Bug #13965: Surprising behavior when using Tempfile.open

• Status: Rejected
• Assignee: -
• Target version: -
• ruby -v: 2.4.1p111
• Backport: 2.3: UNKNOWN, 2.4: UNKNOWN

[davemyron (Dave Myron)]

I was trying to write the contents of an array to a Tempfile using the #open() + block syntax: Tempfile.open { |f| my_array.each { |item| f.puts item }}

However, I was finding that only most of the items were being written to the file which, in my case, was Very Bad.

Witness (with Ruby 2.4.1p111; same behavior in 2.4.3):

>> Tempfile.open() { |f| (1..10000).each { |i| f.puts i }; `tail -n1 #{f.path}` }
=> "8416"
>> Tempfile.open() { |f| (1..100000).each { |i| f.puts i }; `tail -n1 #{f.path}` }
=> "98835"
>> Tempfile.open() { |f| f.puts "Test\n"; `tail -n1 #{f.path}` }
=> ""

All very surprising results! When using block syntax, I expect to be able to interact with the Tempfile inside the block and then forget about it.

Storing the path of the Tempfile and accessing it after the block is closed produces the expected results:

>> path = ""; Tempfile.open() { |f| path = f.path; f.puts "Test\n" }; `tail -n1 #{path}`
=> "Test\n"
>> path = ""; Tempfile.open() { |f| path = f.path; (1..10000).each { |i| f.puts i }}; `tail -n1 #{path}`
=> "10000\n"

I suspect that it's due to some buffering. Nonetheless, it's unexpected to have to store the path to the Tempfile when using the block syntax and to interact with after closing the block.

[davemyron (Dave Myron)]

This appears to be the behavior of File.open + block, as well:

>> File.open("/tmp/test.txt","w+"){|f| (1..10000).each { |i| f.puts i }; `tail -n1 #{f.path}` }
=> "8416"

[davemyron (Dave Myron)]

To workaround the unexpected behavior, either use IO#sync = true or IO#flush after writing to the file.

[nobu (Nobuyoshi Nakada)]

• Status changed from Open to Rejected

Yes, it's a nature of buffering.

[davemyron (Dave Myron)]

Could a small documentation blurb alleviate future headaches for other users?

If the docs had said something like "Due to buffering, the contents of a file may be incomplete until flushed" I would not have been (so) surprised.