[geos-devel] [GEOS] #767: "definitely lost" memory leak in rgeos_Init (rgeos.c:57)

[Roger Bivand]

On Fri, 5 Feb 2016, Sandro Santilli wrote:

> Thanks Roger, could you provide an URL to the "rgeos"
> bug tracker as a reference ? Having this rationale as a comment
> on trac would also be nice. TIA
>
> https://trac.osgeo.org/geos/ticket/767

There is no bug tracker. General practice for all contributed R packages 
(almost 8000) is to email the maintainer, whose email is in the package, 
and on the archive network package page.

> packageDescription("rgeos")$Maintainer
[1] "Roger Bivand <Roger.Bivand at nhh.no>"

This is typically what happens, but in the R-spatial community, it is 
equally OK to post on R-sig-geo:

https://stat.ethz.ch/mailman/listinfo/r-sig-geo

which may draw in other comments about user misunderstandings (being an 
interpreted language, these do happen).

The rgeos development repo is at:

https://r-forge.r-project.org/projects/rgeos/

but the "lists" there are never used, as the developers all follow 
R-sig-geo closely. Questioners may be asked to check attempts to fix 
issues by installing from source from the repo; rgeos fails to build on 
R-Forge because no GEOS binaries are available there so direct installs 
from R-Forge do not work.

In summary - post on R-sig-geo after subscription, and follow the posting 
guide with a reproducible example.

In this case as far as I can see, there is no link between the real issue 
(use of rgeos through another R package under Windows on CP1252 apparently 
switching locale), and the possible valgrind false positive.

valgrind may not be "seeing" all of the behaviour of the R garbage 
collector; the 40 bytes are probably the GEOSContextHandle_t r stored as 
an R_ExternalPtr and under the control of the R garbage collector 
(alternatively something in the errorHandler or the warningHandler). The 
handle is properly registered with a finalizer, so unloading the package 
on R quit will free the pointer when the GC gets to it.

Thanks for taking an interest in an issue which appears to point somewhere 
else - I've asked for a fully reproducible example to see where the locale 
switching kicks in.

Roger

>
> --strk;
>
> On Thu, Feb 04, 2016 at 11:30:23PM +0100, Roger Bivand wrote:
>>>
>>> I would like to file a bug spotted via Valgrind while loading the R
>>> rgeos package. Below are the steps to reproduce it.
>>>
>>
>> This is obviously not a GEOS issue, but should have been sent to the
>> maintainer of the rgeos package, that is to me. I have resolved the ticket
>> as invalid.
>>
>> The identity of "philae" is unknown and untraceable - please contact
>> me directly including the test script, which wasn't attached to the
>> ticket.
>>
>> A minor leakage of 40 bytes isn't a big deal, and its origin isn't clear.
>>
>> Roger
>>
>> --
>> Roger Bivand
>> Department of Economics, Norwegian School of Economics,
>> Helleveien 30, N-5045 Bergen, Norway.
>> voice: +47 55 95 93 55; fax +47 55 95 91 00
>> e-mail: Roger.Bivand at nhh.no
>> http://orcid.org/0000-0003-2392-6140
>> https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en
>> http://depsy.org/person/434412
>>
>>
>> _______________________________________________
>> geos-devel mailing list
>> geos-devel at lists.osgeo.org
>> http://lists.osgeo.org/mailman/listinfo/geos-devel
>
>

-- 
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; fax +47 55 95 91 00
e-mail: Roger.Bivand at nhh.no
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en
http://depsy.org/person/434412