Re: [gentoo-user] Per package /bin/sh selection
Mike Gilbert wrote: > > Well, bosh has been tested to work as /bin/sh on Gentoo. > > BTW: On Solaris, bosh is faster than dash (because Solaris has a fully > > working > > vfork()). On Linux bosh is "only" of the same speed as dash since vfork() on > > Linux does not borrow the parents address space description but copies it. > > Is that also true of clone(CLONE_VM|CLONE_VFORK)? Recent versions of > glibc use this to implement the posix_spawn() function. I guess... from reading the man page. There is a simple test to verify my clain for vfork(): bosh may easily be configured for various behavior. If you like a bosh that does not use vfork() just call: cd sh smake clean smake COPTX=-DNO_VFORK If you run tests like: CONFIG_SHELL=$shell $shell ./configure in the directory "autoconf" from the schilytools: http://sourceforge.net/projects/schilytools/files/ (first run "rm *cache"), there is a significant performance win on Solaris for the version that uses vfork() - 20% less kernel CPU time. On Linux, there is a too little win for the same comparison. On Solaris, a vfork() call (check with truss -c) is takes only 1/3 of the time of a fork(), on Linux the win is only 30%. The total performance win for bosh executing configure using vfork() on Solaris is 10-15%, on Linux the total win is only 3%. Note that Linux copied the Solaris concept to implement a normal fork() with copy on write mmaps to the parent. So the only difference between fork() and vfork() is the handling of the address space description for the processes in the kernel. I would be happy, if there could be a way to get the same vfork() performance win on Linux as seen on Solaris. BTW: I forgot to mention that the unit tests for bosh are used this way: - first compile schilytools by running smake from top level. - Then chdir to "sh" and call "smake tests", see the global code in $SRCROOT/tests and the specific code in sh/tests Jörg -- EMail:jo...@schily.net(home) Jörg Schilling D-13353 Berlin joerg.schill...@fokus.fraunhofer.de (work) Blog: http://schily.blogspot.com/ URL: http://cdrecord.org/private/ http://sf.net/projects/schilytools/files/'
Re: [gentoo-user] Per package /bin/sh selection
On Wed, Apr 8, 2020 at 1:17 PM Joerg Schilling wrote: > > Mike Gilbert wrote: > > > Wikipedia says that dash is a fork of NetBSD's ash, and I do see tests > > in their CVS repo. That might be worth looking into. > > > > http://cvsweb.netbsd.org/bsdweb.cgi/src/tests/bin/sh/ > > I see this is the variant from Rihard Elz, so it may make sense. > The original ash is too buggy as a reference. > > > > When ever I change something in bosh, I run the unit tests to verify that > > > I did > > > not introduce a bug. One of the unit tests is to run a configure and > > > compare > > > the results with the results frm a reference shell. > > > > > > BTW: Did you ever think about replacing dash by bosh? > > > > We use bash as the default /bin/sh, but users are free to replace it > > with whatever shell they like, so long as it is reasonably > > POSIX-compliant. Other shells are obviously less tested in Gentoo. > > Well, bosh has been tested to work as /bin/sh on Gentoo. > BTW: On Solaris, bosh is faster than dash (because Solaris has a fully working > vfork()). On Linux bosh is "only" of the same speed as dash since vfork() on > Linux does not borrow the parents address space description but copies it. Is that also true of clone(CLONE_VM|CLONE_VFORK)? Recent versions of glibc use this to implement the posix_spawn() function.
Re: [gentoo-user] Per package /bin/sh selection
Mike Gilbert wrote: > Wikipedia says that dash is a fork of NetBSD's ash, and I do see tests > in their CVS repo. That might be worth looking into. > > http://cvsweb.netbsd.org/bsdweb.cgi/src/tests/bin/sh/ I see this is the variant from Rihard Elz, so it may make sense. The original ash is too buggy as a reference. > > When ever I change something in bosh, I run the unit tests to verify that I > > did > > not introduce a bug. One of the unit tests is to run a configure and compare > > the results with the results frm a reference shell. > > > > BTW: Did you ever think about replacing dash by bosh? > > We use bash as the default /bin/sh, but users are free to replace it > with whatever shell they like, so long as it is reasonably > POSIX-compliant. Other shells are obviously less tested in Gentoo. Well, bosh has been tested to work as /bin/sh on Gentoo. BTW: On Solaris, bosh is faster than dash (because Solaris has a fully working vfork()). On Linux bosh is "only" of the same speed as dash since vfork() on Linux does not borrow the parents address space description but copies it. Jörg -- EMail:jo...@schily.net(home) Jörg Schilling D-13353 Berlin joerg.schill...@fokus.fraunhofer.de (work) Blog: http://schily.blogspot.com/ URL: http://cdrecord.org/private/ http://sf.net/projects/schilytools/files/'
Re: [gentoo-user] Per package /bin/sh selection
On Wed, Apr 8, 2020 at 8:44 AM Joerg Schilling wrote: > > Michael Orlitzky wrote: > > > Thanks, this will be a PITA for a while (again). Another developer had > > patched /bin/dash so that it was effectively broken, to the point where > > ./configure scripts would decide on their own use bash instead (even if > > you set /bin/sh to point to dash). This "fixed" the errors, but meant > > Are there no unit tests for dash? There do not appear to be any tests in the dash source repo. https://git.kernel.org/pub/scm/utils/dash/dash.git/ Wikipedia says that dash is a fork of NetBSD's ash, and I do see tests in their CVS repo. That might be worth looking into. http://cvsweb.netbsd.org/bsdweb.cgi/src/tests/bin/sh/ > When ever I change something in bosh, I run the unit tests to verify that I > did > not introduce a bug. One of the unit tests is to run a configure and compare > the results with the results frm a reference shell. > > BTW: Did you ever think about replacing dash by bosh? We use bash as the default /bin/sh, but users are free to replace it with whatever shell they like, so long as it is reasonably POSIX-compliant. Other shells are obviously less tested in Gentoo.
