On 03/22/2012 09:38 AM, Simon Urbanek wrote:
>
> On Mar 22, 2012, at 9:45 AM, Terry Therneau <thern...@mayo.edu
> <mailto:thern...@mayo.edu>> wrote:
>
>>
>>>>> strongly disagree. I'm appalled to see that sentence here.
>>>> >
>>>> > Come on!
>>>> >
>>>>> >> The overhead is significant for any large vector and it is in
>>>>> >> particular unnecessary since in .C you have to allocate*and copy*
>>>>> >> space even for results (twice!). Also it is very error-prone, because
>>>>> >> you have no information about the length of vectors so it's easy to
>>>>> >> run out of bounds and there is no way to check. IMHO .C should not be
>>>>> >> used for any code written in this century (the only exception may be
>>>>> >> if you are passing no data, e.g. if all you do is to pass a flag and
>>>>> >> expect no result, you can get away with it even if it is more
>>>>> >> dangerous). It is a legacy interface that dates way back and is
>>>>> >> essentially just re-named .Fortran interface. Again, I would strongly
>>>>> >> recommend the use of .Call in any recent code because it is safer and
>>>>> >> more efficient (if you don't care about either attribute, well, feel
>>>>> >> free ;)).
>>>> >
>>>> > So aleph will not support the .C interface? ;-)
>>>> >
>>> It will look at the timestamp of the source file and delete the package if
>>> it is not before 1980 ;). Otherwise it will send a request for punch cards
>>> with ".C is deprecated, please upgrade to .Call" stamped out :P At that
>>> point I'll be flaming about using the native Aleph interface and not the R
>>> compatibility layer ;)
>>>
>>> Cheers,
>>> S
>> I'll dissent -- I don't think .C is inherently any more dangerous
>> than .Call and prefer it's simplicity in many cases. Calling C at
>> all is what is inherently dangerous -- I can reference beyond the end
>> of a vector, write over objects that should be read only, and branch
>> to random places using either interface.
>
> You can always do so deliberately, but with .C you have no way of
> preventing it since you don't even know what is the length! That is
> certainly far more dangerous than .Call where you can simply loop over
> the length, check that the lengths are compatible etc. Also for types
> like strings .C is a minefield that is hard to not blow up whereas
> .Call it is even more safe than scalar arrays. You can do none of that
> with .C which relies entirely on conventions with no recorded semantics.
>
I've overrun arrays in both .C and .Call routines, and I assure you that
it was never deliberate. Very effective at crashing R with strange
error messages though.
I will have .C("somefun", as.integer(length(x)), x), the .Call version
will skip the second argument and add a line in the C code; no real
difference from my point of view. ( Though the spelling is harder to
remember in .Call. Does R core use dice to decide which things are
upper, lower, and mixed case: LENGTH, asInteger, ncols?) R strings are
a PITA in C and I mostly avoid them so have no arguments about C vs Call
there.
Much of the survival library is .C for historical reasons of course, but
I think it shows that you can be safe in C; though you can't be sloppy.
>
>> If you are dealing with large objects and worry about memory
>> efficiency then .Call puts more tools at your disposal and is worth
>> the effort. However, I did not find the .Call interface at all easy
>> to use at first
>
> I guess this depends on the developer and is certainly a factor.
> Personally, I find the subset of the R API needed for .Call fairly
> small and intuitive (in particular when you are just writing a safer
> replacement for .C), but I'm obviously biased. Maybe in a separate
> thread we could discuss this - I'd be happy to write a ref card or
> cheat sheet if I find out what people find challenging on .Call.
> Nonetheless, my point is that it is more than worth investing the
> effort both in safety and performance.
>
I'm giving a short course at UseR on the design of the survival packages
which promises a document "containing all the details", currently being
written. It has examples of .Call with discussion of what each action
is doing. The final result will certainly be added to the survival
package; hopefully it will be useful enough to earn a place on the CRAN
documentation page as well.
>
>> and we should keep that in mind before getting too pompous in our
>> lectures to the "sinners of .C". (Mostly because the things I needed
>> to know are scattered about in multiple places.)
>>
>> I might have to ask for an exemption on that timestamp -- the first
>> bits of the survival package only reach back to 1986. And I've had
>> to change source code systems multiple times which plays hob with the
>> file times, though I did try to preserve the changelog history to
>> forstall some future litigious soul who claims they wrote it first
>> (sccs -> rcs -> cvs -> svn -> mercurial). :-)
>>
>
> ;) Maybe the rule should be based on the date of the first appearance
> of the package, fair enough :)
>
> Cheers,
> Simon
I think (but not sure) that the first version to get outside our local
group was around 1989.
Terry
[[alternative HTML version deleted]]
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
