A sick idea - mmapped file output

[Zack Weinberg]

On Thu, Nov 02, 2000 at 10:08:59PM -0800, Mark Mitchell wrote:
> 
> Zack --
> 
>   This only affects CPP when *not* integrated with the front-ends,
> right?  In other words, when it is generating a .i file, not when we
> do the fully integrated thing where the front-end just snarfs up the
> tokens directly from the preprocessor?

Correct.  

>   If that's right, I'm not sure that this level of cleverness is
> necessary.  It's only going to be useful in the short term; once we do
> the Right Thing in the front-ends, the preprocessor won't be doing any
> I/O anyhow. 

Neil switched integrated cpplib on by default, today.  "This level of
cleverness," as you say, is already unnecessary.  I brought it up
mainly because it was an interesting experiment.  There are a fair
amount of things out there that use gcc -E directly, so I still do
want to bypass stdio in cpp's output code, but I'll stick to write(2).

It's conceivable that the mmap trick could benefit the output end of
the compiler itself, or of the assembler, especially when emitting
megabytes of debug information.  I'd want to see an actual trace where
it was significant before I did that, though.

[Linux 2.4pre hangs my test process for a couple seconds in exit,
bah.]

zw