[Bioc-devel] cyclic dependency bug in Rdbi

[Hervé Pagès]

Hi Philip,

How many people are still using Rdbi? It has long been superseded by
DBI. One long standing reason for not moving to DBI was that there
was no support for PostgreSQL in the DBI world for a long time. But
this has changed recently with RPostgreSQL now available on CRAN.
Now the question is how good is RPostgreSQL these days? Is it mature
enough to be worth migrating apps from Rdbi/RdbiPgSQL to
DBI/RPostgreSQL? Is it something you've considered/tried?

Maybe it's time for us to deprecate Rdbi/RdbiPgSQL? Feedback from
other "PostgreSQL within R" users would be appreciated.

Thanks!
H.


On 03/21/2011 03:23 AM, Philip Lijnzaad wrote:
> Dear all,
>
> We use Rdbi often, and I keep coming across the rather inconvenient ''cyclic
> name space dependencies are not supported'' error. I decided to get to the
> bottom of this, and basiscally the problem is that Rdbi's .onLoad() function
> always tries to call dbReconnect(obj), even if the .onLoad() is being called
> recursively. The latter happens if an R data file (e.g. .RData) contains an
> object that has a class which is not yet loaded. Once the object is loaded,
> and then inspected, R loads the class definitions of that object. If the
> object's class depends on Rdbi, then it will of course load Rdbi. But in
> doing so, Rdbi's .onLoad() function will try to dbReconnect(obj). But obj
> cannot be accessed without loading all its class definitions, which leads to
> a recursive call to loadNamespace("Rdbi"), triggering the circular
> dependency.
>
> I see four solutions:
>
>    (1) Never put 'Depends: Rdbi' in a class's DESCRIPTION. (not a solution,
>        IMHO)
>
>    (2) Force people to call library(Rdbi) before load()ing anything (no
>        solution either; people forget that, then bother the local R guru's
>        again)
>
>    (3) Don't call dbReconnect() in Rdbi's .onLoad() (see below)
>
>    (4) Make Rdbi's  .onLoad() function a bit cleverer (this will take work)
>
> I would really appreciate if this error would go away; many different people
> seem to be bitten by it time and again (also many GRASS users seem to suffer
> it).  I much prefer solution 3; I frankly think that blindly reconnecting a
> database connection from a saved object is plain wrong. After all, people
> often save an object to disk exactly because they don't want to have to
> connect to a database. In addition, I think it's hard to make this work
> robustly, as evidenced by the cyclic dependency error. (If needed I have a
> simple worked example reproducing the bug).
>
> I may of course have overlooked something; please let me know if that is the
> case.
>
> I contacted the maintainer, but he hasn't responded yet, but the issue maybe
> requires wider consideration anyway.
>
> Regards,
>                                                                         Philip


-- 
Hervé Pagès

Program in Computational Biology
Division of Public Health Sciences
Fred Hutchinson Cancer Research Center
1100 Fairview Ave. N, M2-B876
P.O. Box 19024
Seattle, WA 98109-1024

E-mail: hpages at fhcrc.org
Phone:  (206) 667-5791
Fax:    (206) 667-1319