Adds support for .rds files with readRDS function

[jsams]

I think this would more-or-less close #22 though it doesn't integrate with the load functionality

Note that the tests are ever so slightly different. This is on purpose. rds files do not name the objects they contain. I converted all the rda test files into named lists, and saved them as rds files. So, while the list itself does not contain a name, it contains objects which do have names, which is what you are accessing after conversion into the julia type. With convert=false, a bare R type is returned.

[alyst]

Great, thanks for your contribution! I think we would need to make one more iteration and change readRDS into load (+ necessary changes to FileIO) to ensure API uniformity.

[alyst]

Have you tried to add RDS format to FileIO.jl? It should be straightforward and almost identical to what is there for RData (with the exception of "RDSX" header line). So readRDS() would change into load(s::File{format"RDataSingle"}/Stream{format"RDataSingle"}, ...) pair of methods.

[jsams]

I think I've gotten things in the state you wished: factored out decompress, reduced the test set for the rds to just a data.frame but in binary compressed and decompressed and ascii formats, and duplicated the load interface and removed readRDS.

This all seems to work for me locally with local changes to FileIO to register the format. I couldn't figure out how to get the format to register in the RData package itself to make testing easier before the FileIO pull.

On that note...

If you know anything about the file formats, can you take a look at the detect_rdata_single function I'm proposing in my FileIO branch before requesting the pull? I don't know anything about the formats, and it seems that when the RDSX header line is removed, not much is left before the file format itself begins.

[alyst]

Thanks, I think we're almost there!

I couldn't figure out how to get the format to register in the RData package itself to make testing easier before the FileIO pull.

AFAIR, you can do FileIO.add_format() to add the format to the FileIO registry externally, but I'm not sure it really works since it happens after the package initialization.
Anyway, adding RDS support directly to FileIO is the right thing to do.
Your FileIO branch looks ok to me. Do you know whether ".rds" is the only extension or other alternatives (".RDS") also exist? If so, I would list them to add_format() call too.
For the FileIO PR to be merged you also need to add the test (see test/query.jl for how it was done for RData). As long as detect_rdata_single() works for the simple test file, that function is fine.

[alyst]

Could you please also mention this PR in NEWS.md?

[alyst]

This looks like a duplication of the code from load(format{"RData"}). I think it's a good idea to extract it into a separate function, especially since we are going to extend the supported compression formats. E.g.

# check whether the stream is compressed (Gzip) and return the decompressed stream
function decompress(io::IO)
    gzipped = read(io, UInt8) == 0x1F && read(io, UInt8) == 0x8B # check GZip magic number
    seekstart(io)
    if gzipped
        return GzipDecompressorStream(io) # if compressed, transcode gzipped stream
    else
        return io
    end
end

It should be called then within try block: io = decompress(io)

[alyst]

I don't think it's necessary to have the equivalents of RDA test files as RDS: the serialization format is the same and the code that reads these objects is also the same. So these tests would just test the same code with the same data.
I think it should be enough to test some moderately complex object (say, data.frame), but check that both binary (compressed and uncompressed) and ASCII versions of RDS format are supported.

[alyst]

I guess we need to test both convert=false and convert=true (the default) modes. convert=true should return DataFrame directly

[jsams]

Ok, I've made a PR in FileIO, updated the news (though it's not clear how you see the versions moving forward, please modify to your taste), and had the tests also test the default convert=true condition.

[alyst]

Sorry, missed that: I think before calling eltypes() or comparing with df it makes sense to do @test rdf isa DataFrame (for the other load() as well). If load goes completely wrong, it would help to diagnose it properly, otherwise @test eltype... will fail with MethodError.

[alyst]

@jsams Thanks! Let's see what happens with FileIO PR. I will merge once that one is merged. Maybe by that time there would be some progress with POSIXct, then we can tag 0.3.0 with both these features.

[alyst]

Your PR to FileIO is merged and is available in v0.6.0.
So after updating REQUIRES/NEWS.md and rebasing we should be ready to go.

[jsams]

I've brought this in line with the latest master. runtests throws some errors in the master, but my additions don't add any errors and all of my tests pass. Any reason to not pull?

[coveralls]

Coverage increased (+0.9%) to 67.708% when pulling a5bb2b5 on jsams:rds into 4b3b3e8 on JuliaData:master.