Re: [gentoo-dev] Network configuration and bash
On Friday 09 February 2007, Roy Marples wrote: On Thu, 8 Feb 2007 14:49:57 -0700 Daniel Robbins [EMAIL PROTECTED] wrote: In other words: busybox + single rcS file = fastest and simplest, smallest, best for very small filesystems, not as flexible bash + gentoo baselayout = most flexible, biggest, slower, best for feature-rich systems busybox + gentoo baselayout = ? FreeBSD sh + Gentoo baselayout = cold boot in around 4 seconds Going to multi-user from single user after a boot is under 2 seconds (times measured from when init starts rc - the difference is probably because the all my local mounts are still mounted) I have this running on a 2Ghz P4 Laptop right now. Admittedly, no network scripts are started expect for the loopback interface, but all default scripts + openvpn, ssh, dnsmasq, metalog and vixie-cron are started. Ladies and gentlemen, this has always been about one thing - speed. Ever since I got my 300Mhz Sparc64 to play around with FreeBSD, I've realised that baselayout + bash is just too damn slow. I think that's worth it. If that's what you want, don't use bash in the first place. I would agree that using bash for parsing is a pain in the but Daniel is right in that you're not going to be able to maintain posix compatibility. If you find an acceptable way to add the functionality to the network configuration files, it is ok, but sacrificing usability over an unmaintainable improvement doesn't work. If you want to speed up boot, the dependency generation is probably what's eating most time. Paul -- Paul de Vrieze Gentoo Developer Mail: [EMAIL PROTECTED] Homepage: http://www.devrieze.net pgpS9X5o4MDYx.pgp Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
For speed, there are a *lot* of changes/improvements that could be made. What I would like to see is the ability to get to a login prompt before startup is actually completed. Have all the non-essential startup stuff run in the background. Yes, this would require a sophisticated system since you could log in and start something that requires something that has not yet been started, and it would need to understand and deal with this appropriately and gracefully. But I think it would be cool. The perception of speed is based on how long you have to wait before you can do stuff. -Daniel On 2/16/07, Paul de Vrieze [EMAIL PROTECTED] wrote: On Friday 09 February 2007, Roy Marples wrote: On Thu, 8 Feb 2007 14:49:57 -0700 Daniel Robbins [EMAIL PROTECTED] wrote: In other words: busybox + single rcS file = fastest and simplest, smallest, best for very small filesystems, not as flexible bash + gentoo baselayout = most flexible, biggest, slower, best for feature-rich systems busybox + gentoo baselayout = ? FreeBSD sh + Gentoo baselayout = cold boot in around 4 seconds Going to multi-user from single user after a boot is under 2 seconds (times measured from when init starts rc - the difference is probably because the all my local mounts are still mounted) I have this running on a 2Ghz P4 Laptop right now. Admittedly, no network scripts are started expect for the loopback interface, but all default scripts + openvpn, ssh, dnsmasq, metalog and vixie-cron are started. Ladies and gentlemen, this has always been about one thing - speed. Ever since I got my 300Mhz Sparc64 to play around with FreeBSD, I've realised that baselayout + bash is just too damn slow. I think that's worth it. If that's what you want, don't use bash in the first place. I would agree that using bash for parsing is a pain in the but Daniel is right in that you're not going to be able to maintain posix compatibility. If you find an acceptable way to add the functionality to the network configuration files, it is ok, but sacrificing usability over an unmaintainable improvement doesn't work. If you want to speed up boot, the dependency generation is probably what's eating most time. Paul -- Paul de Vrieze Gentoo Developer Mail: [EMAIL PROTECTED] Homepage: http://www.devrieze.net -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Fri, 16 Feb 2007 22:20:54 +0100 Paul de Vrieze [EMAIL PROTECTED] wrote: If that's what you want, don't use bash in the first place. I would agree that using bash for parsing is a pain in the but Daniel is right in that you're not going to be able to maintain posix compatibility. If you find an acceptable way to add the functionality to the network configuration files, it is ok, but sacrificing usability over an unmaintainable improvement doesn't work. Provided that /bin/sh is /bin/bash then we can maintain the config_* directive. If you want to speed up boot, the dependency generation is probably what's eating most time. And that was solved in baselayout-1.13 where we already have the dep resolver in C. Thanks Roy -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Fri, 16 Feb 2007 16:44:15 -0700 Daniel Robbins [EMAIL PROTECTED] wrote: For speed, there are a *lot* of changes/improvements that could be made. Yes, re-writing the backend in C whilst keeping /etc/init.d/ and /etc/conf.d/ as shell scripts is a good start :) What I would like to see is the ability to get to a login prompt before startup is actually completed. Have all the non-essential startup stuff run in the background. Yes, this would require a sophisticated system since you could log in and start something that requires something that has not yet been started, and it would need to understand and deal with this appropriately and gracefully. But I think it would be cool. The perception of speed is based on how long you have to wait before you can do stuff. Oh, that's easy. I can do this RC_LOGIN_SERVICES=foo bar and rc will wait until sysinit, boot have completed and then wait for foo AND bar to complete. The side effect is at the point we login no service einfo stuff is shown. The obvious downside is that the user needs to state which services are required for logging in, but they would know that much better than us. Of course, that relies on RC_PARALLEL_STARTUP being set. Thanks Roy -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
Mike Frysinger napsal(a): On Thursday 08 February 2007, Roy Marples wrote: The actual scripts themselves can be re-worked if they need to be - this problem only when the arrays are used in config files. i guess my point was i think we really need to be consistent here ... either arrays are OK for init.d scripts or they're not OK did you get a chance to see how hard it would be to integrate the bash array code ? sys-apps/855resolution and sys-apps/915resolution use arrays in config files, and there's not much way around it ATM if you want to replace multiple BIOS modes... snip # to replace more than one mode do something like this #replace=( # 4d 1280 768 24 # 5c 1400 1050 16 #) /snip -- Best regards, Jakub Moc mailto:[EMAIL PROTECTED] GPG signature: http://subkeys.pgp.net:11371/pks/lookup?op=getsearch=0xCEBA3D9E Primary key fingerprint: D2D7 933C 9BA1 C95B 2C95 B30F 8717 D5FD CEBA 3D9E ... still no signature ;) signature.asc Description: OpenPGP digital signature
Re: [gentoo-dev] Network configuration and bash
Quoting Roy Marples [EMAIL PROTECTED]: FreeBSD sh + Gentoo baselayout = cold boot in around 4 seconds Going to multi-user from single user after a boot is under 2 seconds (times measured from when init starts rc - the difference is probably because the all my local mounts are still mounted) I have this running on a 2Ghz P4 Laptop right now. Admittedly, no network scripts are started expect for the loopback interface, but all default scripts + openvpn, ssh, dnsmasq, metalog and vixie-cron are started. And what are the numbers with bash? Just curious. This message was sent using IMP, the Internet Messaging Program. -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
Jakub Moc wrote: Mike Frysinger napsal(a): On Thursday 08 February 2007, Roy Marples wrote: The actual scripts themselves can be re-worked if they need to be - this problem only when the arrays are used in config files. i guess my point was i think we really need to be consistent here ... either arrays are OK for init.d scripts or they're not OK did you get a chance to see how hard it would be to integrate the bash array code ? sys-apps/855resolution and sys-apps/915resolution use arrays in config files, and there's not much way around it ATM if you want to replace multiple BIOS modes... snip # to replace more than one mode do something like this #replace=( # 4d 1280 768 24 # 5c 1400 1050 16 #) How about this? replace= '4d 1280 768 24' '5c 1400 1050 16' signature.asc Description: OpenPGP digital signature
Re: [gentoo-dev] Network configuration and bash
On Fri, 09 Feb 2007 01:03:04 -0800
Donnie Berkholz [EMAIL PROTECTED] wrote:
Jakub Moc wrote:
Mike Frysinger napsal(a):
On Thursday 08 February 2007, Roy Marples wrote:
The actual scripts themselves can be re-worked if they need to be
- this problem only when the arrays are used in config files.