Re: [gentoo-user] Per package /bin/sh selection
Michael Orlitzky wrote: > Thanks, this will be a PITA for a while (again). Another developer had > patched /bin/dash so that it was effectively broken, to the point where > ./configure scripts would decide on their own use bash instead (even if > you set /bin/sh to point to dash). This "fixed" the errors, but meant Are there no unit tests for dash? When ever I change something in bosh, I run the unit tests to verify that I did not introduce a bug. One of the unit tests is to run a configure and compare the results with the results frm a reference shell. BTW: Did you ever think about replacing dash by bosh? Bosh has the advantage to support multi byte characters. Jörg -- EMail:jo...@schily.net(home) Jörg Schilling D-13353 Berlin joerg.schill...@fokus.fraunhofer.de (work) Blog: http://schily.blogspot.com/ URL: http://cdrecord.org/private/ http://sf.net/projects/schilytools/files/'
Re: [gentoo-user] Per package /bin/sh selection
On Tue, Apr 7, 2020 at 2:48 PM Alessandro Barbieri wrote: > > I already filed bugs here: > > https://bugs.gentoo.org/716504 > https://bugs.gentoo.org/716496 > > I need the workaround for the two above plus this > https://bugs.gentoo.org/714094 In the future, please mention you are using dash as /bin/sh and block bug 714092 when you file bugs about this kind of issue. The fontforge bug report doesn't mention anything about it.
Re: [gentoo-user] Per package /bin/sh selection
On 4/7/20 2:48 PM, Alessandro Barbieri wrote: > I already filed bugs here: > > https://bugs.gentoo.org/716504 > https://bugs.gentoo.org/716496 > > I need the workaround for the two above plus > this https://bugs.gentoo.org/714094 > Thanks, this will be a PITA for a while (again). Another developer had patched /bin/dash so that it was effectively broken, to the point where ./configure scripts would decide on their own use bash instead (even if you set /bin/sh to point to dash). This "fixed" the errors, but meant that everyone who thought he switched to dash was actually still using slow old bash for every build. I undid that patch (so that now dash acts like it does upstream), but as a result there's going to be a bunch of previously-hidden bashisms in ./configure scripts that need to be fixed.
Re: [gentoo-user] Per package /bin/sh selection
I already filed bugs here: https://bugs.gentoo.org/716504 https://bugs.gentoo.org/716496 I need the workaround for the two above plus this https://bugs.gentoo.org/714094 Il Mar 7 Apr 2020, 18:10 Michael Orlitzky ha scritto: > On 4/7/20 11:13 AM, Alessandro Barbieri wrote: > > How can I change /bin/sh in a per package way using /etc/portage/env? > > I need to set bash some packages while I run dash for the system. > > If it's the ./configure script failing, you can set > > CONFIG_SHELL="/bin/bash" > > to override it only in that configure script (but please report a bug > first; configure scripts are supposed to be POSIX sh). > >
Re: [gentoo-user] Per package /bin/sh selection
On 4/7/20 11:13 AM, Alessandro Barbieri wrote: > How can I change /bin/sh in a per package way using /etc/portage/env? > I need to set bash some packages while I run dash for the system. If it's the ./configure script failing, you can set CONFIG_SHELL="/bin/bash" to override it only in that configure script (but please report a bug first; configure scripts are supposed to be POSIX sh).
[gentoo-user] Per package /bin/sh selection
How can I change /bin/sh in a per package way using /etc/portage/env? I need to set bash some packages while I run dash for the system.