IO#close should not raise IOError on closed IO objects.
I'd like to change IO#close.
It should not raise IOError on closed IO objects.
We sometimes invoke IO#close only when the IO object is not closed as:
f.close if !f.closed?
If this issue is accepted, we can write it simply as follows.
Simple grep finds many examples.
Following examples are just a little excerpt.
lib/webrick/server.rb: sock.close unless sock.closed? lib/pstore.rb: file.close if !file.closed? lib/mkmf.rb: @log.close if @log and not @log.closed? lib/cgi/session.rb: f.close if f and !f.closed? lib/open-uri.rb: io.close if !io.closed? lib/net/pop.rb: s.close if s and not s.closed? lib/net/http.rb: @socket.close if @socket and not @socket.closed? lib/net/smtp.rb: s.close if s and not s.closed? lib/shell/process-controller.rb: io.close unless io.closed? test/ruby/test_io.rb: w.close unless !w || w.closed? ...
I think there is no problem with the behavior which IO#close doesn't raise an exception on closed IO object.
Because the closed state fulfils the postcondition of the method.
The change means relaxing the precondition which should be harmless.
Moreover raising IOError can smash other exceptions if it is called in ensure clause.
It makes debugging difficult and the proposed behavior ease it.
It also useful to close IO object asynchronously.
Asynchronous close can be used for graceful shutdown.
This issue can eliminate "rescue IOError".
See lib/webrick/server.rb and lib/drb/drb.rb for example.
Note that "double close" is a bad idea in C.
But it is not applicable to Ruby.
A FILE structure is freed on fclose().
But Ruby's IO object is not freed until GC.
So method invocation on closed object doesn't cause invalid memory access.
A file descriptor (FD) can be reused any time because a signal handler or another thread may allocate a FD.
But Ruby's IO object is not reused after close.
So it is impossible to close an IO object unintentionally.
- io.c (rb_io_close_m): Don't raise when the IO object is closed. [ruby-core:67444] [Feature #10718]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@49260 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Updated by normalperson (Eric Wong) over 4 years ago
Eric Wong email@example.com wrote:
Thank you for proposing this. I think this will simplify working with
Ruby IO and make me happier since I work with a lot of IO-related code.
Outside of test cases, I don't forsee compatibility problems either.
Ugh, I take that back...
This can hide existing bugs. For example, I track open connection
counts (persistent HTTP connections are expired in a GC-like manner)
in the multi-threaded yahns server. Something like:
@mtx.synchronize do @count -= 1 io.close end
If I get IOError as in 2.2, I'll there is a major bug (probably race
condition) fix the code. But now, I may need to add extra checks for
io.closed? before every io.close:
@mtx.synchronize do @count -= 1 raise "MAJOR BUG!" if io.closed? io.close end
Anyways, I'm not sure if many people complained about IOError on
IO#close over the past years; so perhaps reverting to 2.2 behavior is
 http://yahns.yhbt.net/README || git clone git://yhbt.net/yahns