Hi, I was recently debugging a hang on the mysql2 gem. The cause turned out to be that the socket file descriptor from libmysql was being passed in to rb_thread_select(): https://github.com/brianmario/mysql2/blob/8c0c350c... This worked fine on linux, but on Windows rb_thread_select would never return. The rb_thread_select function expects CRT fds (made with _open_osfhandle()). Ruby win32 code goes to great lengths to redefine socket(), FD_SET, FD_ISSET, and a pile of other functions, so that extensions like EventMachine create and work with the CRT fds instead of the SOCKET handles transparently. However in cases like like mysql2 and curb (which uses libcurl), the socket is created by an external library, and the handle is the SOCKET, not the CRT fd. Here's the curb case: https://github.com/taf2/curb/blob/5f223a8f2e0de3e6... My question is; why does Ruby go through all this trouble on Windows? Why not pass around and work with SOCKET handles instead? -Greg
on 2011-01-14 04:34
on 2011-01-14 05:04
Hello,
In message "[ruby-core:34483] Windows sockets and CRT fds"
on Jan.14,2011 12:27:48, <ghazel@gmail.com> wrote:
> My question is; why does Ruby go through all this trouble on Windows?
> Why not pass around and work with SOCKET handles instead?
Because, if do so, we must rewrite all ruby core functions and
extensions which expect that socket is a kind of file descriptor.
About mysql2 gem, it should know how MySQL treats Windows SOCKETs
and how Ruby treats them.
And it should fill in the gap. It's the work of wrapper library.
Regards,
on 2011-01-14 05:37
On Thu, Jan 13, 2011 at 8:03 PM, U.Nakamura <usa@garbagecollect.jp> wrote: > In message "[ruby-core:34483] Windows sockets and CRT fds" > on Jan.14,2011 12:27:48, <ghazel@gmail.com> wrote: >> My question is; why does Ruby go through all this trouble on Windows? >> Why not pass around and work with SOCKET handles instead? > > Because, if do so, we must rewrite all ruby core functions and > extensions which expect that socket is a kind of file descriptor. I understand it would have to be rewritten to change it, but why was it written that way to begin with? Aside from the effort to change it, is there anything lost or gained by switching? > About mysql2 gem, it should know how MySQL treats Windows SOCKETs > and how Ruby treats them. > And it should fill in the gap. It's the work of wrapper library. Unfortunately this is quite difficult. Libraries like libmysql create and destroy their own sockets, so it is rarely possible to create and destroy a CRT fd at the same time. Creating a CRT fd when you need it (to pass to rb_thread_select) is fine, but closing it afterwards closes the underlying socket at well. Tracking the associated CRT fds and trying to close them when it's safe is a lot of effort, and maybe not possible if you don't get a destroy event. More importantly: no one has done it correctly. Every author of a ruby extension I can find has completely missed this subtlety of Ruby's Win32 code - since just treating linux fds as fds with rb_select_thread works fine. So, I think by switching to SOCKET handles right now, more extensions would start working on Windows than do today. I have yet to find one which would be broken by the switch. -Greg
on 2011-01-14 07:06
Hello,
In message "[ruby-core:34485] Re: Windows sockets and CRT fds"
on Jan.14,2011 13:35:42, <ghazel@gmail.com> wrote:
> I understand it would have to be rewritten to change it, but why was
> it written that way to begin with? Aside from the effort to change it,
> is there anything lost or gained by switching?
Q: why was it written that way to begin with?
A: Sorry, I don't know. I guess that the author might not want
to put many #ifdef-s into ruby core source.
Q: Aside from the effort to change it, is there anything lost or
gained by switching?
A: Currently the programmer of an extension library doesn't need
to know about Windows if the extension library is not linked
to outer libraries. After the switching, he must write two
code for Unix and Windows. Of course, if the extension library
depends on outer libraries which treat Unix and Windows by
different way, his time will decrease because of omitting
the code for ruby's faked FD.
> So, I think by switching to SOCKET handles right now, more extensions
> would start working on Windows than do today. I have yet to find one
> which would be broken by the switch.
Anyway, the patch is welcomed.
If the patch is better enough than current, it will be applied
naturally. (However, the timing is not certain because we must
take care of ABI compatiblity.)
