Retain attributes when combining cran_deps + remote_deps

[jennybc]

cran_deps bears the type attribute

This fixes the problem I described in slack.

The order of arguments in rbind() affects the attributes of the resulting data frame.

a <- data.frame(a = 1)
attr(a, "yo") <- "hi"

b <- data.frame(a = 2)

attr(rbind(a, b), "yo")
#> [1] "hi"
attr(rbind(b, a), "yo")
#> NULL

^{Created on 2019-02-19 by the reprex package (v0.2.1.9000)}

I don't know why other people aren't seeing this, so perhaps there is more going on? But this fixes things for me.

[jennybc]

This PR passed tests (with 1 skipped test) everywhere but R-devel on travis, so I think that difficulty is orthogonal to this.

[jennybc]

Maybe getting closer to the root cause ...

I think this row indexing is where the type attribute can potentially be lost from either the cran or gh deps.

remotes/R/deps.R

Lines 148 to 152 in 6212345

	# Only keep the remotes that are specified in the cran_deps or are NA
	remote_deps <- remote_deps[is.na(remote_deps$package) | remote_deps$package %in% cran_deps$package, ]
	# If there are remote deps remove the equivalent CRAN deps
	cran_deps <- cran_deps[!(cran_deps$package %in% remote_deps$package), ]

df <- data.frame(a = 1:3)
attr(df, "yo") <- "hi"
attr(df, "yo")
#> [1] "hi"
attr(df[1, ], "yo")
#> NULL

^{Created on 2019-02-20 by the reprex package (v0.2.1.9000)}

So my fix of inverting the order inside rbind() isn't the general solution. I think it just works out better in the examples I'm currently facing.

[jennybc]

Closing in favour of #304