Salut Ralf,
On 06/11/2016 08:27, Ralf Treinen wrote:
> Salut Mehdi,
>
> On Sat, Nov 05, 2016 at 06:36:06PM +0100, Mehdi Dogguy wrote:
>> Hi Ralf,
>>
>> On Tue, Oct 25, 2016 at 09:08:34AM +0200, Ralf Treinen <trei...@free.fr>
>> wrote:
>>> Hi Chris,
>>>
>>> On Sun, Oct 23, 2016 at 09:00:09AM +0100, Chris Lamb wrote:
>>>
>>>> ocamldsort fails to build from source in unstable/amd64:
>>>
>>> it compiles with ocaml 4.03.0 from experimental.
>>>
>>
>> Did you understood what actually makes it fail? It builds fine in a clean
>> chroot on my machine (dunno by which miracle).
>
> No, I do not what this was due to, but if I remember right I could
> reproduce the bug reported by Chris with ocaml 4.02.
I think I understood what happened:
- The first build failure seen by Chris was due to PIE enabled by default in
GCC. Hence, the relocation error emitted by the linker.
- Some time later, OCaml has been binNMUed and that error got "fixed".
- But, a new error appeared (which we see on buildds) and has to do with the
switch to debhelper compat level 10 which enables parallel builds by default.
I've tested on my machine with -j8, and I was able to reproduce the build
failure as seen on buildds. That's the only issue left and I think my -5
upload will fix it.
Eventually, I think we should convince upstream to change the build-system.
I was able to build a functional ocamldsort by using ocamlbuild:
ocamlbuild -pp camlp4o -package unix main.native
It is way simpler and is not subject to failures as seen using the Makefile.
For now, I've just forced parallel=1 in debian/rules to avoid the FTBFS.
Regards,
--
Mehdi
