You can currently rig this together using put_object() calls by passing the relevant query args. I'd gladly take a PR for an R-only implementation that simplifies this; I don't want to add any system requirements, though.

I'm just running across this with the same problem. I've taken a look at the API docs for multipart uploads and may be able to hack something together, but it would be awesome if there were a quick example that @leeper could share here (time permitting), and maybe even an example in the help page for put_object() would be nice.

This should be straightforward. Basically, Amazon recommends multipart uploads if file size is larger than 100MB. So, conditional on that, put_object() will either upload an object directly or, alternatively, initialize and complete a multipart upload where:

1. object is split into chunks
2. initialize the upload: http://docs.aws.amazon.com/AmazonS3/latest/API/mpUploadInitiate.html
3. loop over put_object() to upload parts: http://docs.aws.amazon.com/AmazonS3/latest/API/mpUploadUploadPart.html, adding relevant headers
4. complete the multipart upload: http://docs.aws.amazon.com/AmazonS3/latest/API/mpUploadComplete.html (requires a new post_object() function)

Separately, this requires an abort upload procedure: http://docs.aws.amazon.com/AmazonS3/latest/API/mpUploadAbort.html, which will need to be a separate function.

When I try multipart = TRUE I get a HTTP 403 error:

  $ Code                 : chr "SignatureDoesNotMatch"
 $ Message              : chr "The request signature we calculated does not match the signature you provided. Check your key and signing method."

If I have multipart = FALSE it works as usual.
I use the same env-variables and I am in the same R session.

I'm also getting a signature match error. While looking at the source, I saw this call to readBin:

My understanding of readBin is that it only reads the first byte when it's called without an n argument.

> readLines("message.txt")
[1] "abcd"
> rawToChar(readBin("message.txt", raw()))
[1] "a"
> rawToChar(readBin("message.txt", raw(), file.size("message.txt")))
[1] "abcd\n"

My PR attempts to fix this. Thoughts @leeper? After that change, I'm now getting another error, which I haven't debugged yet but it makes sense given the traceback:

10: stop(txt, obj, call. = FALSE)
9: .errorhandler(paste("Argument object must be of type character", 
       "or raw vector if serialize is FALSE"), mode = errormode)
8: digest::digest(request_body, algo = "sha256", serialize = FALSE)
7: tolower(digest::digest(request_body, algo = "sha256", serialize = FALSE))
6: canonical_request(verb = verb, canonical_uri = action, query_args = query_args, 
       canonical_headers = canonical_headers, request_body = request_body)
5: aws.signature::signature_v4_auth(datetime = d_timestamp, region = region, 
       service = "s3", verb = verb, action = action, query_args = query, 
       canonical_headers = canonical_headers, request_body = request_body, 
       key = key, secret = secret, session_token = session_token) at s3HTTP.R#101
4: s3HTTP(verb = "POST", bucket = bucket, path = paste0("/", object), 
       headers = c(headers, list(`Content-Length` = ifelse(is.character(file) && 
           file.exists(file), file.size(file), length(file)))), 
       request_body = file, ...) at put_object.R#156
3: post_object(file = NULL, object = object, bucket = bucket, query = list(uploads = ""), 
       headers = headers, ...) at put_object.R#101
2: aws.s3::put_object(file, object, space, check_region = FALSE, 
       key = spaces_key, secret = spaces_secret, base_url = spaces_base, 
       ...) at spaces_object_put.R#44
1: spaces_object_put("~/large-file.dmg", space = "test-analogsea", 
       multipart = TRUE)

I can look into that some other time.

This should now be fixed in the version on GitHub.

Thanks @leeper!

Can you add this to CRAN version as well?

The new maintainer of this package is @acolum. Please check with them about release schedule.