KWIC crashes when all results are filtered out by positivelist

[ChristophLeonhardt]

library(polmineR)
use("GermaParl")
x <- partition("GERMAPARL", lp = 17, role = c("gov", "mp")) %>%
  kwic(query = '"[Aa]ttack.*"', positivelist = c("Messer"), cqp = TRUE) 

Leads to a legit crash of RStudio.

polmineR_0.7.11.9024

[ChristophLeonhardt]

In addition, in other corpora, it doesn't crash but throws an error:

x <- partition("BB", lp = 5, role = c("gov", "mp")) %>%
  kwic(query = '"[Aa]ttack.*"', positivelist = c("Messer.*", ".*[Mm]ord.*"))

... get encoding: latin1
... get cpos and strucs
... getting corpus positions
... number of hits: 12
... checking that all p-attributes are available
... getting token id for p-attribute: word
... filtering by positivelist
... number of hits droped due to positivelist: 12
no remaining hits after applying positivelist, returning NULL object... generating contexts
Error in nrow(ctxt@cpos) :
trying to get slot "cpos" from an object of a basic class ("NULL") with no slots

[ChristophLeonhardt]

I think I have a possible explanation:

In the context-method, there is the following line:
if (!is.null(positivelist)) ctxt <- trim(ctxt, positivelist = positivelist, regex = regex, verbose = verbose)

It is possible that all elements within ctxt are getting dropped here. The problem is that this isn't being checked again in the code that follows the positivelist trim:

if (!is.null(stoplist)) ctxt <- polmineR:::trim(ctxt, stoplist = stoplist, regex = regex, verbose = verbose)

And even if ctxt isn't NULL here, it could be after the stoplist is used for filtering. To solve this, it might be possible to just but

if (!is.null(stoplist) && !is.null(ctxt)) ctxt <- polmineR:::trim(ctxt, stoplist = stoplist, regex = regex, verbose = verbose)

and check again afterwards if ctxt is NULL now.

if (is.null(ctxt)) return(NULL)

[ablaette]

Thanks a lot, particularly for developing a potential solution! The check whether ctxt is not NULL is exactly what is required to avoid the error.

For checking whether everything works, I used the following code, which adopts the new workflow to create subcorpora / partitions.

library(polmineR)
use("GermaParl")

corpus("GERMAPARL") %>%
  subset(lp == "17" & role %in% c("gov", "mp")) %>%
  kwic(query = '"[Aa]ttack.*"', positivelist = "Messer", cqp = TRUE) 

corpus("GERMAPARL") %>%
  kwic(query = '"[Aa]ttack.*"', cqp = TRUE, positivelist = "Messer")

The bugfix is included in polmineR v0.7.11.9026 I just pushed to the dev branch.

[ablaette]

I have introduced a test that will check that the return value is NULL if all matches have been dropped due to a positivelist. As this may safeguard the confidence that everything works as intended, I will close the issue.