Bug #22537
open`IO#gets(nil, limit)` stops right after a 0xFF byte
Description
IO#gets(nil, limit) (no separator, with a limit) returns early when the data contains a 0xFF byte: it returns the bytes up to and including the first 0xFF, even though neither the limit nor EOF has been reached. IO#gets(nil) and IO#read(limit) return the whole data.
puts RUBY_DESCRIPTION
File.binwrite("t.bin", "ab\xFFcd")
File.open("t.bin", "rb") { |f| p f.gets(nil, 100) }
File.open("t.bin", "rb") { |f| p f.gets(nil) }
File.open("t.bin", "rb") { |f| p f.read(100) }
File.open("t.bin", "rb") { |f| p f.each_line(nil, 100).to_a }
File.binwrite("t.bin", "ab\xFEcd")
File.open("t.bin", "rb") { |f| p f.gets(nil, 100) }
ruby 4.1.0dev (2026-11-09T07:44:48Z HEAD 4f0a0a0eca) +PRISM [x86_64-linux]
"ab\xFF"
"ab\xFFcd"
"ab\xFFcd"
["ab\xFF", "cd"]
"ab\xFEcd"
Expected: "ab\xFFcd" for gets(nil, 100) and ["ab\xFFcd"] for each_line(nil, 100), because nil means "no separator" and the data is shorter than the limit.
The output is the same on 4.0.0 and 3.0.0, and gets(nil, 100) also returns "ab\xFF" on 2.7.0, 2.0.0 and 1.9.3. Ruby 1.9.1 returns "ab\xFFcd". IO#readline(nil, 100) and a pipe behave the same way as the file above. StringIO#gets(nil, 100) returns the whole string. Other bytes such as 0xFE are not affected.
It looks like the cause is the -1 used as "no delimiter" in rb_io_getline_0:
https://github.com/ruby/ruby/blob/v4.0.0/io.c#L4200
When rs is nil, newline stays -1 and is passed to appendline as delim, and search_delim calls memchr(p, delim, len) when the minimum character length of the encoding is 1:
https://github.com/ruby/ruby/blob/v4.0.0/io.c#L3882-L3887
memchr converts its second argument to unsigned char, so -1 matches 0xFF.
I found this through the csv library: CSV.read(path) reads a sample with @input.gets(nil, chunk_size) to detect the row separator, and a file that contains 0xFF gets a truncated sample (https://github.com/ruby/csv/issues/385).
No data to display