Regards,
on 2011-01-14 23:17
On Thu, Jan 13, 2011 at 10:03 PM, U.Nakamura <usa@garbagecollect.jp> wrote: > to put many #ifdef-s into ruby core source. win32.c and win32.h have a ton of ifdefs in them. What I'm proposing would remove many of them, and not add any to the rest of Ruby code. Who is the author of the _open_osfhandle code? > Q: Aside from the effort to change it, is there anything lost or > gained by switching? > A: Currently the programmer of an extension library doesn't need > to know about Windows if the extension library is not linked > to outer libraries. After the switching, he must write two > code for Unix and Windows. Of course, if the extension library > depends on outer libraries which treat Unix and Windows by > different way, his time will decrease because of omitting > the code for ruby's faked FD. I'm not sure which parts you're saying need to be rewritten for Unix and Windows. Currently there needs to be different code for Unix and Windows, because a socket handle you might come across on Windows has to be adapted to the sort of fd that ruby uses. After the hypothetical switch to using regular Windows sockets in Ruby, you could take a socket handle from anywhere and select on it in Ruby and have no Windows specific code (except inside of Ruby). So, I only see it reducing the amount of platform specific code (and bugs) in extensions. -Greg
on 2011-01-15 01:48
At 6:49 AM +0900 1/15/11, ghazel@gmail.com wrote: >> A: Sorry, I don't know. I guess that the author might not want >> to put many #ifdef-s into ruby core source. > >win32.c and win32.h have a ton of ifdefs in them. What I'm proposing >would remove many of them, and not add any to the rest of Ruby code. >Who is the author of the _open_osfhandle code? [Sometimes I find playing with git a bit like a crossword puzzle ... there's almost always some way to do what I want ...] Here's a summary of all the locations in win32/win32.c that reference open_osfhandle. The first part of each listing uses git blame to list the last author who changed that line. This section also includes the 3 lines before and after the reference. Each of these lines starts with the first 8 characters of the git commit SHA of thfollowed by the author, the date, line number and then the line itself. The second part displays the commit message for the fourth line where open_osfhandle is referenced. The subversion revisionnumber is on the line in this section starting with git-svn-id. [ruby-git (trunk)]$ git grep -nh open_osfhandle win32/win32.c | cut -d : -f 1 | xargs -n 1 ruby -e 'linenum=ARGV[0].to_i;blame=`git blame -c -L #{linenum-3},+7 win32/win32.c`; puts"\n\n#{blame}\n"; puts `git log -1 #{blame.split("\n")[3][0..7]}`' d35fdf33 ( usa 2010-09-02 00:02:52 +0000 66)static structChildRecord *CreateChild(const WCHAR *, const WCHAR *, SECURITY_ATTRIBUTES *, HANDLE, HANDLE, HANDLE); b1469c13 ( usa 2004-06-09 03:47:08 +0000 67)static int has_redirection(const char *); bb022bed ( usa 2007-02-23 09:33:53 +0000 68)int rb_w32_wait_events(HANDLE *events, int num, DWORD timeout); 1aa54c6d ( usa 2007-06-27 11:47:53 +0000 69)static int rb_w32_open_osfhandle(intptr_t osfhandle, int flags); a8209ffc ( wanabe 2010-04-30 21:45:33 +0000 70)static int wstati64(const WCHAR *path, struct stati64*st); 2c01a07b ( nobu 2010-05-08 03:25:17 +0000 71)VALUE rb_w32_conv_from_wchar(const WCHAR *wstr, rb_encoding *enc); edf2e9b7 ( matz 1999-01-20 04:45:32 +0000 72) commit 1aa54c6db6a10bf671a80eddf93630fbbea4f810 Author: usa <usa@b2dd03c8-39d4-4d8f-98ff-823fe69b080e> Date: Wed Jun 27 11:47:53 2007 +0000 * include/ruby/win32.h, win32/Makefile.sub, win32/configure.bat, win32/mkexports.rb, win32/setup.mak, win32/win32.c: import x64-mswin64 port. git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@12635 