Need methods package on the search path

[krlmlr]

> Rscript -e "dplyr::memdb_frame(a=1)"
Error in UseMethod("db_has_table") : 
  no applicable method for 'db_has_table' applied to an object of class "SQLiteConnection"
Calls: <Anonymous> ... copy_to -> copy_to.src_sql -> isTRUE -> db_has_table
5: db_has_table(dest$con, name)
4: isTRUE(db_has_table(dest$con, name)) at tbl-sql.r#334
3: copy_to.src_sql(src_memdb(), data_frame(...), name = .name) at copy-to.r#14
2: copy_to(src_memdb(), data_frame(...), name = .name) at src-sqlite.r#135
1: dplyr::memdb_frame(a = 1)
> Rscript -e "library(methods); dplyr::memdb_frame(a=1)"
Source:   query [?? x 1]
Database: sqlite 3.11.1 [:memory:]

      a
  <dbl>
1     1

Importing all symbols from methods in the NAMESPACE does not seem to change this behavior.

[hadley]

This came up with another package I was debugging in person - I could've sworn this used to work. Is it a regression in R?

[krlmlr]

I'll bisect R later today.

[krlmlr]

Looks like this particular problem is not a regression with R. I tried trunk1000 (mid-August last year), with the same symptoms. Now trying trunk3000 (from 2014), will report here if anything changes.

This will be a lot of fun to track down :-o

I can replicate in a similar setting when implementing an S3 method for class "Matrix"; it is also not dispatched for Matrix::Matrix() (which has class "lsyMatrix"):

#' @export
test <- function(x) UseMethod("test", x)

#' @export
test.Matrix <- function(x) "Hi!"

[kevinushey]

Why not just add methods to the Depends: field?

[krlmlr]

Will this fix the example where dplyr is loaded not attached?

[kevinushey]

Ah, you're right :/ It would need to be more aggressive, e.g. call library(methods) in .onLoad().

FWIW, Rscript is weird compared to e.g. R -e in that it explicitly does not attach methods, but I'm guessing you've bumped into this before as well?

[krlmlr]

Yes, even asked on SO a couple of years ago ;-)

A workaround would be to up the affected methods to S4 generics. Still, I'm curious why the current code doesn't work, even when all of "methods" is imported into the NAMESPACE.

[krlmlr]

After adding this for db_has_table() and importing "methods", I'm getting an error about another method:

#' @export
setGeneric("db_has_table",
           def = function(con, ...) standardGeneric("db_has_table")
)

#' @export
setMethod("db_has_table", signature("DBIConnection"),
          function(con, table) dbExistsTable(con, table))

In the C code for UseMethod(), there are too many uses of R_GlobalEnv for my taste, but I haven't investigated further. (Oops: R_GlobalContext.)

[krlmlr]

It works flawlessly if methods::extends is copied into the global environment:

Rscript -e "extends <- methods::extends; dplyr::memdb_frame(a = 1)"

But it also works if extends is assigned an arbitrary value:

Rscript -e "extends <- 42; dplyr::memdb_frame(a = 1)"

To me, it looks like (at least) this code in R is responsible for this behavior. Weird...

[kevinushey]

It sounds like it's worth reporting to R-devel, or filing a bug -- likely that sanity check could just check that methods is within the set of loaded namespaces?

[krlmlr]

Done: https://stat.ethz.ch/pipermail/r-devel/2016-April/072592.html

[krlmlr]

Should be fixed upstream eventually.