🚀 go-pugleaf

kr> Von : kjb@cgl.rmit.oz.au (Fr, 15.05.92 03:30)
kr> Name: Kendall Bennett
kr> Box : RMIT Computer Centre, Melbourne Australia.
kr> MId : <kjb.705897037@godzilla.cgl.citri.edu.au>
kr> RId : <92-05-062@gnusenet.comp.compilers>

kr> thing I did when I first installed DJGPP for my PC (running under DOS
kr> naturally) was to run tests on this program. It took about twice as long
kr> or more to compile under DJGPP, and the resulting program ran
kr> horrendously slow. I set up file approx. 1Mb in size, to be translated,
kr> and the times were...
kr>    BC++: 9.2 seconds
kr>    DJGPP: > 40 seconds

I'm not astonished. Did you look at how DJGPP does I/O? No? Don't look there,
you might get nightmares! :-)

What it does is, first, everything goes through the DJ/GNU/BSD-Libraries; then,
when it arrives at the actual read()/write(), it calls an interrupt which is
intercepted by the driver. The driver, in turn, calls Borland's read()/write()
routines, which may even do a (very slow) cr/lf<->lf translation.

No wonder it's slow.

By the way, last I checked, the same handling is what broke timezone handling
for DJGPP: because it's broken in Borland's Turbo C.

In both cases, double libs make the problem much worse.

MfG Kai