i guess my point was i think we really need to be consistent
here ... either arrays are OK for init.d scripts or they're not OK
did you get a chance to see how hard it would be to integrate the
bash array code ?
sys-apps/855resolution and sys-apps/915resolution use arrays in
config files, and there's not much way around it ATM if you want to
replace multiple BIOS modes...
snip
# to replace more than one mode do something like this
#replace=(
# 4d 1280 768 24
# 5c 1400 1050 16
#)
How about this?
replace=
'4d 1280 768 24'
'5c 1400 1050 16'
Actually, that may work better than my delimited with ; approach.
We could then do
eval set -- ${replace}
for x in $@ ; do
done
Thanks
Roy
--
gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Thu, 8 Feb 2007 20:27:06 -0500 Mike Frysinger [EMAIL PROTECTED] wrote: On Thursday 08 February 2007, Ned Ludd wrote: Please read over what's been talked about elsewhere in this thread. He is not trying to break existing functionality at all. Only extend it to be posix aware (additionally) erm, no ... our code is a superset of POSIX, so technically yes he is breaking existing functionality and doing the quite opposite of extending -mike True, but remember that this only applies for when /bin/sh isn't bash. So this will mainly affect FreeBSD and embedded, but it at least gives them a better chance of working from the get-go if baselayout at least supports it. Roy -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Fri, 09 Feb 2007 17:27:58 +0900 Georgi Georgiev [EMAIL PROTECTED] wrote: Quoting Roy Marples [EMAIL PROTECTED]: FreeBSD sh + Gentoo baselayout = cold boot in around 4 seconds Going to multi-user from single user after a boot is under 2 seconds (times measured from when init starts rc - the difference is probably because the all my local mounts are still mounted) I have this running on a 2Ghz P4 Laptop right now. Admittedly, no network scripts are started expect for the loopback interface, but all default scripts + openvpn, ssh, dnsmasq, metalog and vixie-cron are started. And what are the numbers with bash? Just curious. About a second or two slower. ie not much difference at all in the grand scheme of things. But that's to be expected as most of the gain has come from moving existing bash code to C - except for the init script's themselves. Also note that when this gets released at some point, Linux will be slower as it has to load udev. FreeBSD has loaded devfs and done its magic before init, so there will always be some skew as to which is faster/better. But that's why *I* am doing it. Roy -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Fri, Feb 09, 2007 at 09:29:04AM +, Roy Marples wrote:
On Fri, 09 Feb 2007 01:03:04 -0800
Donnie Berkholz [EMAIL PROTECTED] wrote:
Jakub Moc wrote:
Mike Frysinger napsal(a):
On Thursday 08 February 2007, Roy Marples wrote:
The actual scripts themselves can be re-worked if they need to be
- this problem only when the arrays are used in config files.
i guess my point was i think we really need to be consistent
here ... either arrays are OK for init.d scripts or they're not OK
did you get a chance to see how hard it would be to integrate the
bash array code ?
sys-apps/855resolution and sys-apps/915resolution use arrays in
config files, and there's not much way around it ATM if you want to
replace multiple BIOS modes...
snip
# to replace more than one mode do something like this
#replace=(
# 4d 1280 768 24
# 5c 1400 1050 16
#)
How about this?
replace=
'4d 1280 768 24'
'5c 1400 1050 16'
Actually, that may work better than my delimited with ; approach.
We could then do
eval set -- ${replace}
for x in $@ ; do
done
That works with your ; approach too, as well as with the
newline-separated approach:
replace=
4d 1280 768 24
5c 1400 1050 16
IFS=';
'
#set -f
set -- ${replace}
#set +f
unset IFS
for x ; do
done
pgprrqWQ2pIf0.pgp
Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
On Fri, 9 Feb 2007 10:51:29 +0100
Harald van Dijk [EMAIL PROTECTED] wrote:
That works with your ; approach too, as well as with the
newline-separated approach:
replace=
4d 1280 768 24
5c 1400 1050 16
IFS=';
'
#set -f
set -- ${replace}
#set +f
unset IFS
for x ; do
done
Yes, but consider this
iwpriv_ESSID=
set AuthMode=WPAPSK;
set EncrypType=TKIP;
set WPAPSK=yourpasskey
One issue is that any valid character is allowed in WPAPSK, including ;
$ foo='one '\'' two' \
'three ; four' \
'five \ six' \
$ eval set -- ${foo} ; for x in $@ ; do echo $x; done
one ' two
three ; four
five \ six
$
Demonstrates that we can achieve multi-line arrays like so, without the
need to set shell opts or alter IFS
Thanks
Roy
--
gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Fri, 9 Feb 2007 12:05:52 +0100 Harald van Dijk [EMAIL PROTECTED] wrote: One issue is that any valid character is allowed in WPAPSK, including ; Good point, but excluding newline, right? I can't try it, but rt2500's own settings file format does not allow for newline in WPAPSK. No it doesn't, but the actual specification for IEEE 802.11 allows ANY character in SSID and keys - newlines, tabs, ;, etc. Just because the rt2500 proprietary config doesn't allow it doesn't mean we shouldn't either. Roy -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Fri, Feb 09, 2007 at 11:49:05AM +, Roy Marples wrote: On Fri, 9 Feb 2007 12:05:52 +0100 Harald van Dijk [EMAIL PROTECTED] wrote: One issue is that any valid character is allowed in WPAPSK, including ; Good point, but excluding newline, right? I can't try it, but rt2500's own settings file format does not allow for newline in WPAPSK. No it doesn't, but the actual specification for IEEE 802.11 allows ANY character in SSID and keys - newlines, tabs, ;, etc. Okay, thanks for the info. The only easy alternative I can think of is using printf %b, but IIRC printf is not universally supported yet, so yeah, eval is probably the only thing that will offer all the needed options. I still hope you can get some decent sanity checks for it. Just because the rt2500 proprietary config doesn't allow it doesn't mean we shouldn't either. Proprietary? pgp512noaYxaT.pgp Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
On Wed, 7 Feb 2007 23:42:14 -0500 Mike Frysinger [EMAIL PROTECTED] wrote: On Wednesday 07 February 2007, Roy Marples wrote: In the current code I'm running it's only the network stuff that uses arrays. If you're thinking about /sbin/functions.sh, well that can stay as bash as it's not used by baselayout anymore. some init.d scripts use arrays as well -mike Do we know which ones? The actual scripts themselves can be re-worked if they need to be - this problem only when the arrays are used in config files. Thanks Roy -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Thursday 08 February 2007, Roy Marples wrote: Mike Frysinger [EMAIL PROTECTED] wrote: On Wednesday 07 February 2007, Roy Marples wrote: In the current code I'm running it's only the network stuff that uses arrays. If you're thinking about /sbin/functions.sh, well that can stay as bash as it's not used by baselayout anymore. some init.d scripts use arrays as well Do we know which ones? grep for it :p netmount for sure right now The actual scripts themselves can be re-worked if they need to be - this problem only when the arrays are used in config files. i guess my point was i think we really need to be consistent here ... either arrays are OK for init.d scripts or they're not OK did you get a chance to see how hard it would be to integrate the bash array code ? -mike pgpc7BSebhvSu.pgp Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
