On Wed, Apr 20, 2016 at 07:46:38AM +0200, Anselm R Garbe wrote: > On 20 April 2016 at 05:17, Sylvain BERTRAND <sylvain.bertr...@gmail.com> > wrote: > > For my personnal use, I needed a small http server. All "mini" http servers > > out > > there I had a look were, IMHO, bloaty (SDK included). > > Did you also look at http://git.suckless.org/quark/tree/quark.c ?
Nope. Missed that one. :( As a member of the suckless collective: shame on me. > quark is fork() based (and POSIX compliant in that sense) though, not > epoll_wait() based. So quark will perform accordingly, but it supports > a bit more HTTP features (incl. cgi execution), but its SLOC is > considerable equal to lnanohttpd. lnanohttp is enough for my personal use (static content). cgi is already too much. It's easy to add clone(fork/thread)/cgi/(and more). In ulinux/patterns there is code of a "network server" which does the cloning stuff, even "pre-cloning" (pre-fork/pre-thread spawning). Just need to adapt the code. But it's way beyond my needs. > > lnanohttp is really small (including dependencies and SDK), straight on > > linux kernel > > syscalls with a thin layer. Tested only on x86 and with a gcc/binutils > > toolchain for the moment (planing to buy an armv8 raspberry board). > > > > Can be use easily as a base for a beefier http server. > > Some quick remarks: it uses a lot of #define's (all over the place) > which I consider very weird. Also for type declarations it uses > #define's of types which seem to be very weird as well. (e.g. #define > sl ulinux_sl). The defines are used to manage namespaces. ulinux thin layer has its own namespace in order to be mixed with other namespages (libc/posix/etc). > Also is relying on -errno checks all over the place really secure? I > would rather check for -1 and compare with errno directly. Or is it a > linux pattern I wasn't aware of? quark pulls the posix libc. lnanohttp has 0 deps: this is straight linux syscalls programming, there are no libc syscall wrappers. Now, I need to find bugs... :) -- Sylvain
