Bug#972097: Fwd: [conradsand at gmail.com: mlpack stuck in debian unstable]

[Barak A. Pearlmutter]

Package: mlpack
Version: 3.4.1-1

I'm forwarding this to the Debian bug tracking system, it will be
allocated a number which we should CC on further correspondence about
the issue. For posterity etc.

---------- Forwarded message ---------
From: Ryan Curtin <ryan at ratml.org>
Date: Mon, 12 Oct 2020 at 16:19
Subject: Re: [conradsand at gmail.com: mlpack stuck in debian unstable]
To: Barak A. Pearlmutter <barak at pearlmutter.net>
Cc: <conradsand at gmail.com>


On Mon, Oct 12, 2020 at 04:01:21PM +0100, Barak A. Pearlmutter wrote:
> The big issue is compilation issues on various architectures, rather
> than the Julia bindings.
> Have a look,
>
> - https://tracker.debian.org/pkg/mlpack
> - https://buildd.debian.org/status/package.php?p=mlpack
>
> Help welcome, absolutely!

Awesome, thanks for the links.  I can try 'debugging-from-afar' for some
of these.  Please do feel free to open bugs in the mlpack repository for
this, just like the ensmallen repository.  I actually had no idea there
were problems here.

Anyway, quick hits:

 - armel: maybe add `-latomic` to CMAKE_EXE_LINKER_FLAGS?  e.g. add this
       to the CMake configuration command:
       "-DCMAKE_EXE_LINKER_FLAGS='-latomic'"

 - mips64el: related to this bug?
       https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=965247
       You could also disable LTO by adding -fno-lto to CMAKE_CXX_FLAGS
       or CMAKE_LINKER_FLAGS.  I'm not sure if that will fix it.

 - mipsel: "LLVM ERROR: out of memory"; is there anything we can do
       here?  `make -j1`, maybe remove '-pipe' from CMAKE_CXX_FLAGS (if
       it's set) and set --param ggc-min-heapsize=32768 --param
       ggc-min-expand=1?  That's what I do for underpowered arm7hl
       builders on Fedora.

 - alpha: ".got subsegment exceeds 64K".  I have no idea what could be
       wrong here.  Can we just disable alpha?  There's very little
       information I can find on this one, other than a suggestion
       somewhere that the linker table sizes need to be increased, plus
       what look like a bunch of debian packages where they disabled
       alpha because of this exact issue.

 - m68k: is the version of Doxygen here different?  A workaround could
       be to regex `WARN_AS_ERROR[ ]*= YES` to `WARN_AS_ERROR = NO` in
       `Doxyfile` before the build starts.

I'm not sure about the dependency installability problems, but in the
worst case, those are all related to Python, so you could for those
architectures just disable the Python bindings.

> Meantime I'll turn off the Julia bindings.

My experience with Debian-packaged Julia actually hasn't been great, but
perhaps it's improved over time.  I think once Julia is available in
stable, perhaps the situation will be better?

--
Ryan Curtin    | "Get off my lawn!"
ryan at ratml.org |   - Kowalski