[ruby-core:81591] Re: [ruby-cvs:66197] ko1:r59023 (trunk): revert r59020 because it may fail some tests sometimes on some environment (http://ci.rvm.jp/). This revert is to check the reason of failures.

[Eric Wong <normalperson@yhbt.net>]

Eric Wong <normalperson@yhbt.net> wrote:
> ko1@ruby-lang.org wrote:
> > ko1	2017-06-06 15:58:54 +0900 (Tue, 06 Jun 2017)
> > 
> >   New Revision: 59023
> > 
> >   https://svn.ruby-lang.org/cgi-bin/viewvc.cgi?view=revision&revision=59023
> > 
> >   Log:
> >     revert r59020 because it may fail some tests sometimes on some environment (http://ci.rvm.jp/). This revert is to check the reason of failures.
> 
> Any progress on tracking these failures?  I suspect there are
> existing bugs which r59020 exposed, causing the failures.
> I will check the IO#gets and buffering code, too.
> 
> Initial investigation shows r57224 looks like it may lose data...

Ah, r59020 interacts badly with r57422.

I think the API for close notification (and FD waiting)
needs to be reworked to avoid races in MT.

On a related note, the new C API in [Feature #13618] (auto
fiber) uses "int *fd" and "rb_io_t *fptr" (not "int fd") to
avoid race conditions, but I think there is more work to
be done in MT.