Bug #12989: Passing `binmode: true` to `IO.pipe` makes `binmode?` return `true` but encoding is not binary

• Status: Closed
• Assignee: nobu (Nobuyoshi Nakada)
• Target version: -
• ruby -v: ruby 2.4.0dev (2016-11-16 gc-compact 56805) [x86_64-darwin16]
• Backport: 2.1: UNKNOWN, 2.2: UNKNOWN, 2.3: UNKNOWN

[tenderlovemaking (Aaron Patterson)]

Here is a sample program:

Encoding.default_external = Encoding::UTF_8
Encoding.default_internal = Encoding::UTF_8

reader, writer = IO.pipe(binmode: true)
reader.binmode?           # => true
reader.external_encoding  # => #<Encoding:UTF-8>
writer.binmode?           # => true
writer.external_encoding  # => #<Encoding:UTF-8>

reader, writer = IO.pipe
reader.binmode
writer.binmode

reader.binmode?           # => true
reader.external_encoding  # => #<Encoding:ASCII-8BIT>
writer.binmode?           # => true
writer.external_encoding  # => #<Encoding:ASCII-8BIT>

I think that passing binmode: true to IO.pipe should behave the same way as calling binmode on each file. Today, passing binmode: true to IO.pipe puts the files in a strange state: they are binary and not binary.

I've attached a patch to fix the problem.

[nobu (Nobuyoshi Nakada)]

• Description updated (diff)

Encoding argument should be prior to binmode option.

open(IO::NULL, "r", binmode: true){|f| p [f.binmode?, f.external_encoding]} #=> [true, #<Encoding:ASCII-8BIT>]
open(IO::NULL, "r:utf-8", binmode: true){|f| p [f.binmode?, f.external_encoding]} #=> [true, #<Encoding:UTF-8>]

Your patch always makes external_encoding ASCII-8BIT when binmode is set, even if an encoding is given.

[jeremyevans0 (Jeremy Evans)]

• File io-pipe-binmode.patch io-pipe-binmode.patch added

nobu (Nobuyoshi Nakada) wrote:

Encoding argument should be prior to binmode option.

open(IO::NULL, "r", binmode: true){|f| p [f.binmode?, f.external_encoding]} #=> [true, #<Encoding:ASCII-8BIT>]
open(IO::NULL, "r:utf-8", binmode: true){|f| p [f.binmode?, f.external_encoding]} #=> [true, #<Encoding:UTF-8>]

Your patch always makes external_encoding ASCII-8BIT when binmode is set, even if an encoding is given.

Here's an updated patch based on tenderlove's original patch that does not change the encoding of the returned IO objects if encoding arguments are given in addition to the binmode option. Is this acceptable?

[ko1 (Koichi Sasada)]

• Status changed from Open to Assigned
• Assignee set to nobu (Nobuyoshi Nakada)

[nobu (Nobuyoshi Nakada)]

jeremyevans0 (Jeremy Evans) wrote:

Here's an updated patch based on tenderlove's original patch that does not change the encoding of the returned IO objects if encoding arguments are given in addition to the binmode option. Is this acceptable?

I think so.
In general, dedicated assertion methods e.g.,assert_predicate would be preferable to mere assert, I think.

• Status changed from Assigned to Closed

Applied in changeset git|ebc99e026d0ae770b297a93d1f1c1ceeffd13bfc.

---

Do not change IO.pipe encodings if encodings explicitly given

This commit makes it so that if the binmode option is given with
any encoding arguments, the reader and writer IO objects are
not set to binary encoding.

Fixes [Bug #12989]