[3886] in linux-net channel archive

home help back first fref pref prev next nref lref last post

Re: EVEN MORE 2.0 vs 1.2 and INN config.

daemon@ATHENA.MIT.EDU (Miquel van Smoorenburg)
Sun Jul 28 10:24:49 1996

To: submit-linux-dev-net@ratatosk.yggdrasil.com
From: miquels@drinkel.cistron.nl (Miquel van Smoorenburg)
Date: 	28 Jul 1996 15:53:07 +0200

In article <960727150713_72254.3077_IHC54-1@CompuServe.COM>,
PATRICK MAIN  <72254.3077@compuserve.com> wrote:
>
> But first a suggestion, i noticed your path to overview iss the same as
> distributed. From FAQ4 and the NOV faq they advise placing the overview
> files in a directory separate from the news directory with a standard
> location becoming  /path-to-news-spool/over.view

That's what Debian does, and it _really_ helps if you have a busy news
server.

> 2.  what is the type of an object size ..usually size_t or unsigned int
>     you have SIZE_T    size_t
>   The Faq also has size_t while Debian uses int .... size_t may be right <?>

Yup, this is a bug in the debian config.data.. will be fixed.

> 3.  for file locking , LOCK_STYLE  select NONE, FLOCK,  FCNTL or LOCKF
>     you have FCNTL, Debian also has FCNTL
>     while the FAQ suggests using FLOCK  .... wonder which is best if it
>matters <?>

flock() used to be emulated in the C library since Linux didn't have it.
That's why I chose for fcntl(). It doesn't really matter which one you
use anymore if you have a 2.0.x kernel and libc >= 5.2.18. I'll stick to
fcntl for now though.

> 4.  for use waitpid instead of wait3 <?>
>      you have HAVE_WAITPID      DONT
>     both the FAQ and Debian sets this to DO.....    is DO the correct choice

Again it doesn't matter. Chances are waitpid and wait3 are the same internally
anyway (both emulated in the libc with a call to wait4).

> 5.  for how to fork pick FOR or VFORK
>      you have FORK   vfork ..... and also set  HAVE_VFORK  DO
>      both the FAQ and Debian set FORK for  and HAVE_VFORK DONT
>      what is the difference and is fork the correct choice for 2.0 <?>

Linux doesn't have vfork, and in the BSD compat library headers there's just a

#define vfork fork

> 6.  for "does AF_UNIX use sizeof for the socket size"
>      you have BIND_USE_SIZEOF  DO    as does Debian too!
>      while the FAQ say to et this to DONT... maybe for 2.0 hould be DO  <?>

In Linux this can be DO. It means that for UNIX domain sockets you can just
do a bind(s, (struct sockaddr *)&un, sizeof(un)) instead of all the mucking
around with strlen() etc. It's a bit faster too.

> 7.  everybody agrees that close on exec needs  FCNTL and not IOCTL
>     but for "how should non-blocking I/O be done"
>     you have NBIO_STYLE   FCNTL   as does also DEBIAN
>     while the FAQ says to use IOCTL...maybe FCNTL is correct "NOW" <?>

Again this doesn't matter. The result is the same.

> 8.  for "how to get number of available descripters"
>      select from GETDTAB, GETRLIMIT, SYSCONF, ULIMIT, CONSTANT!
>      you have  FDCOUNT_STYLE   GETRLIMIT  as does Debian!
>      The FAQ says to use SYSCONF as does and INN setup for 1.2.13
>       i have no idea but think maybe 2.0 allows GETRLIMIT now <?>

sysconf() is just a libc wrapper that calls getrlimit as far as I know,
so it's reasonable to just call getrlimit directly.

> 9.  for "what predicate <ctype.h> macros need
>      you have  CTYPE  isXXXXX((c))  which is also suggested by the FAQ
>      DEBIAN uses   (isascii((c))  &&  ixXXXXX((c)))   .... i don't have any

Yeah. I don't think it matters, and I can't remember why I used
(isascii((c))  &&  ixXXXXX((c))).. perhaps it was needed with an older
version of libc. I could take it out but it doesn't hurt I guess.

> 10. that actually does it now...all are using MMAP and UNIX DOMAIN SOCKETS

Great isn't it?

Mike.
-- 
+   Miquel van Smoorenburg   + Cistron Internet Services +  Living is a     |
|  miquels@cistron.nl (SP6)  | Independent Dutch ISP     |   horizontal     |
+ miquels@drinkel.cistron.nl + http://www.cistron.nl/    +      fall        +


home help back first fref pref prev next nref lref last post