0

This might be a hard question to field because I can't easily offer a reproducible SNAFU without providing my access token to the Gmail API, but my hope is that I am tripping over an issue that will be clear enough from my description. We'll see...

I wrote a function that takes the gmail message ID of a Google Scholar Alert and parses it into a data frame with a columns for article title, authors, publication, date, etc. My conundrum is that the same code that works when I load it interactively (i.e., load the function into session RAM a la "my_fancy_function <- function(arguments){blah blah})" does not work when I load the function as part of an R package (a la 'devtools::load_all("/mypackage/")'). It's worth noting that the other main function in my package works fine either way, but this one trips up when I try to use it from the package version after loading the package via devtools::load_all("my_misbehaving_package/").

Below is the .R file content of the function that won't come to heel when loaded as part of a package -- and I hereby acknowledge in advance that it's some ugly mess of code, if that will spare me some finger wagging in your answers. This feels like the package version of the classic "strings as factors" SNAFU, but you tell me. The problem seems to occur very early in the function evaluation, and the error I get reads as follows:

Error in UseMethod("read_xml") : no applicable method for 'read_xml' applied to an object of class "NULL"

And here is my whole sick code for this function:

library(stringr)
library(rvest)
library(plyr)
library(dplyr)
library(lubridate)
library(gmailr)

GScholar_alert_msg_to_df <- function(message_id){
     one_message <- message(message_id)
     msg_html <- body(one_message)
     title <- read_html(msg_html) %>% html_nodes("h3") %>% html_text()
     link <- read_html(msg_html) %>% html_nodes("h3 a") %>% html_attr("href")
     msg_chunks <- msg_html %>% str_split("<a href") %>% unlist
     msg_chunks <- msg_chunks[2:(length(msg_chunks)-2)]
     excerpt <- msg_chunks %>% str_replace_all(fixed("<b>"), "") %>%
          str_replace_all(fixed("</b>"), "") %>%
          str_replace_all(fixed("<br>"), "") %>%
          str_extract("<(font size=2 color=#222222|font size=\"-1\")>(.*?)</font>") %>%
          unlist %>% str_replace_all("<(font size=2 color=#222222||font size=\"-1\")>", "") %>%
          str_replace_all("</font>", "")
     author_pub_field <- msg_chunks %>% str_replace_all(fixed("<b>"), "") %>%
     str_replace_all(fixed("</b>"), "") %>%
     str_extract("<font size=(2|\"-1\") color=#(006621|009933|008000)>(.*?)</font>") %>%
     unlist %>%
     str_replace_all("<font size=(2|\"-1\") color=#(006621|009933|008000)>", "") %>%
     str_replace_all("</font>", "")
     one_message_df <- data.frame(title, excerpt, link, author_pub_field, stringsAsFactors = FALSE)
     one_message_df$date <- date(one_message) # needs reformatting
     one_message_df$date %>% str_replace("[[:alpha:]]{3}, ", "") %>%
          str_extract("^.{11}") %>% dmy -> one_message_df$date
     one_message_df$author_only <- str_detect(one_message_df$author_pub_field, " - ")
     one_message_df$author <- one_message_df$author_pub_field %>%
          str_extract("^(.*?) - ") %>% str_replace(" - ", "")
     one_message_df$author <- ifelse(one_message_df$author_only == 1, one_message_df$author, one_message_df$author_pub_field)
     one_message_df$publication <- one_message_df$author_pub_field %>%
          str_extract(" - (.*?)$") %>% str_replace(" - ", "") %>%
          str_replace(", [0-9]{4}$", "") %>% str_replace("^[0-9]{4}$", NA)
     one_message_df$publication <- str_replace(one_message_df$publication, "^[0-9]{4}$", NA)
     one_message_df$author_MIAs <- str_detect(one_message_df$author, "&hellip;")
     one_message_df$author %>% str_replace("&hellip;", " \\.\\.\\.") ->      one_message_df$author
     one_message_df$pub_name_clipped <- one_message_df$publication %>% str_detect("&hellip;")
     one_message_df$publication %>% str_replace("&hellip;", " \\.\\.\\.") -> one_message_df$publication
     return(one_message_df)

Yeah, it's ugly, but again, I promise it works when the code is entered interactively. As for the misbehaving package version, according to RStudio traceback the error seems to happen way up top, I think either the body() or the message() function I'm using from the gmailr package. (Yes, I've tried the code via Terminal to no happier conclusion.) Help me, oh be anyone, you're my only hope.

Ben
  • 1
  • How exactly dod you create your package? What's in the NAMESPACE file? Did you list all these packages as "import"s? – MrFlick Aug 03 '16 at 13:53
  • @MrFlick I initiated the package from within RStudio using the "New Project..." option from the dropdown menus. I haven't even opened the NAMESPACE file so it only has that one generic regex line in it. I have that list of packages at the top of one of the package's .R files so they load when I load the package. – Ben Aug 03 '16 at 15:49
  • You shouldn't have `library()` statements in code in a package. You should be setting up the NAMESPACE differently. Maybe checkout [writing R packages](http://r-pkgs.had.co.nz/). Also can't you reduce this example to make it *minimally* reproducible? Exactly what line causes the error (remove any not directly related to the error)? – MrFlick Aug 03 '16 at 17:50
  • I know library() statements in packaged code is a no-no, but haven't had problems with it until now so I didn't see the harm if it's a package for my own use. I will do that properly and report back whether it fixes the problem. // I don't see how I can provide reproducible code for this because it relies on having a client_id.json file to access the Gmail API. – Ben Aug 03 '16 at 19:44

1 Answers1

0

Thank you to @MrFlick for indicating a way out of the woods here. I was able to resolve the problem by moving that list of packages to the 'Depends' field of the DESCRIPTION file, and I also had to use the '::' operator to make explicit calls to the 'body' and 'date' functions from the gmailr package. Those two names also belong to nonstandard functions that are part of the 'base' namespace and they can't be masked by loading gmailr within the package environment. According to Hadley, I should be using '::' for every use of a function from another package when I'm writing my own.

Ben
  • 1