textstat_dist() is faling with dfm with empty rows

[koheiw]

require(quanteda)
require(readtext)

data_twitter <- readtext("/home/kohei/packages/quanteda.tutorials/content/data/twitter.json", 
                         source = "twitter")

corp_tweets <- corpus(data_twitter)
dfmat_tweets <- dfm(corp_tweets,
                    remove_punct = TRUE, remove_url = TRUE,
                    remove = c('*.tt', '*.uk', '*.com', 'rt', '#*', '@*')) %>% 
    dfm_remove(stopwords('en'))

dfmat_users <- dfm_group(dfmat_tweets, groups = 'screen_name')

sum(rowSums(dfmat_users_prop) == 0)
# [1] 200

dfmat_users_prop <- dfmat_users %>% 
    dfm_select(min_nchar = 2) %>% 
    dfm_trim(min_termfreq = 10) %>% 
    dfm_weight('prop')

textstat_dist(dfmat_users_prop)
# Error in rep_len(value, lenRepl) : attempt to replicate non-vector

Because of this bug, we cannot build quanteda.tutorials.io

[kbenoit]

What's the dim here of dfmat_users_prop?

[koheiw]

> dim(dfmat_users_prop)
[1] 5061  942

[kbenoit]

So the problem is an NA in the dfm? I can't understand the problem without the source data, but then I'm sure you are on top of this.

[koheiw]

data_twitter <- readtext("content/data/twitter.json", source = "twitter")

[koheiw]

This block is failing

quanteda/R/textstat_simil.R

Lines 574 to 577 in f72848a

	if (any(na1))
	result[na1, , drop = FALSE] <- NA
	if (any(na2))
	result[, na2, drop = FALSE] <- NA