Re: TVM packaging
On Fri, 2022-02-18 at 03:29 +0000, Wookey wrote:
> However the doc build described in docs/REDAME.txt is not just a
> simple doxygen run and starts trying to download things off the net
> (see log below). My question is, what is an intersphinx inventory,
> and do we really need one to build the docs? Is anyone familiar with
> what's going on here, or do I just need to investigate?
Looks like a cross-documentation reference extension.
https://www.sphinx-doc.org/en/master/usage/extensions/intersphinx.html
> On the other hand the top-level 'make doc' just runs doxygen
> docs/Doxyfile (and doesn't need any of the sphinx stuff, but does need
> some obsolete stuff in the Doxyfile updating). This works.
> I have no idea how useful this is in comparison to whatever the fancy sphinx-based docs.
Maybe it just wants to reference the sphinx doc of its dependency project.
> I'm not quite sure which doc-build is the 'right' one. Or if both or right for different purposes?
>
I only found one doc-build in the mail and did not figure out
another doc-build to be compared...
But if you mean doxygen vs sphinx ... doxygen is likely for C++,
and sphinx is surely for python.
>
> loading intersphinx inventory from https://docs.python.org/3/objects.inv...
> loading intersphinx inventory from https://numpy.org/doc/stable/objects.inv...
> loading intersphinx inventory from https://docs.scipy.org/doc/scipy/objects.inv...
> loading intersphinx inventory from https://matplotlib.org/objects.inv...
It seems that lots of debian packages has enabled this extension and
shipped their objects.inv file in the corresponding doc packages.
for example
~/Downloads ❯❯❯ apt-file search objects.inv | rg numpy
python-numpy-doc: /usr/share/doc/python-numpy-doc/html/objects.inv
~/Downloads ❯❯❯ apt-file search objects.inv | rg matplotlib
python-matplotlib-doc: /usr/share/doc/python-matplotlib-doc/html/objects.inv
~/Downloads ❯❯❯
>
>
> Assuming we ignore the above complication and take the basic doxygen docs, then I have a working package for
> libtvm0
> libtvm-dev
> tvm-doc
>
> I guess we need a python3-tvm too and it's good enough for now.
We should have a python3-tvm as it's the language interface used by their
official starter examples. C++ does not make any starter example in
the ML/DL fields.
If you look into TensorFlow, PyTorch, etc, you will find all of them
write terrible documentation for their C++ interface. Their first
priority is the documentation for python interface.
If we have to choose one from doxygen and sphinx documentation to
drop ... just drop doxygen. The python doc is much more useful
than C++ documentation generally.
Reply to: