Make sure that dfm_* are robust against empty dfm

[koheiw]

nulldfm <- quanteda:::make_null_dfm()

# These look fine
dfm_select(nulldfm)
dfm_select(nulldfm, 'a')
dfm_trim(nulldfm)
dfm_weight(nulldfm)
dfm_sample(nulldfm)
dfm_subset(nulldfm)

# These are not
dfm_lookup(nulldfm, dictionary(list(A = 'a')))
dfm_group(nulldfm)
dfm_compress(nulldfm)

[kbenoit]

Remember the grammar rules: dfm_* should only accept inputs as dfms, and output a dfm (never NULL).

Maybe that’s what this issue is designed to enforce, but I cannot check since in a meeting and currently on ipad only!

[koheiw]

It is still a dfm, but no features or documents.

> nulldfm
Document-feature matrix of: 0 documents, 0 features.
0 x 0 sparse Matrix of class "dfm"
<0 x 0 matrix>