[R] read_csv_arrow fails when a string contains a backslash-escaped quote mark followed by a comma

[asfimport]

read_csv_arrow() incorrectly parses CSV files when a string value contains a comma that appears after a backslash-escaped quote mark. Originally noted by Thomas Klebel https://scicomm.xyz/@tklebel/109270436511066953

This is an example that throws the error:

x <- tempfile()
readr::write_lines(
'
id,text
1,"some text on \\"BLAH
" and X, and Y also"
', x)

cat(system(paste('cat', x), intern = TRUE), sep = "\n")
#> 
#> id,text
#> 1,"some text on \"BLAH\" and X, and Y also"
arrow::read_csv_arrow(x, escape_backslash = TRUE)
#> Error:
#> ! Invalid: CSV parse error: Expected 2 columns, got 3: 1,"some text on \"BLAH\" and X, and Y also"

#> Backtrace:
#> ▆
#> 1. └─arrow (local) `<fn>`(file = x, escape_backslash = TRUE, delim = ",")
#> 2. └─base::tryCatch(...) at r/R/csv.R:217:2
#> 3. └─base (local) tryCatchList(expr, classes, parentenv, handlers)
#> 4. └─base (local) tryCatchOne(expr, names, parentenv, handlers[[1L]])
#> 5. └─value[[3L]](cond)
#> 6. └─arrow:::augment_io_error_msg(e, call, schema = schema) at r/R/csv.R:222:6
#> 7. └─rlang::abort(msg, call = call) at r/R/util.R:251:2

^{Created on 2022-11-02 with reprex v2.0.2}

This version includes four lines that might potentially error but do not:

x <- tempfile()
readr::write_lines(
'
id,text
2,"some text on X and Y"
3,"some text on X, and Y"
4,"some text on \\"BLAH
"
5,"some text on X and Y, and \\"BLAH
" also"
', x)

cat(system(paste('cat', x), intern = TRUE), sep = "\n")
#> 
#> id,text
#> 2,"some text on X and Y"
#> 3,"some text on X, and Y"
#> 4,"some text on \"BLAH\"
#> 5,"some text on X and Y, and \"BLAH\" also"
arrow::read_csv_arrow(x, escape_backslash = TRUE)
#> # A tibble: 4 × 2
#> id text 
#> <int> <chr> 
#> 1 2 "some text on X and Y" 
#> 2 3 "some text on X, and Y" 
#> 3 4 "some text on \\BLAH\\\"" 
#> 4 5 "some text on X and Y, and \\BLAH\\\" also\""

^{Created on 2022-11-02 with reprex v2.0.2}

I'm not sure if the problem is R specific. I've partially reproduced the error using reticulate and pyarrow as follows, but notice that this errors at a different point: the pyarrow version appears to fail with the comma preceding the backslash-escaped quote mark:

x <- tempfile()
readr::write_lines(
'
id,text
1,"some text on X and Y"
2,"some text on X, and Y"
3,"some text on \\"BLAH
"
4,"some text on X and Y, and \\"BLAH
" also"
5,"some text on \\"BLAH
" and X, and Y also"
', x)

cat(system(paste('cat', x), intern = TRUE), sep = "\n")
#> 
#> id,text
#> 1,"some text on X and Y"
#> 2,"some text on X, and Y"
#> 3,"some text on \"BLAH\"
#> 4,"some text on X and Y, and \"BLAH\" also"
#> 5,"some text on \"BLAH\" and X, and Y also"

csv <- reticulate::import("pyarrow.csv")
opt <- csv$ParseOptions(escape_char='
')
csv$read_csv(x, parse_options = opt)
#> Error in py_call_impl(callable, dots$args, dots$keywords): pyarrow.lib.ArrowInvalid: CSV parse error: Expected 2 columns, got 3: 3,"some text on \"BLAH\"
#> 4,"some text on X and Y, and \"BLAH\" also"

^{Created on 2022-11-02 with reprex v2.0.2}

Reporter: Danielle Navarro / @djnavarro

_{Note: This issue was originally created as ARROW-18219. Please see the migration documentation for further details.}

[asfimport]

Antoine Pitrou / @pitrou:
How was the original CSV file produced? By which utility?

[asfimport]

Antoine Pitrou / @pitrou:
Note the CSV is incorrect: the third line misses an ending quote before the newline.

[asfimport]

Antoine Pitrou / @pitrou:
(for the record: the Python CSV reader also fails reading this CSV file "correctly")

[asfimport]

Thomas Klebel:
The original CSV I was having issues looked like this:

id,text
1,Some interesting text
2,"Some text on: \"how to break arrow\" by X, and Y" 

It was created using spark_write_csv, with

sparklyr v1.7.5

Spark v2.3.2

HDFS v3.1.1

[asfimport]

Antoine Pitrou / @pitrou:
[~tklebel] That file is parsed correctly by PyArrow using the following options:

>>> s
'id,text\n1,Some interesting text\n2,"Some text on: \\"how to break arrow\\" by X, and Y" \n'
>>> parse_options = csv.ParseOptions(escape_char='\\', quote_char='"', double_quote=False)
>>> csv.read_csv(io.BytesIO(s.encode()), parse_options=parse_options).to_pandas()
   id                                             text
0   1                            Some interesting text
1   2  Some text on: "how to break arrow" by X, and Y 

[asfimport]

Antoine Pitrou / @pitrou:
cc @thisisnic @paleolimbot

[thisisnic]

Reproducible in 23.0.1 and re-triaged with Claude - verify before fixing

---

The C++ CSV parser supports combining quoting + escaping, and Python passes both options correctly. However, the R binding in csv___ParseOptions__initialize() sets escape_char but never sets the escaping boolean — it's missing entirely. So escaping stays at its C++ default (false) regardless of what R users pass.

Relevant code:

arrow/r/src/csv.cpp

Lines 60 to 72 in b3362f2

	std::shared_ptr<arrow::csv::ParseOptions> csv___ParseOptions__initialize(
	cpp11::list options) {
	auto res =
	std::make_shared<arrow::csv::ParseOptions>(arrow::csv::ParseOptions::Defaults());
	res->delimiter = cpp11::as_cpp<char>(options["delimiter"]);
	res->quoting = cpp11::as_cpp<bool>(options["quoting"]);
	res->quote_char = cpp11::as_cpp<char>(options["quote_char"]);
	res->double_quote = cpp11::as_cpp<bool>(options["double_quote"]);
	res->escape_char = cpp11::as_cpp<char>(options["escape_char"]);
	res->newlines_in_values = cpp11::as_cpp<bool>(options["newlines_in_values"]);
	res->ignore_empty_lines = cpp11::as_cpp<bool>(options["ignore_empty_lines"]);
	return res;
	}

One-line fix: add res->escaping = cpp11::as_cpp(options["escaping"]); after the escape_char line.