On Thu, 8 Feb 2007 13:01:08 -0500 Mike Frysinger [EMAIL PROTECTED] wrote: some init.d scripts use arrays as well Do we know which ones? grep for it :p netmount for sure right now Well, netmount is baselayout, so that will kinda be done by default :) i guess my point was i think we really need to be consistent here ... either arrays are OK for init.d scripts or they're not OK I say no, as I'd like eventually anything in /etc/init.d,conf.d to be useable by any shell. It's not especially hard to make existing scripts in the tree work either. I'm happy to do this myself. did you get a chance to see how hard it would be to integrate the bash array code ? Integrate into what? You mean integrate into other shells? Roy -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
I sort of missed this conversation, so apologies in advance if this has already been covered, but wanted to say that gentoo's initscripts are generally not suited for embedded systems. So making baselayout busybox-compatible doesn't seem to be worth the disruption and headaches it would cause. It would be disruptive for gentoo developers who would need to be extra-careful in maintaining their initscripts to ensure busybox compatibility. Not to mention the potential disruption for users. If you are building an embedded system using busybox, then generally you will want a single /etc/init.d/rcS script that starts all the stuff you need. -Daniel On 2/8/07, Mike Frysinger [EMAIL PROTECTED] wrote: On Thursday 08 February 2007, Roy Marples wrote: Mike Frysinger [EMAIL PROTECTED] wrote: On Wednesday 07 February 2007, Roy Marples wrote: In the current code I'm running it's only the network stuff that uses arrays. If you're thinking about /sbin/functions.sh, well that can stay as bash as it's not used by baselayout anymore. some init.d scripts use arrays as well Do we know which ones? grep for it :p netmount for sure right now The actual scripts themselves can be re-worked if they need to be - this problem only when the arrays are used in config files. i guess my point was i think we really need to be consistent here ... either arrays are OK for init.d scripts or they're not OK did you get a chance to see how hard it would be to integrate the bash array code ? -mike -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Thursday 08 February 2007, Roy Marples wrote: Mike Frysinger [EMAIL PROTECTED] wrote: did you get a chance to see how hard it would be to integrate the bash array code ? Integrate into what? You mean integrate into other shells? mmm i thought you were looking to parse the config files directly from baselayout and not via shell, so i was referring to integrating arrays into baselayout -mike pgp6VXSWIz5rl.pgp Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
On Thu, 2007-02-08 at 13:23 -0700, Daniel Robbins wrote: I sort of missed this conversation, so apologies in advance if this has already been covered, but wanted to say that gentoo's initscripts are generally not suited for embedded systems. So making baselayout busybox-compatible doesn't seem to be worth the disruption and headaches it would cause. Please read over what's been talked about elsewhere in this thread. He is not trying to break existing functionality at all. Only extend it to be posix aware (additionally) It would be disruptive for gentoo developers who would need to be extra-careful in maintaining their initscripts to ensure busybox compatibility. Not to mention the potential disruption for users. There is no reason this has to be disruptive to the users who don't care about this functionality. If you are building an embedded system using busybox, then generally you will want a single /etc/init.d/rcS script that starts all the stuff you need. As somebody that's had to hand write many of those kinds of scripts. A single rcS is not very ideal. Our init scripts are in fact mostly usable by busybox. Granted there are a few special special cases, but then Roy is offering to update those for free. One of the larger problems really boils down to many packages provide default init.d scripts and these expect the existing baselayout only. That will be a bigger feat to deal with later on down the road. -Daniel On 2/8/07, Mike Frysinger [EMAIL PROTECTED] wrote: On Thursday 08 February 2007, Roy Marples wrote: Mike Frysinger [EMAIL PROTECTED] wrote: On Wednesday 07 February 2007, Roy Marples wrote: In the current code I'm running it's only the network stuff that uses arrays. If you're thinking about /sbin/functions.sh, well that can stay as bash as it's not used by baselayout anymore. some init.d scripts use arrays as well Do we know which ones? grep for it :p netmount for sure right now The actual scripts themselves can be re-worked if they need to be - this problem only when the arrays are used in config files. i guess my point was i think we really need to be consistent here ... either arrays are OK for init.d scripts or they're not OK did you get a chance to see how hard it would be to integrate the bash array code ? -mike -- Ned Ludd [EMAIL PROTECTED] Gentoo Linux -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On 2/8/07, Ned Ludd [EMAIL PROTECTED] wrote: As somebody that's had to hand write many of those kinds of scripts. A single rcS is not very ideal. Our init scripts are in fact mostly usable by busybox. Granted there are a few special special cases, but then Roy is offering to update those for free. One of the larger problems really boils down to many packages provide default init.d scripts and these expect the existing baselayout only. That will be a bigger feat to deal with later on down the road. Developers will then need to test their init.d scripts to ensure that they are compatible with busybox. This is asking a lot of work of people just so you can use Gentoo's initscripts for something they are not really ideal for. Any time a script is updated a new rev of a package is required, and this does impact users and will cause packages to be rebuilt when a user does emerge -u. So I think this should be weighed against the potential benefits of baselayout + busybox. If you are targetting something smaller than 32MB, then maybe busybox is appropriate. But you are trying to go really small, then you probably don't want all the extra junk in our initscripts. And if you are _not_ trying to go really small, then put bash in your filesystem, not busybox, and the initscripts will work. If bash isn't fast enough from a boot time perspective, then the gentoo initscripts certainly aren't going to be fast enough either. In other words: busybox + single rcS file = fastest and simplest, smallest, best for very small filesystems, not as flexible bash + gentoo baselayout = most flexible, biggest, slower, best for feature-rich systems busybox + gentoo baselayout = ? I think that in 99 out of 100 cases, if you have room for baselayout, then you probably have room for bash too. And in 99 out of 100 cases, if you can deal with the load time of baselayout, then you can deal with the overhead that might be incurred from having bash. I'm just pointing out that it's not an obviously good combination. In the grand scheme of things, maybe it's not a great use of developer resources. Or, maybe I'm wrong and it is a great idea. Personally I think that baselayout + busybox may be cool, but adding an aftermarket sunroof to your car can be cool too. But that doesn't mean it's worth the effort :) Really, it's hard for me to imagine many scenarios where you really need the flexibility of baselayout but can't squeeze in bash. And I have a pretty good imagination. -Daniel -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Thu, 2007-02-08 at 14:49 -0700, Daniel Robbins