Fail to read multiple-line string with multiple tasks

[huangyxi]

CSV.read failed when CSV files contain multiple-line strings with multiple tasks.

Reproduction

Threads.nthreads() # 240
using DataFrames
using CSV

n = 4000 # change to larger number if unable to reproduce
text = "123
abc" # multiple-line string
a = DataFrame(id=1:n, text=fill(text, n))
CSV.write("test.csv", a)
b = CSV.read("test.csv", DataFrame) # Error
# CSV.read("test.csv", DataFrame, ntasks=1) is OK

[TheLostLambda]

I've hit this as well — very frustrating! I was personally getting a "fatal error, encountered an invalidly quoted field while parsing" error on a file with no quoting issues.

That's presumably because the file chunking for the multithreaded operation has split the input file in the middle of one of those multiline strings.

Maybe the first line of each chunk should try to read the first row, see if there is a quoting error, and if so, backtrack until it finds the opening quote? The last row of each chunk can look for unclosed quotes and drop the row quietly if there are any (since that other chunk would be backtracking to pick it up? The first and last chunks would need to work a little differently though. This retry behavior should keep the already-working cases running at the same speed whilst still fixing this issue?

[quinnj]

This should be fixed in #1184