updated patch follows...

[vbraun]

Updated patch

[jdemeyer]

Looks good to me.