2 (related) problems with RODBC in 64 bit Windows

Dear Uwe,
Many thanks for the reply.
On 1, the problem is that RODBC on 32 bit ' interprets'  factors correctly, whereas on 64 bit it gives the error below.  On both systems forcing characters (via colClasses = "character" in read.csv), results in no problems.  I still see this as a problem of implementation on 64 bit.
On 2, many thanks, once I gather the courage to address Prof. Ripley I will send him a recollection of my experience.

Kind regards,
Ivan
On 29 Aug 2012, at 15:08, Uwe Ligges <ligges at statistik.tu-dortmund.de> wrote:

On 24.08.2012 21:53, Ivan Alves wrote:

Hi all,

I am encountering an RODBC problem in R 2.15.1 in windows 64 bit which I do not encountered in the same set up in windows 32 bit (the latest binary version of RODBC in both cases from the same depository gotten by install.packages(?RODBC?), Oracle ODBC client software installed in 64 and 32 bit respectively)

1.      The code looks like


library(RODBC)

credentials <- read.csv("~/credentials.csv", head=T, row.names=1)

db <- odbcConnect(dsn="DSN", uid=credentials["DSN", "username"], pwd=credentials["DSN", "password"], rows_at_time=1024)


on which the odbcConnect call fails with the following error code

Error in nchar(uid) : 'nchar()' requires a character vector

(

credentials are processed correctly and credentials["DSN", "username"] correctly returns ? by the way a factor ?

[1] _username_

Levels: ?

).


When I run the equivalent call with direct arguments


db <- odbcConnect("DSN", uid="_username_", pwd="_password_", rows_at_time=1024)


it works just fine.  Furthermore both work just fine on windows 32 bit, or on both systems when the colClasses = "character" option is used. Is this perhaps a problem with RODBC in 64 bit when dealing with factors that is not a problem in 32 bit?

I think 32-bit and 64-bit behave the same way (but you have not compared exactly), reading

credentials <- read.csv("~/credentials.csv", head=T, row.names=1)

results in factors for username and password that have to be converted to character. It is unrelated to RODBC.

2.      Furthermore (and as reported in http://stackoverflow.com/questions/3407015/querying-oracle-db-from-revolution-r-using-rodbc), there are issues with using sqlQuery without the option believeNRows=FALSE, as RODBC seems to still have issues with signed vs. unsigned integer (or sizeof(long) between 32 and 64 bit.

Don't know, but that is something you may want to report (preferrably including patches) to the package maintainer.

Uwe ligges

Any chance the problems have the same source in RODBC code and could be addressed in the near future (after apparently years of making difficult the transition to 64 bit for work with Oracle servers)? (is there an implicit encouragement to use RJDBC when combining 64 bit R use and Oracle databases?)

Many thanks in advance for any guidance.

Ivan
     [[alternative HTML version deleted]]