wrote: On 2/8/07, Ned Ludd [EMAIL PROTECTED] wrote: As somebody that's had to hand write many of those kinds of scripts. A single rcS is not very ideal. Our init scripts are in fact mostly usable by busybox. Granted there are a few special special cases, but then Roy is offering to update those for free. One of the larger problems really boils down to many packages provide default init.d scripts and these expect the existing baselayout only. That will be a bigger feat to deal with later on down the road. Developers will then need to test their init.d scripts to ensure that they are compatible with busybox. This is asking a lot of work of people just so you can use Gentoo's initscripts for something they are not really ideal for. I don't think anybody can/would expect that. They don't test for hardened or uClibc now before stabilizing a pkg. What would be nice to see is if a maintainer is offered a posix compliant init.d script that they merge it or allow those to be merged for a pkg they maintain as long as it does not degrade functionality. Any time a script is updated a new rev of a package is required, and this does impact users and will cause packages to be rebuilt when a user does emerge -u. So I think this should be weighed against the potential benefits of baselayout + busybox. busybox is not Roys underlying goal as far as I understand. I think he simply mentioned it as an example of another group who wishes to unify efforts and have an interest in getting away from arrays where feasible. If you are targetting something smaller than 32MB, then maybe busybox is appropriate. But you are trying to go really small, then you probably don't want all the extra junk in our initscripts. And if you are _not_ trying to go really small, then put bash in your filesystem, not busybox, and the initscripts will work. If bash isn't fast enough from a boot time perspective, then the gentoo initscripts certainly aren't going to be fast enough either. In other words: busybox + single rcS file = fastest and simplest, smallest, best for very small filesystems, not as flexible bash + gentoo baselayout = most flexible, biggest, slower, best for feature-rich systems busybox + gentoo baselayout = ? It's been done in the past by end users. Before there were only about 4 changes needed to make it work. That all changed when bash arrays were introduced. I think that in 99 out of 100 cases, if you have room for baselayout, then you probably have room for bash too. And in 99 out of 100 cases, if you can deal with the load time of baselayout, then you can deal with the overhead that might be incurred from having bash. I'm just pointing out that it's not an obviously good combination. In the grand scheme of things, maybe it's not a great use of developer resources. Or, maybe I'm wrong and it is a great idea. His time and resources. His itch Personally I think that baselayout + busybox may be cool, but adding an aftermarket sunroof to your car can be cool too. But that doesn't mean it's worth the effort :) I don't think those who are not interested in this will be burden by any extra effort. Worst case is maybe getting a bug assigned to you which offers a posix replacement/update for the default init.d a pkg you maintain might provide. Really, it's hard for me to imagine many scenarios where you really need the flexibility of baselayout but can't squeeze in bash. And I have a pretty good imagination. baselayout is only about a half of a meg these days and probably getting smaller/faster with the addition of the multicall rc/runscript work he has been doing. Adding bash also requires ncurses which in turn mostly requires having a c++ aware compiler or using the nocxx,minimal flags. Even with those flags enabled I'm seeing 3M going to ncurses+bash. So I can for sure see the benefits. Also for a moment lets stop and think. Some XYZ update breaks ncurses/bash. Supporting this gives us a nice alternative way to still boot our boxes for rescue using ash or another shell which might not have such big deps. -- Ned Ludd [EMAIL PROTECTED] Gentoo Linux -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Thu, 8 Feb 2007 14:49:57 -0700 Daniel Robbins [EMAIL PROTECTED] wrote: In other words: busybox + single rcS file = fastest and simplest, smallest, best for very small filesystems, not as flexible bash + gentoo baselayout = most flexible, biggest, slower, best for feature-rich systems busybox + gentoo baselayout = ? FreeBSD sh + Gentoo baselayout = cold boot in around 4 seconds Going to multi-user from single user after a boot is under 2 seconds (times measured from when init starts rc - the difference is probably because the all my local mounts are still mounted) I have this running on a 2Ghz P4 Laptop right now. Admittedly, no network scripts are started expect for the loopback interface, but all default scripts + openvpn, ssh, dnsmasq, metalog and vixie-cron are started. Ladies and gentlemen, this has always been about one thing - speed. Ever since I got my 300Mhz Sparc64 to play around with FreeBSD, I've realised that baselayout + bash is just too damn slow. I think that's worth it. Thanks Roy -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Friday 09 February 2007, Ned Ludd wrote: baselayout is only about a half of a meg these days and probably getting smaller/faster with the addition of the multicall rc/runscript work he has been doing. Adding bash also requires ncurses which in turn mostly requires having a c++ aware compiler or using the nocxx,minimal flags. Even with those flags enabled I'm seeing 3M going to ncurses+bash. So I can for sure see the benefits. Also for a moment lets stop and think. Some XYZ update breaks ncurses/bash. Supporting this gives us a nice alternative way to still boot our boxes for rescue using ash or another shell which might not have such big deps. From where I stand I can see Ned's point just fine: I'm interested in both having a sane baselayout that doesn't break on bash upgrade (I've seen the breakage with 3.1, 3.2.. I also masked bash 3.1 while Roy fixed baselayout for that version), and in a baselayout that can run on medium embedded systems (and not just for fun, trust me), so I wouldn't dismiss Roy's work as unneeded and/or not useful to anyone. He's not going to waste someone else's time, and as he said there will be compatibility with current configuration files, I don't think there's any downside to users. -- Diego Flameeyes Pettenò - http://farragut.flameeyes.is-a-geek.org/ Gentoo/Alt lead, Gentoo/FreeBSD, Video, Sound, ALSA, PAM, KDE, CJK, Ruby ... pgpg13D9POXYc.pgp Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
He's not going to waste someone else's time, and as he said there will be compatibility with current configuration files, I don't think there's any downside to users. FWIW, speaking as a user, I value stability over speed. But if I have a promise of stability (i.e. my current configs will still work), I have no problem with Roy scratching his speed itch. :) Thanks, Marty -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
Roy Marples wrote: Hi List As some of you may be aware, I've started work on baselayout-2 which is basically re-tooling it in C. One of the side goals is to eliminate the need for using bash. You'll be pleased to know that it's working well enough to boot Gentoo/FreeBSD. Hm... So, to free baselayout of forcing bash down our throats I/we am/are looking at re-writing our network setup, including configuration. What about adding array support in the C retool? What about adding this particular feature in busybox and just use it? lu - against having a completely pointless transition... -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
