bug-gsl
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Bug-gsl] bug with gsl_integration


From: Patrick Alken
Subject: Re: [Bug-gsl] bug with gsl_integration
Date: Fri, 19 Oct 2018 08:22:27 +0000
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Thunderbird/52.9.1

Thank you for finding a solution Johannes. I've never worked with
mex/GSL myself.

If the default handler is used, then the program will abort/segfault
anyway, so I don't think it should matter if you are calling from mex or
from a command line.

On 10/18/2018 11:26 PM, Johannes Hendriks wrote:
> Dear sir/madam,
>
> I would like to close this thread as it has been resolved. The problem
> ended up being the combination of the default error handler along with the
> mex compiler. The default error handler would do use 'printf' to output
> error message to screen and abort the function. Unfortunately this would
> cause matlab to crash as output to display requires mexPrintf when using
> mex compiler.
>
> So I recognise this is not a bug in your functions. However, maybe it would
> be good to highlight somewhere that if mex is being used then the default
> error handler needs to be disabled / replaced. I am sure there are quite a
> few academic and research users who might come across this problem.
>
> Regards,
> Johannes
>
> On Wed, Oct 17, 2018 at 1:18 PM Johannes Hendriks <
> address@hidden> wrote:
>
>> Dear sir/madam
>>
>> I have found a possible issue with the gsl_integration_qng
>> and gsl_integration_qag functions. Under particular conditions when these
>> functions fail to meet the specified relative or absolute tolerance,
>> instead of failing gracefully they are causing a seg fault.
>>
>> I am running on osx 10.12.6 and compiling using the matlab mex compiling.
>> I have matlab 2017b, and am compiling with the command
>>
>> mex intTwoK_mex.c -I/usr/local/include/ -lmwlapack -lmwblas -lgsl
>>
>> I have attached a zip file containing a matlab script that will cause the
>> integration routine to fault.
>>
>> Let me know if you require any further infomation.
>>
>> Regards,
>> Johannes Hendriks
>>
>>
>>


reply via email to

[Prev in Thread] Current Thread [Next in Thread]