[ARROW-3547] [R] Protect against Null crash when reading from RecordBatch

• Type: Improvement
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 0.12.0
• Component/s: R
• External issue URL: https://github.com/apache/arrow/issues/19861
• Assignee: Romain Francois
• Reporter: Javier Luraschi
• Created: 17/Oct/18 19:03
• Updated: 11/Jan/23 07:28
• Resolved: 22/Oct/18 21:57

Description

Reprex:

 

  tbl <- tibble::tibble(
    int = 1:10, dbl = as.numeric(1:10),
    lgl = sample(c(TRUE, FALSE, NA), 10, replace = TRUE),
    chr = letters[1:10]
  )

  batch <- record_batch(tbl)
  bytes <- write_record_batch(batch, raw())

  stream_reader <- record_batch_stream_reader(bytes)
  batch1 <- read_record_batch(stream_reader)

  batch2 <- read_record_batch(stream_reader)
  
  # Crash
  as_tibble(batch2)

 

While users should check for Null entries by running:

 

if(!batch2$is_null()) as_tibble(batch2)

It's harsh to trigger a crash, we should consider protecting all functions that use RecordBatch pointers to return NULL instead, for instance:

 

List RecordBatch__to_dataframe(const std::shared_ptr<arrow::RecordBatch>& batch) {
   if (batch->get() == nullptr) Rcpp::stop("Can't read from NULL record batch.")
}