Date: Thu, 7 Oct 2004 22:40:32 GMT From: Mark Gooderum <mark@verniernetworks.com> To: freebsd-threads@FreeBSD.org Subject: Re: threads/72429: threads blocked in stdio (fgets, etc) are not cancellable in 5.3 (works in 4.x) Message-ID: <200410072240.i97MeWLj098967@freefall.freebsd.org>
next in thread | raw e-mail | index | archive | help
The following reply was made to PR threads/72429; it has been noted by GNATS. From: Mark Gooderum <mark@verniernetworks.com> To: Daniel Eischen <deischen@gdeb.com> Cc: freebsd-gnats-submit@freebsd.org, freebsd-threads@freebsd.org, archie@dellroad.org Subject: Re: threads/72429: threads blocked in stdio (fgets, etc) are not cancellable in 5.3 (works in 4.x) Date: Thu, 07 Oct 2004 17:36:44 -0500 This is a multi-part message in MIME format. --------------080409080603080600050704 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Uncle. Okay - you're right, sigh. It's not so much that we're using fgets() as many/most C based parsing libraries use stdio, bleh. As for why libc_r is cancellable... In 4.x _foo() sets the cancellation state and calls __foo() (for read, write, et.al), in 5.3 it's reversed. -- Mark >On Thu, 7 Oct 2004, Mark Gooderum wrote: > > > >>But this is a major change in behavior from FreeBSD 4 and also a >>difference from Linux. I'm not a Linux bigot at all but there is a >>recurring theme that XX threaded apps works on Linux but is unstable on >>FreeBSD and these sort of major behavior deltas contribute to the >>perception of FreeBSD threading as unstable by some. >> >> > >If you want to be portable, you should be using select() or >poll(). It's not like there is no portable way of doing >what you want, and in fact relying on fgets() to be cancellable >is not portable. > > > >>In fact it means that any thread doing blocking stdio is uncancellable - >>the standard may not require it but many applications might expect it. >>Given that the functionality was there in 4.x and lost in 5.x I'd call >>it a regression. >> >> > >fgets() was not supposed to be cancellable in 4.x. If it >is, it is not by intention. Anything that calls _foo (single >underscore versions of system calls) is intentionally doing >it that way to avoid entering undesired cancellation points >(and blocking points in the case of libc_r). There are other >uses of _read() within libc and those may not want to be >cancellation points. > >The overall design of libc is that all internal uses of system >calls use the single underscore versions and let the threads >library override them if it wants. _foo() is not supposed to >be cancellable, and, in the case of libc_r, is also not supposed >to allow the process to block when hit (libc_r converts _read() >and _write() to poll(), then switches to another thread). I >don't know why fgets() ends up as a cancellation point in libc_r, >but it shouldn't be by design -- it is a bug. > > > --------------080409080603080600050704 Content-Type: text/html; charset=us-ascii Content-Transfer-Encoding: 7bit <!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"> <html> <head> <meta http-equiv="Content-Type" content="text/html;charset=ISO-8859-1"> <title></title> </head> <body text="#000000" bgcolor="#ffffff"> Uncle. Okay - you're right, sigh.<br> <br> It's not so much that we're using fgets() as many/most C based parsing libraries use stdio, bleh.<br> <br> As for why libc_r is cancellable...<br> <br> In 4.x _foo() sets the cancellation state and calls __foo() (for read, write, et.al), in 5.3 it's reversed. <br> --<br> Mark<br> <br> <blockquote type="cite" cite="midPine.GSO.4.43.0410071721330.1700-100000@sea.ntplx.net"> <pre wrap="">On Thu, 7 Oct 2004, Mark Gooderum wrote: </pre> <blockquote type="cite"> <pre wrap="">But this is a major change in behavior from FreeBSD 4 and also a difference from Linux. I'm not a Linux bigot at all but there is a recurring theme that XX threaded apps works on Linux but is unstable on FreeBSD and these sort of major behavior deltas contribute to the perception of FreeBSD threading as unstable by some. </pre> </blockquote> <pre wrap=""><!----> If you want to be portable, you should be using select() or poll(). It's not like there is no portable way of doing what you want, and in fact relying on fgets() to be cancellable is not portable. </pre> <blockquote type="cite"> <pre wrap="">In fact it means that any thread doing blocking stdio is uncancellable - the standard may not require it but many applications might expect it. Given that the functionality was there in 4.x and lost in 5.x I'd call it a regression. </pre> </blockquote> <pre wrap=""><!----> fgets() was not supposed to be cancellable in 4.x. If it is, it is not by intention. Anything that calls _foo (single underscore versions of system calls) is intentionally doing it that way to avoid entering undesired cancellation points (and blocking points in the case of libc_r). There are other uses of _read() within libc and those may not want to be cancellation points. The overall design of libc is that all internal uses of system calls use the single underscore versions and let the threads library override them if it wants. _foo() is not supposed to be cancellable, and, in the case of libc_r, is also not supposed to allow the process to block when hit (libc_r converts _read() and _write() to poll(), then switches to another thread). I don't know why fgets() ends up as a cancellation point in libc_r, but it shouldn't be by design -- it is a bug. </pre> </blockquote> </body> </html> --------------080409080603080600050704--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200410072240.i97MeWLj098967>