On 13-08-30 7:22 AM, Оботуров Артем wrote:
This essentially locks up the version of each package at the time of
writing an adapter. This makes you to make review of changes before
introducing an updated version of those packages into the system.
Yes, that's why I said this is hard.
Also there could a register of converters written by package developers
to cope with diversity of types.
Converters aren't sufficient. For example, many functions that take
input from files accept a filename as a character string or a connection
object for the input parameter. The character string is converted to a
connection using file() (sometimes "" is converted to stdin() or
stdout()), but often other things are done as well, e.g. code to close
the connection at the end of the function call.
For example, here is the code that examines the "con" argument to the
readLines() function:
if (is.character(con)) {
con <- file(con, "r")
on.exit(close(con))
}
The cat() function is a little more complicated, and the parse()
function has much more extensive calculations involving its "file"
argument, because it needs to set debugging information and produce
diagnostic messages.
Duncan Murdoch
On Friday, August 30, 2013, Duncan Murdoch wrote:
On 13-08-30 5:19 AM, Оботуров Артем wrote:
Hello.
One of my clients asked if it would be possible to have an IDE
which could
use type discovery for R functions to make flow-like
construction of an R
program. Something like a LabView graphical composition of
processing
elements.
They called this type of program composition a "workflow".
I looked at some of this programs, like:
* Orange http://orange.biolab.si/
* RedR http://www.red-r.org/
* Rattle http://rattle.togaware.com/
* Rpad https://code.google.com/p/__rpad/
<https://code.google.com/p/rpad/>
and all of them did type introspection (they made mapping between R
functions and their IDE's visual framework) by hand for each
function of
supported packages which is time and resource consuming.
So, to reduce an amount of code to be written for adapters
between R and
IDE some kind of annotations could be introduced over parameters
and return
values. Those could be optional and will help to discover type
information
for support of dynamic composition of programs.
Do you have any suggestions on the topic?
It's very common for R functions to accept many different types for
the same argument on input, and somewhat common for the type of
output to depend on the inputs, so this looks hard: that's why
those front ends need work by hand.
Duncan Murdoch
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
