Bug#825236: RFS: openfst/1.5.3-1 -- weighted finite-state transducers library

[Giulio Paci <giuliopaci@gmail.com>]

On 31/05/2016 11:52, Jakub Wilk wrote:
> * Giulio Paci <giuliopaci@gmail.com>, 2016-05-31, 09:33:
>> I just checked the building status of openfst and noticed failures on arm architectures and on kfreebsd-amd64.
>>
>> On arm it seems to me that a timeout occurred,
> 
> One suspicious thing I found in armhf build log is:
> 
> /usr/bin/make -C . -j4 -k distclean
> 
> but this machine has only 4GB of RAM. So our parallelism limits don't seem to be enforced.

Indeed I was wondering how much memory was available on that build machine. :-)

> This line
> 
> ifneq "$(wildcard /build/buildd-*/)" ""
> 
> was supposed to detect if the package is being built on a buildd, but buildds do longer use build directories like this. I don't think there's any reliable and future-proof
> way to detect this, so I'd suggest to drop this ifneq (i.e., start limiting parallelism everywhere).

I was thinking about removing this line in any case, as the limit is useful everywhere.


>> on kfreebsd-amd64 it seems a compiler failure.
> 
> I suspect that cc1plus was killed by OOM killer, for the same reason as above.

Probably yes.
The limit we set (~2GB) is not enough when we have to deal with 64bit build machines building the test suite with optimization flags turned on.

So I implemented a workaround that sets different limits for the build phase and the check phase.

Can you review my changes and eventually upload them?

Bests,
Giulio