b2dd03c8-39d4-4d8f-98ff-823fe69b080e 65a51625 ( matz 1999-08-13 05:45:20 +0000 2056)#define FTEXT 0x80 /* file handle is in text mode */ 65a51625 ( matz 1999-08-13 05:45:20 +0000 2057) 6d53b692 ( eban 2000-05-13 16:13:31 +0000 2058)static int 1aa54c6d ( usa 2007-06-27 11:47:53 +0000 2059)rb_w32_open_osfhandle(intptr_t osfhandle, int flags) 65a51625 ( matz 1999-08-13 05:45:20 +0000 2060){ 65a51625 ( matz 1999-08-13 05:45:20 +0000 2061) int fh; 65a51625 ( matz 1999-08-13 05:45:20 +0000 2062) char fileflags; /* _osfile flags */ commit 1aa54c6db6a10bf671a80eddf93630fbbea4f810 Author: usa <usa@b2dd03c8-39d4-4d8f-98ff-823fe69b080e> Date: Wed Jun 27 11:47:53 2007 +0000 * include/ruby/win32.h, win32/Makefile.sub, win32/configure.bat, win32/mkexports.rb, win32/setup.mak, win32/win32.c: import x64-mswin64 port. git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@12635 b2dd03c8-39d4-4d8f-98ff-823fe69b080e 3d6fde33 ( matz 2000-02-18 06:59:36 +0000 2076) 7345d0ef ( nobu 2002-09-29 09:31:58 +0000 2077) /* attempt to allocate a C Runtimefile handle */ 7345d0ef ( nobu 2002-09-29 09:31:58 +0000 2078) hF =CreateFile("NUL", 0, 0, NULL, OPEN_ALWAYS, 0, NULL); 150b4efa ( nobu 2010-10-12 14:58:23 +0000 2079) fh =_open_osfhandle((intptr_t)hF, 0); 7345d0ef ( nobu 2002-09-29 09:31:58 +0000 2080) CloseHandle(hF); 7345d0ef ( nobu 2002-09-29 09:31:58 +0000 2081) if (fh == -1) { 7345d0ef ( nobu 2002-09-29 09:31:58 +0000 2082) errno = EMFILE; /* too many openfiles */ commit 150b4efa5510d77fdde7b9692b3c391fbde6ac19 Author: nobu <nobu@b2dd03c8-39d4-4d8f-98ff-823fe69b080e> Date: Tue Oct 12 14:58:23 2010 +0000 * win32/win32.c (rb_w32_open_osfhandle, rb_w32_wopen, rb_w32_pipe): use uintptr_t instead of long for win64. * win32/win32.c (socketpair_internal): suppress warnings. * win32/win32.c (ftruncate): use HANDLE instead of long for win64. git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@29467 b2dd03c8-39d4-4d8f-98ff-823fe69b080e 65a51625 ( matz 1999-08-13 05:45:20 +0000 2141) 160bfa5b ( ocean 2005-01-06 11:05:31 +0000 2142)#ifdef __BORLANDC__ 160bfa5b ( ocean 2005-01-06 11:05:31 +0000 2143)static int 92cad430 ( nobu 2007-07-21 08:32:34 +0000 2144)rb_w32_open_osfhandle(intptr_t osfhandle, int flags) 160bfa5b ( ocean 2005-01-06 11:05:31 +0000 2145){ 160bfa5b ( ocean 2005-01-06 11:05:31 +0000 2146) int fd = _open_osfhandle(osfhandle, flags); 160bfa5b ( ocean 2005-01-06 11:05:31 +0000 2147) if (fd == -1) { commit 92cad4305b1751fcea0bcfe1386e8f8cba265167 Author: nobu <nobu@b2dd03c8-39d4-4d8f-98ff-823fe69b080e> Date: Sat Jul 21 08:32:34 2007 +0000 * bcc32/{Makefile.sub,configure.bat,setup.mak: configure_args support. * bcc32/setup.mak: check runtime version. * win32/win32.c (rb_w32_open_osfhandle): prototype has changed in bcc 5.82. git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@12831 b2dd03c8-39d4-4d8f-98ff-823fe69b080e 160bfa5b ( ocean 2005-01-06 11:05:31 +0000 2143)static int 92cad430 ( nobu 2007-07-21 08:32:34 +0000 2144)rb_w32_open_osfhandle(intptr_t osfhandle, int flags) 160bfa5b ( ocean 2005-01-06 11:05:31 +0000 2145){ 160bfa5b ( ocean 2005-01-06 11:05:31 +0000 2146) int fd = _open_osfhandle(osfhandle, flags); 160bfa5b ( ocean 2005-01-06 11:05:31 +0000 2147) if (fd == -1) { 160bfa5b ( ocean 2005-01-06 11:05:31 +0000 2148) errno = EMFILE; /* too many openfiles */ 160bfa5b ( ocean 2005-01-06 11:05:31 +0000 2149) _doserrno = 0L; /* not an OS error */ commit 160bfa5befa7de9a8675afe4efb05882e5020301 Author: ocean <ocean@b2dd03c8-39d4-4d8f-98ff-823fe69b080e> Date: Thu Jan 6 11:05:31 2005 +0000 * win32/win32.c (rb_w32_close): didn't close socket handle. * win32/win32.c (rb_w32_open_osfhandle): bcc32's _open_osfhandle never set EMFILE. git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@7737 