On 2011-02-06 4:02 AM, Philip Prindeville wrote: > On 2/5/11 9:49 AM, Felix Fietkau wrote: >> On 2011-02-05 12:26 PM, Bastian Bittorf wrote: >>> Some weeks ago, i started (undesired) a discussion[1,2] >>> about how and why i think the base-files should get >>> a refurbishment. I send several patches as a proposal, >>> but it was not possible to apply them - the reasons are: >>> >>> - the code changed nearly nothing, only the style >>> - some people mentioned, that the old code was not good, >>> but it worked since years and changes in base are risky >>> - it is not clear, if lowmem-targets are part of openwrt >>> (most of the patches uses lower mem and are faster) >>> - there is no official coding style, that enforces the >>> use of specific ways how to do things >>> - i mixed the reasons, why i wrote the new code >>> >>> Here some words about this: >>> >>> I like openwrt. There are many concepts i want to see >>> in other distros. I'm not satisfied with many scripts >>> and the way how people make a solution. Ofcourse this >>> is my personal opinion, but there are good reasons to >>> speak loud: >>> >>> - clean code (not obfuscated) >>> >>> This should be the most imported thing. >>> It helps to debug. >>> It lowers the entry-border. >>> It makes it easier to change the code and do refactoring. >>> (i have some ideas to modularize the functions/files, at the >>> moment we often include files which we don't need) >>> >>> It does NOT help to make a oneliner and 1000 users asks >>> themself: what does THIS line? [3] >>> >>> But while we are at it: >>> - don't fork, if not needed. examples from /etc/init.d/compache: >>> >>> [ "`cat /proc/swaps | grep 'ramzswap0'`" != "" ]&& swapoff /dev/ramzswap0 >>> >>> better use: >>> fgrep -q "ramzswap0" /proc/swaps || ... >> I fully agree with changes like this. >> >>> - use function_names to say what you want to do >>> - use better shell builtins, instead of invoking another language >>> (e.g. awk, sed, lua) - only if there is no other option >> I disagree with that. awk and sed are part of most typical shell >> scripts, they've been around for ages and if awk and sed make something >> expressible in one line instead of 10 without being unnecessarily >> complex, then I consider that an improvement over writing weird >> while/read/case loops. >> >>> - using of shell builtins, leads to lower mem usage and are (mostly) faster >>> >>> In the first discussion, i explained that awk is big and should not >>> be in base-files. During experiments it turned out, that these are only >>> some kilobytes. But the real reasons are explained above. >> I think many of these scripts are simply prototypes - trying to evolve >> them to better suit devices with 8 MB RAM is mostly a pointless >> excercise, as many of them will be replaced with compact and much more >> efficent c-code soon-ish. We will integrate new infrastructure for >> starting/stopping/monitoring services and for generating config files >> soon, which could work for even complex services without writing any >> complex shell scripts - simply by having a proper description language. > > That's true, but at the same time the cost of a fork()/exec() pair > is non-trivial, not to mention all of the cacheline misses that will incur. > If you can keep things in the same process space, that's a big win. > > The "hairiness" of the scripting can be abstracted away by using > function wrappers in shell. awk/sed calls do not result in an exec() because they're handled by busybox internally. All it does is a fork, and thanks to copy-on-write, most of the memory stays shared.
Not that I don't agree with the general goal of making things more efficient, but as mentioned earlier, a lot of this code will be replaced with non-shell stuff anyway. - Felix _______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
