Date: Thu, 12 Jun 1997 19:59:34 +0400 (MSD)
From: "Alexander V. Lukyanov" <lav AT video DOT yars DOT free DOT net>
Message-Id: <199706121559.TAA26813@video.yars.free.net>
To: billc AT blackmagic DOT tait DOT co DOT nz, eliz AT is DOT elta DOT co DOT il
Cc: djgpp-workers AT delorie DOT com, robert DOT hoehne AT mathematik DOT tu-chemnitz DOT de
Subject: Re: Latest stub
Precedence: bulk

Eli:
> Yes, that's easy, now that I'm aware of that.  But try to explain this 
> one: several people have reported (it's even in the FAQ) that stubediting 
> ld.exe to 64KB makes linking *much* faster when the libraries are on a 
> networked drive.

I think I can explain that, as I hacked stdio and experimented with that.
ld looks through the library, doing lots of fseek's. As stdio default
buffer size defaults to transfer buffer size, the size of chunk read
is determined by it. But ld often needs just a few kilobytes (common
size of object modules) the other data can be unneeded and will be
discarded on the next fseek. (It was even worse sometime - fseek used to
always discard data evein if it could be reused).

This all explains the strange diagram of ld performance depending on
TB size, which I have posted here sometime ago.

   Alexander.