b2dd03c8-39d4-4d8f-98ff-823fe69b080e edf2e9b7 ( matz 1999-01-20 04:45:32 +0000 2687) } 62280094 ( usa 2002-09-28 21:39:57 +0000 2688) RUBY_CRITICAL({ d544a3d6 ( usa 2008-11-11 08:54:07 +0000 2689) HANDLE h = CreateFile("NUL", 0, 0, NULL, OPEN_ALWAYS, 0, NULL); d544a3d6 ( usa 2008-11-11 08:54:07 +0000 2690) fd = rb_w32_open_osfhandle((intptr_t)h, O_RDWR|O_BINARY|O_NOINHERIT); d544a3d6 ( usa 2008-11-11 08:54:07 +0000 2691) if (fd != -1) { d544a3d6 ( usa 2008-11-11 08:54:07 +0000 2692) r = accept(TO_SOCKET(s), addr, addrlen); d544a3d6 ( usa 2008-11-11 08:54:07 +0000 2693) if (r != INVALID_SOCKET) { commit d544a3d6c721b435151a74cb00aa32142a284097 Author: usa <usa@b2dd03c8-39d4-4d8f-98ff-823fe69b080e> Date: Tue Nov 11 08:54:07 2008 +0000 * win32/win32.c (rb_w32_accept): secure fd before accept because if error causes in securing, cannot restore the state of accepted socket. fixed [ruby-core:19728] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@20189 b2dd03c8-39d4-4d8f-98ff-823fe69b080e 7345d0ef ( nobu 2002-09-29 09:31:58 +0000 3214) fd = -1; 7345d0ef ( nobu 2002-09-2909:31:58 +0000 3215) } 7345d0ef ( nobu 2002-09-29 09:31:58 +0000 3216) else { 4532c0d8 ( usa 2008-07-11 07:18:29 +0000 3217) fd = rb_w32_open_osfhandle(s, O_RDWR|O_BINARY|O_NOINHERIT); 84658471 ( usa 2008-07-24 07:33:00 +0000 3218) if (fd != -1) 84658471 ( usa 2008-07-24 07:33:00 +0000 3219) st_insert(socklist, (st_data_t)s, (st_data_t)0); 84658471 ( usa 2008-07-24 07:33:00 +0000 3220) else commit 4532c0d81ef7aeb6d75d7bbe099682ef90165e29 Author: usa <usa@b2dd03c8-39d4-4d8f-98ff-823fe69b080e> Date: Fri Jul 11 07:18:29 2008 +0000 * win32/win32.c (rb_w32_accept, rb_w32_socket, rb_w32_socketpair): prohibit inheritance of sockets, too. [experimental] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@18024 b2dd03c8-39d4-4d8f-98ff-823fe69b080e c3ce319d ( usa 2004-12-07 15:47:59 +0000 3440) c3ce319d ( usa 2004-12-07 15:47:59 +0000 3441) if (socketpair_internal(af, type, protocol, pair) < 0) c3ce319d ( usa 2004-12-07 15:47:59 +0000 3442) return -1; 4532c0d8 ( usa 2008-07-11 07:18:29 +0000 3443) sv[0] = rb_w32_open_osfhandle(pair[0], O_RDWR|O_BINARY|O_NOINHERIT); 84658471 ( usa 2008-07-24 07:33:00 +0000 3444) if (sv[0] == -1) { 84658471 ( usa 2008-07-24 07:33:00 +0000 3445) closesocket(pair[0]); 84658471 ( usa 2008-07-24 07:33:00 +0000 3446) closesocket(pair[1]); commit 4532c0d81ef7aeb6d75d7bbe099682ef90165e29 Author: usa <usa@b2dd03c8-39d4-4d8f-98ff-823fe69b080e> Date: Fri Jul 11 07:18:29 2008 +0000 * win32/win32.c (rb_w32_accept, rb_w32_socket, rb_w32_socketpair): prohibit inheritance of sockets, too. [experimental] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@18024 b2dd03c8-39d4-4d8f-98ff-823fe69b080e 84658471 ( usa 2008-07-24 07:33:00 +0000 3446) closesocket(pair[1]); 84658471 ( usa 2008-07-24 07:33:00 +0000 3447) return -1; 84658471 ( usa 2008-07-24 07:33:00 +0000 3448) } 4532c0d8 ( usa 2008-07-11 07:18:29 +0000 3449) sv[1] = rb_w32_open_osfhandle(pair[1], O_RDWR|O_BINARY|O_NOINHERIT); 84658471 ( usa 2008-07-24 07:33:00 +0000 3450) if (sv[1] == -1) { 84658471 ( usa 2008-07-24 07:33:00 +0000 3451) rb_w32_close(sv[0]); 84658471 ( usa 2008-07-24 07:33:00 +0000 3452) closesocket(pair[1]); commit 4532c0d81ef7aeb6d75d7bbe099682ef90165e29 Author: usa <usa@b2dd03c8-39d4-4d8f-98ff-823fe69b080e> Date: Fri Jul 11 07:18:29 2008 +0000 * win32/win32.c (rb_w32_accept, rb_w32_socket, rb_w32_socketpair): prohibit inheritance of sockets, too. [experimental] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@18024 b2dd03c8-39d4-4d8f-98ff-823fe69b080e 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4829) /* allocate a C Runtime file handle */ 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4830) RUBY_CRITICAL({ 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4831) h = CreateFile("NUL", 0, 0, NULL, OPEN_ALWAYS, 0, NULL); 150b4efa ( nobu 2010-10-12 14:58:23 +0000 4832) fd = _open_osfhandle((intptr_t)h, 0); 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4833) CloseHandle(h); 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4834) }); 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4835) if (fd == -1) { commit 150b4efa5510d77fdde7b9692b3c391fbde6ac19 Author: nobu <nobu@b2dd03c8-39d4-4d8f-98ff-823fe69b080e> Date: Tue Oct 12 14:58:23 2010 +0000 * win32/win32.c (rb_w32_open_osfhandle, rb_w32_wopen, rb_w32_pipe): use uintptr_t instead of long for win64. * win32/win32.c (socketpair_internal): suppress warnings. * win32/win32.c (ftruncate): use HANDLE instead of long for win64. git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@29467 b2dd03c8-39d4-4d8f-98ff-823fe69b080e 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4948) RUBY_CRITICAL(do { 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4949) ret = 0; 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4950) h = CreateFile("NUL", 0, 0, NULL, OPEN_ALWAYS, 0, NULL); 150b4efa ( nobu 2010-10-12 14:58:23 +0000 4951) fdRead = _open_osfhandle((intptr_t)h, 0); 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4952) CloseHandle(h); 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4953) if (fdRead == -1) { 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4954) errno = EMFILE; commit 150b4efa5510d77fdde7b9692b3c391fbde6ac19 Author: nobu <nobu@b2dd03c8-39d4-4d8f-98ff-823fe69b080e> Date: Tue Oct 12 14:58:23 2010 +0000 * win32/win32.c (rb_w32_open_osfhandle, rb_w32_wopen, rb_w32_pipe): use uintptr_t instead of long for win64. * win32/win32.c (socketpair_internal): suppress warnings. * win32/win32.c (ftruncate): use HANDLE instead of long for win64. git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@29467 b2dd03c8-39d4-4d8f-98ff-823fe69b080e 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4968) 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4969) RUBY_CRITICAL(do { 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4970) h = CreateFile("NUL", 0, 0, NULL, OPEN_ALWAYS, 0, NULL); 150b4efa ( nobu 2010-10-12 14:58:23 +0000 4971) fdWrite = _open_osfhandle((intptr_t)h, 0); 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4972) CloseHandle(h); 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4973) if (fdWrite == -1) { 3c8dca55 ( usa 2008-08-28 12:46:58 +0000 4974) errno = EMFILE; commit 150b4efa5510d77fdde7b9692b3c391fbde6ac19 Author: nobu <nobu@b2dd03c8-39d4-4d8f-98ff-823fe69b080e> Date: Tue Oct 12 14:58:23 2010 +0000 * win32/win32.c (rb_w32_open_osfhandle, rb_w32_wopen, rb_w32_pipe): use uintptr_t instead of long for win64. * win32/win32.c (socketpair_internal): suppress warnings. * win32/win32.c (ftruncate): use HANDLE instead of long for win64. git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@29467 b2dd03c8-39d4-4d8f-98ff-823fe69b080e [ruby-git (trunk)]$
on 2011-01-15 20:00
On Fri, Jan 14, 2011 at 4:47 PM, Stephen Bannasch <stephen.bannasch@deanbrook.org> wrote: >>> Q: why was it written that way to begin with? >>> A: Sorry, I don't know. I guess that the author might not want >>> to put many #ifdef-s into ruby core source. >> >>win32.c and win32.h have a ton of ifdefs in them. What I'm proposing >>would remove many of them, and not add any to the rest of Ruby code. >>Who is the author of the _open_osfhandle code? > > [Sometimes I find playing with git a bit like a crossword puzzle ... there's almost always some way to do what I want ...] > > Here's a summary of all the locations in win32/win32.c that reference open_osfhandle. Thank you for encouraging me to dig. The use of _open_osfhandle I'm referring to dates back to 2000-11-15 when it was