Roy Marples wrote: This email is about network configuration. Before I joined Gentoo, network configuration was done in bash arrays like so (note, that the variable name was changed in baselayout-1.11) ifconfig_eth0=( 10.1.1.1 netmask 255.255.255.0 10.1.1.2 netmask 255.255.255.0 ) This is all well and good, but only bash and zsh can use it. Infact I'm guilty as I embraced arrays and used them as the basis for our entire network scripts. They have also served us well, but as they are also array based they will have to be re-written too. So, to free baselayout of forcing bash down our throats I/we am/are looking at re-writing our network setup, including configuration. How about leaving baselayout depending on bash and make baselayout-lite do whatever you want instead? Such an upgrade will be a major pita for our entire user base, unless of course you make conf.d/net syntax transition as easy as running dispatch-conf. signature.asc Description: OpenPGP digital signature
Re: [gentoo-dev] Network configuration and bash
On Tue, 6 Feb 2007 22:27:46 -0500 Mike Frysinger [EMAIL PROTECTED] wrote: On Tuesday 06 February 2007, Roy Marples wrote: No, I'm just interesting in killing bash array requirements in the network config file. you need to kill them everywhere then ... network config isnt the only file that utilizes arrays -mike In the current code I'm running it's only the network stuff that uses arrays. If you're thinking about /sbin/functions.sh, well that can stay as bash as it's not used by baselayout anymore. Roy -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tue, Feb 06, 2007 at 11:09:11PM +, Roy Marples wrote: On Tue, 6 Feb 2007 21:14:49 + Ciaran McCreesh [EMAIL PROTECTED] wrote: Why? What's wrong with requiring a shell that supports various features beyond what POSIX specifies? Granted, choice of shell is good, but not if it's at the expense of functionality or ease of use. H, just how many features should a config file have beyond the setting of variables? I make reasonable use of some of the advanced stuff already: http://www.orbis-terrarum.net/~robbat2/conf.d-multihomed -- Robin Hugh Johnson Gentoo Linux Developer E-Mail : [EMAIL PROTECTED] GnuPG FP : 11AC BA4F 4778 E3F6 E4ED F38E B27B 944E 3488 4E85 pgpU3IyoJ7lQL.pgp Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
Roy Marples ha scritto:
Hi List
As some of you may be aware, I've started work on baselayout-2 which is
basically re-tooling it in C. One of the side goals is to eliminate the
need for using bash. You'll be pleased to know that it's working well
enough to boot Gentoo/FreeBSD.
[...]
what about to keep compatibility with at least this form:
ifconfig_eth0=(
10.1.1.{1,2} netmask 255.255.255.0
)
and maybe hooks to external code to be executed
{pre,post}ip_link_set
{pre,post}ip_addr_{add,del}
etc...
A fast look at Bash code show that i should be maintenable
~/tmp/bash-3.2 $ wc braces.* array*
630 2208 14019 braces.c
932 2843 19720 array.c
111 498 3817 array.h
837 2896 20407 arrayfunc.c
58 300 2289 arrayfunc.h
2568 8745 60252 total
Regards,
Francesco
--
gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Wednesday 07 February 2007, Roy Marples wrote: In the current code I'm running it's only the network stuff that uses arrays. If you're thinking about /sbin/functions.sh, well that can stay as bash as it's not used by baselayout anymore. some init.d scripts use arrays as well -mike pgpimDq3QEneT.pgp Description: PGP signature
[gentoo-dev] Network configuration and bash
Hi List As some of you may be aware, I've started work on baselayout-2 which is basically re-tooling it in C. One of the side goals is to eliminate the need for using bash. You'll be pleased to know that it's working well enough to boot Gentoo/FreeBSD. Now, this email isn't about the merits of bash, nor the fact that it's in base system profile so we can use it anyway, blah blah blah. embedded has a vested interest in not using bash and I have a personal interest as Gentoo/FreeBSD on Sparc64 takes a very long time to boot. This email is about network configuration. Before I joined Gentoo, network configuration was done in bash arrays like so (note, that the variable name was changed in baselayout-1.11) ifconfig_eth0=( 10.1.1.1 netmask 255.255.255.0 10.1.1.2 netmask 255.255.255.0 ) This is all well and good, but only bash and zsh can use it. Infact I'm guilty as I embraced arrays and used them as the basis for our entire network scripts. They have also served us well, but as they are also array based they will have to be re-written too. So, to free baselayout of forcing bash down our throats I/we am/are looking at re-writing our network setup, including configuration. Who's got any bright ideas for a new config then? Lets brain storm! Thanks Roy -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tue, Feb 06, 2007 at 06:17:16PM +, Roy Marples wrote: Hi List Hi Roy, [snip] So, to free baselayout of forcing bash down our throats I/we am/are looking at re-writing our network setup, including configuration. Who's got any bright ideas for a new config then? Lets brain storm! It is a good format and easy to write and parse in C, why change it ? Just tell people to stop treating config files like bash scripts and everything will be fine. Or am I missing something? - ferdy -- Fernando J. Pereda Garcimartín Gentoo Developer (Alpha,net-mail,mutt,git) 20BB BDC3 761A 4781 E6ED ED0B 0A48 5B0C 60BD 28D4 pgpVCA2ayOize.pgp Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
On Tue, 6 Feb 2007 19:27:40 +0100 Fernando J. Pereda [EMAIL PROTECTED] wrote: On Tue, Feb 06, 2007 at 06:17:16PM +, Roy Marples wrote: Hi List Hi Roy, [snip] So, to free baselayout of forcing bash down our throats I/we am/are looking at re-writing our network setup, including configuration. Who's got any bright ideas for a new config then? Lets brain storm! It is a good format and easy to write and parse in C, why change it ? Just tell people to stop treating config files like bash scripts and everything will be fine. Or am I missing something? Well, the stuff in /etc/init.d/ and /etc/conf.d will stay as shell scripts - just not bash specific scripts. The stuff that handles our networking maybe written in A.N. Other-Language (Mrs.), but keeping /etc/conf.d/net readable by a shell script does have advantages. Thanks Roy -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tue, 6 Feb 2007 19:09:26 + Roy Marples [EMAIL PROTECTED] wrote: | The stuff that handles our networking maybe written in A.N. | Other-Language (Mrs.), but keeping /etc/conf.d/net readable by a shell | script does have advantages. He didn't say make it not readable by a shell. He was suggesting making it in some format that is readable both by C and the shell -- that is to say, a subset of shell. -- Ciaran McCreesh Mail: ciaranm at ciaranm.org Web : http://ciaranm.org/ Paludis, the secure package manager : http://paludis.pioto.org/ signature.asc Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
