Bug #16143: BOM UTF-8 is not removed after rewind

• Status: Feedback
• Assignee: -
• Target version: -
• ruby -v: ruby 2.6.2p47 (2019-03-13 revision 67232) [x86_64-darwin17]
• Backport: 2.5: UNKNOWN, 2.6: UNKNOWN
• Tags: core, patch

[Dirk (Dirk Meier-Eickhoff)]

I have a CSV file with "forced quotes" and UTF-8 BOM (\xEF\xBB\xBF) which CSV can not read after a rewind. I get "CSV::MalformedCSVError: Illegal quoting in line 1."

My UTF-8 CSV file with BOM:

File.open('bom_test.csv', 'w') do |io|
  io.write("\xEF\xBB\xBF\"Name\",\"City\"\n\"John Doe\",\"New York\"")
end

Reproduce error:

# Case 1
csv = CSV.open('bom_test.csv', 'r:BOM|UTF-8', {headers: true})
csv.shift
# => #<CSV::Row "Name":"John Doe" "City":"New York">
csv.rewind
csv.shift
# => CSV::MalformedCSVError (Illegal quoting in line 1.)

# Case 2
csv = CSV.open('bom_test.csv', 'r:BOM|UTF-8', {headers: true})
csv.readline
# => #<CSV::Row "Name":"John Doe" "City":"New York">
csv.rewind
csv.readline
# => CSV::MalformedCSVError (Illegal quoting in line 1.)

Sutou Kouhei has posted other reproducable code to my first issue at CSV gem: https://github.com/ruby/csv/issues/103

File.open("/tmp/a.txt", "w") do |x|
  x.puts("\xEF\xBB\xBFa,b,c")
end
File.open("/tmp/a.txt", "r:BOM|UTF-8") do |x|
  p x.gets.unpack("U*") # => [97, 44, 98, 44, 99, 10]
  x.rewind
  p x.gets.unpack("U*") # => [65279, 97, 44, 98, 44, 99, 10]
end

He said: "This [CSV] library rely on Ruby's BOM processing. It seems that Ruby's BOM processing doesn't support rewind."

My expectation is that reading a file with BOM always return the same content, regardless of first reading or after a rewind.

[nobu (Nobuyoshi Nakada)]

I'm afraid if the spec of BOM is such simple and obvious.

Implemented but I'm sure that something is overlooked.
https://github.com/ruby/ruby/pull/2430

[kou (Kouhei Sutou)]

I've reviewed the pull request. I found a problem.

[akr (Akira Tanaka)]

I feel changing the default behavior of IO#rewind is dangerous.

We use IO#rewind when we modify a file in place.

open(filename, "r+") {|f|
  f.read
  f.rewind
  f.truncate(0)
  f.write "..."
}

If IO#rewind moves the file pointer to just after BOM,
BOM is changed to NULs in above code.

I think adding a keyword argument for IO#rewind is better for compatibility.

[mame (Yusuke Endoh)]

• Status changed from Open to Feedback

I think this issue can be easily worked around by using IO#set_encoding_by_bom which was introduced by #15210.

csv = CSV.open('bom_test.csv', 'r:BOM|UTF-8', headers: true)
p csv.shift  #=> #<CSV::Row "Name":"John Doe" "City":"New York">

# workaround
csv.rewind
csv.binmode
csv.to_io.set_encoding_by_bom

p csv.shift  #=> #<CSV::Row "Name":"John Doe" "City":"New York">

Do we really need any change? It is surprising to me if IO#pos is non-zero after IO#rewind. IO#rewind(bom: true) or something, which @akr (Akira Tanaka) proposes, may be less surprising. But IMHO, IO#set_encoding_by_bom is enough.