On 12.05.2020 15:05, Yasuhito FUTATSUKI wrote: > On 2020/05/12 20:14, Daniel Shahaf wrote: >> Branko Čibej wrote on Tue, 12 May 2020 07:48 +0200: >>> On 11.05.2020 16:47, Daniel Shahaf wrote: >>>> Yasuhito FUTATSUKI wrote on Mon, 11 May 2020 17:55 +0900: >>>>> On 2020/05/10 1:39, Daniel Shahaf wrote: >>>>>> A minor issue in current trunk: >>>>>> >>>>>> [[[ >>>>>> % …/configure -q --enable-maintainer-mode --without-swig 'RUBY=none' >>>>>> configure: WARNING: Python.h not found; disabling python swig bindings >>>>>> ]]] >>>>>> >>>>>> That warning should not be printed when --without-swig is passed. >>>>>> >>>>>> It's printed by SVN_FIND_SWIG, which _does_ know whether that flag was >>>>>> passed (it was passed iff «$where = no»), but I'm not sure what the >>>>>> right fix is. >>>>> I think it is not the problem of --without-swig option but the problem >>>>> of usage of PYTHON environment variable, or how to specify the target >>>>> of Python bindings. >>>>> >>>>> Even if users don't want to build Python bindings, they can't set >>>>> 'PYTHON=none', like Ruby or Perl bindings. >>>>> >>>>> On the other hand, they can build Python bindings without SWIG, >>>>> if there exists pre-generated bindings source files. >>>> Makes sense. I guess we should add a --enable-swig-python argument, >>>> then? >>> >>> We can build Perl and Ruby bindings without Swig, too, and there are no >>> extra knobs there. >> On the contrary: «./configure RUBY=none» disables the analogous Ruby >> warning. I understand from Yasuhito that PERL=none does the same for >> Perl. >> >>> The --without-swig option just means "don't use swig", it doesn't >>> mean "don't build bindings." The latter is covered by not running >>> 'make swig-X'. >> Yes, forget about --without-swig; that was my mistake and a red >> herring. The point is that there should be a way to tell configure >> "I don't intend to build swig-py so don't warn me about those >> dependencies being missing", like RUBY=none disables the "Ruby not >> found" warning and --without-apxs disables the "apxs not found" warning. >> >> As Yasuhito said, passing PYTHON=none is not a satisfactory workaround >> since it breaks «make check». Hence, I proposed adding a dedicated >> knob. Do you see a better solution? > One of the solutions is to make an other variable to specifiy > Python path for SWIG Python bindings, like TARGET_PYTHON or more > good name. (I tried it for the purpose to enable build Python 2 and > Python 3 bindings in a same build tree[1], though it used bad names > for this purpose, PYTHON2 and PYTHON3) > > Additionally I think it is nice to add argument like > --with-swig-python(=bindings target python path)|--without-swig-python > --with-swig-ruby(=bindings targe ruby path)|--without-swig-ruby > --with-swig-perl(=bindings target perl path)|--without-swig-perl > > I think symmetricalness of those options/variables are important, > but I have no idea how to resolve of asymmetric of PERL, PYTHON, > and RUBY variables.
Can we just say that it was a mistake to look at the environment (e.g., PERL=none) instead of configure arguments (--without-swig-perl)? I agree that adding the proposed flags to configure is the correct solution. We could also add a warning for usage of PERL=none and RUBY=none, pointing users to the new options instead. -- Brane