Forgive me if this is just noise, but I just wanted to say I agree fully with ferdy. As I was reading Roy's email, and I looked at the net config sample he had in there, I thought well, what's actually wrong with this? Keeping it as is has the advantage that an upgrade/downgrade cycle wouldn't change much in functionality based on config, which is pretty good (ie, backwards compatibility). In this case, I'm not sure legacy is all that bad, simply because it's expressive and concise and easily understood :) thanks, -- Seemant Kulleen Developer, Gentoo Linux -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tuesday 06 February 2007, Michael Hanselmann wrote: XML! Actually, no. For me, libconfig[1] turned out to be very easy to work with. Its config file format is easy to write by hand and the parser resides in the library. On a simialr note there's libconfuse[1], which uses one of the most common formats available: ini-like :) [1] http://www.nongnu.org/confuse/ -- Diego Flameeyes Pettenò - http://farragut.flameeyes.is-a-geek.org/ Gentoo/Alt lead, Gentoo/FreeBSD, Video, Sound, ALSA, PAM, KDE, CJK, Ruby ... pgpYWPdBzIKuX.pgp Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
On Tue, 06 Feb 2007 14:18:43 -0500 Seemant Kulleen [EMAIL PROTECTED] wrote: Forgive me if this is just noise, but I just wanted to say I agree fully with ferdy. As I was reading Roy's email, and I looked at the net config sample he had in there, I thought well, what's actually wrong with this? Cannot be parsed by shells other than bash or zsh Keeping it as is has the advantage that an upgrade/downgrade cycle wouldn't change much in functionality based on config, which is pretty good (ie, backwards compatibility). In this case, I'm not sure legacy is all that bad, simply because it's expressive and concise and easily understood :) So you're saying we patch other shells to support bash arrays? -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tue, 6 Feb 2007 20:31:03 + Roy Marples [EMAIL PROTECTED] wrote: | On Tue, 6 Feb 2007 19:27:15 + | Ciaran McCreesh [EMAIL PROTECTED] wrote: | | On Tue, 6 Feb 2007 19:09:26 + Roy Marples [EMAIL PROTECTED] | wrote: | | The stuff that handles our networking maybe written in A.N. | | Other-Language (Mrs.), but keeping /etc/conf.d/net readable by a | shell | script does have advantages. | | He didn't say make it not readable by a shell. He was suggesting | making it in some format that is readable both by C and the shell -- | that is to say, a subset of shell. | | Right, and bash arrays are not shell | http://www.opengroup.org/onlinepubs/009695399/toc.htm Sure they're shell. They're just not POSIX. -- Ciaran McCreesh Mail: ciaranm at ciaranm.org Web : http://ciaranm.org/ Paludis, the secure package manager : http://paludis.pioto.org/ signature.asc Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
On Tue, 2007-06-02 at 20:34 +, Roy Marples wrote: Keeping it as is has the advantage that an upgrade/downgrade cycle wouldn't change much in functionality based on config, which is pretty good (ie, backwards compatibility). In this case, I'm not sure legacy is all that bad, simply because it's expressive and concise and easily understood :) So you're saying we patch other shells to support bash arrays? What so wrong with bash? Apart from the fact that its not included in the unmodified base system of FreeBSD? Last I checked, we add lots of Gentooish stuff already. And that its not friendly to tiny embedded systems, which dont use a complex and powerful init system like Gentoo's (but prefer very simple ash scripts). -- Olivier Crête [EMAIL PROTECTED] Gentoo Developer -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tue, 6 Feb 2007 20:58:52 + Ciaran McCreesh [EMAIL PROTECTED] wrote: | Right, and bash arrays are not shell | http://www.opengroup.org/onlinepubs/009695399/toc.htm Sure they're shell. They're just not POSIX. Maybe I should have been more clear. Anything in /etc/conf.d/ should be able to be read by a POSIX compliant shell. This means no arrays. At this point in time, the only thing that uses arrays in /etc/conf.d/net - or any other bashisms or non POSIX shell stuff is out net config. Hence the need to change it. -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tue, 06 Feb 2007 16:03:50 -0500 Olivier Crete [EMAIL PROTECTED] wrote: What so wrong with bash? Unsuited to an init system that wants to work everywhere, like embedded systems. Also, being tied to one shell causes problems when that shell breaks. Witness baselayout problems regarding bash-3.0, 3.1 and 3.2 Lastly, in my first email I said Now, this email isn't about the merits of bash, nor the fact that it's in base system profile so we can use it anyway, blah blah blah. embedded has a vested interest in not using bash and I have a personal interest as Gentoo/FreeBSD on Sparc64 takes a very long time to boot. I'm assuming you did't read that bit :) Thanks Roy -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tue, 6 Feb 2007 21:04:59 + Roy Marples [EMAIL PROTECTED] wrote: | On Tue, 6 Feb 2007 20:58:52 + | Ciaran McCreesh [EMAIL PROTECTED] wrote: | | Right, and bash arrays are not shell | | http://www.opengroup.org/onlinepubs/009695399/toc.htm | | Sure they're shell. They're just not POSIX. | | Maybe I should have been more clear. | | Anything in /etc/conf.d/ should be able to be read by a POSIX | compliant shell. This means no arrays. Why? What's wrong with requiring a shell that supports various features beyond what POSIX specifies? Granted, choice of shell is good, but not if it's at the expense of functionality or ease of use. -- Ciaran McCreesh Mail: ciaranm at ciaranm.org Web : http://ciaranm.org/ Paludis, the secure package manager : http://paludis.pioto.org/ signature.asc Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
On Tue, 2007-06-02 at 21:11 +, Roy Marples wrote: On Tue, 06 Feb 2007 16:03:50 -0500 Olivier Crete [EMAIL PROTECTED] wrote: What so wrong with bash? Unsuited to an init system that wants to work everywhere, like embedded systems. Also, being tied to one shell causes problems when that shell breaks. Witness baselayout problems regarding bash-3.0, 3.1 and 3.2 Lastly, in my first email I said Now, this email isn't about the merits of bash, nor the fact that it's in base system profile so we can use it anyway, blah blah blah. embedded has a vested interest in not using bash and I have a personal interest as Gentoo/FreeBSD on Sparc64 takes a very long time to boot. I'm assuming you did't read that bit :) Actually I did, but I really dont see what kind of embedded system would be complex enough to require the Gentoo init system, yet not enough to use bash. And I don't see why g/fbsd sparc64 is problematic? Or is it just that you have a 1995 vintage system and you want to run Gentoo on it? -- Olivier Crête [EMAIL PROTECTED] Gentoo Developer -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tue, 6 Feb 2007 21:11:31 + Roy Marples [EMAIL PROTECTED] wrote: | Now, this email isn't about the merits of bash, nor the fact that | it's in base system profile so we can use it anyway, blah blah blah. | embedded has a vested interest in not using bash and I have a | personal interest as Gentoo/FreeBSD on Sparc64 takes a very long | time to boot. | | I'm assuming you did't read that bit :) I think it's more that you're expected to justify *why* the bash requirement is so bad, given the cost of changing. -- Ciaran McCreesh Mail: ciaranm at ciaranm.org Web : http://ciaranm.org/ Paludis, the secure package manager : http://paludis.pioto.org/ signature.asc Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
On Tue, 6 Feb 2007 21:14:49 + Ciaran McCreesh [EMAIL PROTECTED] wrote: Why? What's wrong with requiring a shell that supports various features beyond what POSIX specifies? Granted, choice of shell is good, but not if it's at the expense of functionality or ease of use. H, just how many features should a config file have beyond the setting of variables? -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tue, 6 Feb 2007 21:28:04 +
Ciaran McCreesh [EMAIL PROTECTED] wrote:
I think it's more that you're expected to justify *why* the bash
requirement is so bad, given the cost of changing.