added with this helpful comment: C:\ruby\win32>svn log -r 1040 ------------------------------------------------------------------------ r1040 | matz | 2000-11-15 23:24:21 -0800 (Wed, 15 Nov 2000) | 2 lines matz ------------------------------------------------------------------------ I believe, from reading the commit, that the intention was to implement fddup. It seems according to this next commit though, that fddup was removed and the need for it disappeared in ruby 1.7: C:\ruby\win32>svn log -r 2916 ------------------------------------------------------------------------ r2916 | nobu | 2002-10-01 08:26:39 -0700 (Tue, 01 Oct 2002) | 7 lines * ext/socket/socket.c (init_sock): no need for special finalizer, socket descriptor is no longer duplicated in 1.7. [ruby-talk:50732] * win32/win32.c, win32/win32.h (rb_w32_fddup, rb_w32_fdclose): delete.
on 2011-01-17 01:51
Hello,
In message "[ruby-core:34509] Re: Windows sockets and CRT fds"
on Jan.15,2011 06:49:55, <ghazel@gmail.com> wrote:
> Who is the author of the _open_osfhandle code?
"svn ann" will show you it.
> I'm not sure which parts you're saying need to be rewritten for Unix
> and Windows. Currently there needs to be different code for Unix and
> Windows, because a socket handle you might come across on Windows has
> to be adapted to the sort of fd that ruby uses. After the hypothetical
> switch to using regular Windows sockets in Ruby, you could take a
> socket handle from anywhere and select on it in Ruby and have no
> Windows specific code (except inside of Ruby). So, I only see it
> reducing the amount of platform specific code (and bugs) in
> extensions.
If you think so, could you show the patch that shows the concept?
Especially, I want know how about the main body of ruby, ext/socket,
and ext/openssl.
Regards,
on 2011-01-17 02:10
On Sun, Jan 16, 2011 at 4:51 PM, U.Nakamura <usa@garbagecollect.jp> wrote: > If you think so, could you show the patch that shows the concept? > Especially, I want know how about the main body of ruby, ext/socket, > and ext/openssl. Those will not change. It would only affect the Win32 code. If I make a patch against 1.8.7, assuming everyone likes it, is there any hope of it being accepted for 1.8.7 or 1.8.8? The problem is even worse in Ruby 1.9. IS_SOCKET() checks whether the handle is in an internal list or not, so the socket created by libmysql will not appear to be a socket, and there is nothing the extension writer can do to work around that since the list is not exposed in any way.. Thankfully 1.9 has rb_thread_blocking_region so the extension writer can use real select(), but there is no way to write it to be compatible with both versions without an ifdef. I could write a patch which fixes this for 1.9 as well, if the 1.8 patch is accepted. -Greg
on 2011-06-11 12:07
Attempt 2 (perhaps using quotation marks in my message caused it to kill the spam scanner?) > and curb (which uses libcurl), the socket is created by an external > library, and the handle is the SOCKET, not the CRT fd. So is there currently any known way to pass a (library's socket) into select in a ruby extension in windows? I suppose one could just surround a read call in a thread blocking region and get a similar effect, they just couldn't do that for more than one socket per thread I suppose... ref: https://github.com/brianmario/mysql2/issues/142 Thanks! -roger-
Enable email notification
Enable multi-page viewPlease log in before posting. Registration is free and takes only a minute.
Existing account
(Switch to SSL-encrypted connection)
NEW: Do you have a Google/GoogleMail or Yahoo account? No registration required!
Log in with Google account | Log in with Yahoo account
Log in with Google account | Log in with Yahoo account
No account? Register here.