[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: lynx-dev lynx2.8.2dev.7
From: |
dickey |
Subject: |
Re: lynx-dev lynx2.8.2dev.7 |
Date: |
Mon, 23 Nov 1998 22:55:22 -0500 (EST) |
> > 1998-11-23 (2.8.2dev.7)
> > * add/use new functions HTAA_UidToName(), HTAA_NameToUid(), HTAA_GidToName()
> > and HTAA_NameToGid() to hide details of code which uses pwd.h and grp.h,
> > as well as to cache the returned user/group names, improving performance
> > in the dired screen - TD
> HTAAProt.c fails to build with DJGPP (probably no groups on DOS?) -
> lines 688,713,717, etc - "redefining pointer to incomplete type"
ok (I missed the NOUSERS ifdef).
> > * modify HTCheckForInterrupt() to check for interrupt no more than one per
> > second, since this check is comparatively slow - TD
>
> I usually test lynx on slow 386: no difference in speed detected,
> but 1 sec delay while scrolling in partial mode is annoying (with any CPU).
well, on the unix side, it's a big hit (probably slows Lynx down by a
factor of 2-3 when just reading a directory). I noticed that it was
slow starting up, and (as usual ;-) compared it to 2.7.2, then ran both
with strace. About half of all the system calls were being made as
as result of the interrupt-check. (This also slows down retrieval of
files).
I would think that a 1 second delay in stopping output wouldn't be
a bad thing - it's a cheap solution (and the alternatives all appear
expensive).
At the same time, I saw that on Linux (and SunOS) at least, each of those
calls to getpwnam, etc., caused the system to reread the whole /etc/passwd
(or /etc/group). Probably Solaris caches the information. (We may make
the caching by Lynx of user and group info an optional feature - it's a
simple ifdef).
--
Thomas E. Dickey
address@hidden
http://www.clark.net/pub/dickey