1) Lack of choice.
Gentoo is all about giving the user choice. baselayout even supports
other init systems when requested.
2) Speed.
Bash is one of the slowest shells around for looping.
However, it also requires less forking due to it's nice built-ins. This
does of course only work with bash and not other shells.
3) What's the cost of *not* changing away from bash?
I would say that bash is the best shell around in terms of features and
ease of use, however that is not without cost. That cost is new bash
versions consistently breaking baselayout, ebuilds and configure files.
4) Size.
Because bash has all these nice features it's large, hence unsuitable
for low memory or low disk space environments.
5) I'm *just* talking about config files here.
If users want to run bash, that's fine and I won't stop them. They can
also use bash in their init script if they so wish as I plan to support
something like so
depend() {
shell bash
}
And voila, problem solved. Of course, that's just an idea I just had.
However, I also think that baselayout provided services should not
require bash for the above reasons, hence the need for a new config.
Hopefully I've justified this enough :)
Thanks
Roy
--
gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tue, 6 Feb 2007 19:09:26 + Roy Marples [EMAIL PROTECTED] wrote: The stuff that handles our networking maybe written in A.N. Other-Language (Mrs.), but keeping /etc/conf.d/net readable by a shell script does have advantages. You need to define what shell (or subset) you want to parse it. 'sh' itself varies from platform to platform. The one we have is a softlink to bash so doesn't make any difference for Gentoo Linux except for limiting what can be written. I just tried variable redirection for example (which can be used to implement pseudo-arrays without using eval) - I was surprised it works in sh here - dunno if it works in BSD sh (doesn't on busybox). What you have on FreeBSD may be different from what's on Solaris. Perhaps busybox sh might be a practical set to choose, for basic posix compliance. You could simply do something like: ifconfig_eth0=\ 10.1.1.1 netmask 255.255.255.0;\ 10.1.1.2 netmask 255.255.255.0 which means standard shell interpretation doesn't lose information, even if it's actually normally parsed by something else (chose ';' as a separator since ':' is used in ipv6 strings). It seems to me that the problem you're trying to solve, is the implementation of baselayout on restricted systems. Reducing all init.d/conf.d and so on to a common denominator for everyone isn't necessarily the best way forward. A different approach could be to provide more than one baselayout; one for large systems, where expecting to have bash available isn't such a big deal, and one for limited systems, restricted to busybox-standard sh. Actually I kinda assumed that's what baselayout-lite was all about... -- Kevin F. Quinn signature.asc Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
On Tuesday 06 February 2007, Roy Marples wrote: Hopefully I've justified this enough :) justified what ? this thread started off killing bash array requirements in the network config file and now it looks like your killing the shell everywhere ? to be honest, unless the new code is really good, it's still going to be not suitable for real embedded systems ... sure, i run Gentoo on all my embedded machines, but lets be honest, these arent even close to real embedded -mike pgpIvD4ML3jyD.pgp Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
On Tuesday 06 February 2007, Kevin F. Quinn wrote: You need to define what shell (or subset) you want to parse it. 'sh' itself varies from platform to platform. our standard has always (always is relative here; let's say current) been the bash superset of POSIX ... if a request comes up where someone wants to change code because it breaks in their shell but the code in question is POSIX compliant, the answer is simple: blow me if the request is reasonable like stop using some bashism in favor of the POSIX form and the change isnt invasive, then sure we'll generally make the change -mike pgpG0S2atZLMK.pgp Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
Or perhaps, Something a little more explicit might work? In instances such as the ifconfig lines, it's to create eth0 aliases (such as eth0:0), so perhaps that could look like: ifconfig_eth0:0 = 10.1.1.1 netmask 255.255.255.0 ifconfig_eth0:1 = 10.1.1.2 netmask 255.255.255.0 For uses that really do require a list (such as modules), something like: modules = wpa_supplicant, ifconfig or redefinition: modules = wpa_supplicant modules = ifconfig might work? If the comma separated syntax is acceptable, it may be that brackets around it are acceptable too, in which case the bash syntax is fine. If the normal shells will still require *some* form of processing to deal with the list, could they not be given processing to deal with lists that bash already has builtin? Mike 5:) -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tue, 2007-02-06 at 19:42 -0500, Mike Frysinger wrote: On Tuesday 06 February 2007, Kevin F. Quinn wrote: You need to define what shell (or subset) you want to parse it. 'sh' itself varies from platform to platform. our standard has always (always is relative here; let's say current) been the bash superset of POSIX ... if a request comes up where someone wants to change code because it breaks in their shell but the code in question is POSIX compliant, the answer is simple: blow me if the request is reasonable like stop using some bashism in favor of the POSIX form and the change isnt invasive, then sure we'll generally make the change /me likes the direction Roy is heading with this. -mike -- Ned Ludd [EMAIL PROTECTED] Gentoo Linux -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tue, 6 Feb 2007 23:26:32 +
Roy Marples [EMAIL PROTECTED] wrote:
On Tue, 6 Feb 2007 21:28:04 +
Ciaran McCreesh [EMAIL PROTECTED] wrote:
I think it's more that you're expected to justify *why* the bash
requirement is so bad, given the cost of changing.
1) Lack of choice.
Gentoo is all about giving the user choice. baselayout even supports
other init systems when requested.
