[R-pkg-devel] Failing Valgrind Passing R CMD check

[Duncan Murdoch]

On 07/03/2022 6:00 p.m., Dirk Eddelbuettel wrote:
> 
> On 6 March 2022 at 15:00, Duncan Murdoch wrote:
> | What does simpleCall.R do?
> 
> I supplied sources so you could have check -- but it's not relevant. It was a
> four-element vector leaking 4 x 8 bytes "by design". But make it a 4 million
> element vector and you leak 32 mb.
> 
> | I took a quick look but didn't spot where (if?) R analyzes the valgrind
> | output.
> 
> That was my point. If R supports valgrind, why does it ignore its findings?
> 

Sorry, I missed the link to the full package, but I did try to answer 
your question: maybe trivial losses are ignored.  Your example would be 
more definitive if it really did lose 32 MB.

It's also possible that only some losses are reported (which would be a 
bug, I think).  Your package had the loss in test code, but not in 
example code.

I'd make the changes myself to see, but I don't have valgrind installed, 
and don't have a Linux system handy right now.

Duncan Murdoch