kwic does not return data.frame when there are no matches, but should.

[conjugateprior]

In quanteda 0.9.7-20:

k1 <- kwic(inaugCorpus, "brexit")
class(k1) ## logical
k1 ## NA
k2 <- kwic(inaugCorpus, "independence")
class(k2) ## kwic data.frame

This breaks some natural aggregation procedures, such as

res <- list()
for (i in 1:N)
  res[[ i ]] <- kwic(subcorpus[i], "independence")
do.call(rbind, res) ## stack kwics 

One workaround is to check is.data.frame(k1) before doing anything (not to check is.na(k1) which will trigger warnings when evaluated on a kwic with matches).

As implied by the help page, kwic should return an empty data.frame, not NA.

┆Issue is synchronized with this Asana task

[kbenoit]

Agreed in general that this is odd behaviour, but it was designed that way to be consistent with other, similar R function, e.g.:

> grep("exit", inaugTexts)
[1] 17 23
> grep("brexit", inaugTexts)
integer(0)
> match("brexit", c(LETTERS, "brexit"))
[1] 27
> match("brexit", LETTERS)
[1] NA
> stringi::stri_match_all_regex(inaugTexts[1], "union")
[[1]]
     [,1]   
[1,] "union"
[2,] "union"
> stringi::stri_match_all_regex(inaugTexts[1], "brexit")
[[1]]
     [,1]
[1,] NA  

[conjugateprior]

stringi is an odd one to make the argument because it behaves like I think kwic should, i.e. gives you the data structure you're expecting, even if it's empty.

class(stringi::stri_match_all_regex(inaugTexts[1], "brexit"))
[1] "list"

Also, base R is not consistent on this theme, e.g.

class(Filter(is.numeric, strsplit(LETTERS, '')))
[1] "list"

Maybe they feel you're allowed to do this with lists but not with other things. (Or maybe that's the kind of generalization from two cases that never works in a language organized like the Celestial Emporium of Benevolent Knowledge...)

[kbenoit]

Fair enough - what should kwic(inaugCorpus, "brexit") return then?

[conjugateprior]

Ideally, a completely empty kwic object. That would behave correctly in any downstream processing.
A data.frame would be enough to make rbind, select, indexing etc. work:

k1 <- data.frame(docname=as.factor(character()), position=numeric(), 
                 contextPre=character(), keyword=character(), contextPost=character(), 
                 stringsAsFactors=FALSE)
k1
[1] docname     position    contextPre  keyword     contextPost
<0 rows> (or 0-length row.names)
class(k1)
data.frame

but it might be sensible to add the attributes. Depends on whether and what you plan to do with them elsewhere, I guess.

[kbenoit]

Note that this already works:

> rbind(NA, kwic(inaugTexts[1], "union"))
                                                contextPre keyword                     contextPost
               [NA, NA]                               NA [    NA                              ] NA
 [1789-Washington, 964] course of nature an indissoluble [   union ] between virtue and happiness;
[1789-Washington, 1502]        for the security of their [   union ] and the advancement of their 
> rbind(kwic(inaugTexts, "Brexit"), kwic(inaugTexts[1], "union"))
                                                contextPre keyword                     contextPost
               [NA, NA]                               NA [    NA                              ] NA
 [1789-Washington, 964] course of nature an indissoluble [   union ] between virtue and happiness;
[1789-Washington, 1502]        for the security of their [   union ] and the advancement of their 

[kbenoit]

Fixed. Now returns an empty data.frame. The print.kwic() displays a NULL but the object is actually an empty data.frame.

testkwic <- kwic(paste(LETTERS, collapse=' '), 'É')
testkwic
# NULL
str(testkwic)
# Classes ‘kwic’ and 'data.frame':	0 obs. of  0 variables
#  - attr(*, "valuetype")= chr "glob"
#  - attr(*, "ntoken")= Named int 26
#   ..- attr(*, "names")= chr "text1"
#  - attr(*, "keywords")= chr "É"
#  - attr(*, "tokenize_opts")= list()