Surely you would provide choice by providing different baselayout
packages; one tailored for embedded systems that only have busybox, one
for general purpose use, etc. That's what the virtual is for, isn't it?
2) Speed.
Bash is one of the slowest shells around for looping.
However, it also requires less forking due to it's nice built-ins.
This does of course only work with bash and not other shells.
Restricting everything to 'sh' I think is more likely to slow things
down than anything else. Apart from the forking issue that you
mention, builtins are different - '[' for example is about 30% slower
than '[[' in bash (which is what's implementing sh on Gentoo Linux). I
wonder how much time would be saved on Gentoo Linux by replacing [ with
[[ throughout the init scripts; maybe a percentage point?
If there really is a big speed penalty from using bash on BSD compared
to the native shell, wouldn't it be better to supply a Gentoo/FreeBSD
baselayout? They don't have to be completely independent; smart use of
the vcs would allow you to share scripts between baselayout branches,
with specific variants where it makes a difference.
3) What's the cost of *not* changing away from bash?
I would say that bash is the best shell around in terms of features
and ease of use, however that is not without cost. That cost is new
bash versions consistently breaking baselayout, ebuilds and configure
files.
w.r.t new bash versions, we should certainly be conservative in marking
new versions stable. It's worth noting the breakage isn't always
100% the fault of bash. The recent problem with '=~' and quoting for
example is down to glibc behaving differently to everyone else's libc
when it comes to accepting quoted characters for the regex interfaces
(a point where the POSIX standard is open to interpretation).
4) Size.
Because bash has all these nice features it's large, hence unsuitable
for low memory or low disk space environments.
But you only get one bash text image in memory at any one time (~825k).
So space isn't a real issue, except on small-memory systems.
5) I'm *just* talking about config files here.
If users want to run bash, that's fine and I won't stop them. They can
also use bash in their init script if they so wish as I plan to
support something like so
depend() {
shell bash
}
Making that sort of requirement explicit is a good idea. I wouldn't use
'depend()', as in init scripts it's quite cleanly only to do with the
order of services. You could make it an option to runscript:
#!/bin/runscript --shell=/bin/bash
or something along those lines - the shebang is clearly all about how
the script is executed, and the shell used falls nicely into that.
And voila, problem solved. Of course, that's just an idea I just had.
However, I also think that baselayout provided services should not
require bash for the above reasons, hence the need for a new config.
I think the argument for conf.d files is better than that for init.d
scripts; you could have multiple baselayout setups that share conf.d
file formats.
--
Kevin F. Quinn
signature.asc
Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
On Tue, 6 Feb 2007 19:39:03 -0500 Mike Frysinger [EMAIL PROTECTED] wrote: On Tuesday 06 February 2007, Roy Marples wrote: Hopefully I've justified this enough :) justified what ? this thread started off killing bash array requirements in the network config file and now it looks like your killing the shell everywhere ? No, I'm just interesting in killing bash array requirements in the network config file. Once that is done, then you can use whatever shell you wish as far as the init scripts allow, which will hopefully be all of them in portage. to be honest, unless the new code is really good, it's still going to be not suitable for real embedded systems ... sure, i run Gentoo on all my embedded machines, but lets be honest, these arent even close to real embedded -mike I cannot argue that as I have no embedded systems of which to speak. However, some users and devs who use embedded systems have expressed an interest so I'll let them argue as they will. Thanks Roy -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tue, 6 Feb 2007, Roy Marples wrote: This email is about network configuration. Before I joined Gentoo, network configuration was done in bash arrays like so (note, that the variable name was changed in baselayout-1.11) ifconfig_eth0=( 10.1.1.1 netmask 255.255.255.0 10.1.1.2 netmask 255.255.255.0 ) This is all well and good, but only bash and zsh can use it. That's not true. Only bash and zsh can *source* it. Aside from the fact that most other shells don't have any way of storing the results of reading such a file in an accessible way, that format is easy enough to parse with a bit of sed. Who's got any bright ideas for a new config then? Lets brain storm! If it's necessary to change at all, I'd vote for sections in square brackets, applying to everything until the next square bracket, and lines with a key, followed by a ':' or '=', followed by optional whitespace, followed by any number of values, optionally in quotes with backslash-escaped quotes and backslashes, separated by whitespace, with the value lists for duplicate keys being concatenated. It's not hard to parse (assuming, again, that you have some way to store the results), users coming from Windows can write it as ini files, users who like Java can write it as resource files, and old-school Unix types can write RFC822 headers. And they're all mutually comprehensible if the parser is just reasonably lenient. -Daniel *This .sig left intentionally blank* -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
On Tuesday 06 February 2007, Roy Marples wrote: No, I'm just interesting in killing bash array requirements in the network config file. you need to kill them everywhere then ... network config isnt the only file that utilizes arrays -mike pgpj1mYCh71a3.pgp Description: PGP signature
Re: [gentoo-dev] Network configuration and bash
Roy Marples [EMAIL PROTECTED] writes: H, just how many features should a config file have beyond the setting of variables? In the case of networking, the ability to define the functions for the various hooks. In most systems these will not be needed, but where policy routing etc is used the hooks need to be defined somewhere, but not necessarily the configuration file as they are at the moment. -- gentoo-dev@gentoo.org mailing list
Re: [gentoo-dev] Network configuration and bash
Roy Marples napsal(a):
This email is about network configuration. Before I joined Gentoo,
network configuration was done in bash arrays like so (note, that the
variable name was changed in baselayout-1.11)
ifconfig_eth0=(
10.1.1.1 netmask 255.255.255.0
10.1.1.2 netmask 255.255.255.0
)
This is all well and good, but only bash and zsh can use it. Infact I'm
guilty as I embraced arrays and used them as the basis for our entire
network scripts. They have also served us well, but as they are also
array based they will have to be re-written too.
To be honest, I hate the idea of rewriting the network config; being
able to specify things like
config_eth0=( 10.1.1.{1..10}/24 brd + )
(yes, it really expands correctly and works!) is extremely useful and
something you don't find in any other distro out there last time I've
checked. You can assign tons of IP address on a single line, or just
with a couple of lines if you need different subnets, instead of going
thru the boring and error-prone task of specifying every single address
separately.
Why kill such useful features just to get rid of arrays and bash?
Kevin F. Quinn napsal(a):
Surely you would provide choice by providing different baselayout
packages; one tailored for embedded systems that only have busybox,
one for general purpose use, etc. That's what the virtual is for,
isn't it?
+1
--
Best regards,
Jakub Moc
mailto:[EMAIL PROTECTED]
GPG signature:
http://subkeys.pgp.net:11371/pks/lookup?op=getsearch=0xCEBA3D9E
Primary key fingerprint: D2D7 933C 9BA1 C95B 2C95 B30F 8717 D5FD CEBA 3D9E
... still no signature ;)
signature.asc
Description: OpenPGP digital signature
