ThreadedX

[Robert Collins]

> -----Original Message-----
> From: Alan Hourihane [ mailto:alanh@fairlite.demon.co.uk ]
> Sent: Thursday, September 20, 2001 6:02 PM
> To: cygwin-xfree@cygwin.com
> Subject: Re: ThreadedX
> 
> 
> Mmm, in the thread Chuck asks that 'ld' shouldn't generate 
> bad load addresses,

The load address's are fine. The impact on cygwin isn't.

> and Chris doesn't know the answer, apart from saying it's a 
> WinXP tradeoff.

This is one of the big tradeoffs cygwin makes. The heap *must* be at the
same address consistently, as must cygwin1.dll itself. (It can be
different across reboots and machines though). 

> As WinXP ships late October and we've got until early-mid 
> November before 4.2.0
> ships. If it shows up problems then, yes, I'll put it back. But it

Problems will occur now. Randomly - per .exe. Consistently - if exe foo
fails it will fail consistently.

> does make the compilation a lot cleaner. Unless the consensus 
> here is to
> put it back now. Suhaib - do you agree with the above ?
> 
> I tend to agree that 'ld' shouldn't be generating bad code though.

The thing is, that it's good code. It just *will not* work reliably with
cygwin.

The problem occurs on *all* cygwin linked binaries. WinXP only affects
this insofar as Chris committed a fix for cygwin - which is in 1.3.3 -
to make it work with XP. The impact affects win95->win2k with cygwin.

It's up to you, but I know that The early KDE beta Ralf worked on had
this problem and thats why he couldn't run with Cygwin 1.3.2 until I
debugged it and let him know.

My 2c is, just do not set a base address. There is little need to do so.
If you are going to do so, write a script to prevent colliding with
cygwin.

Rob
 
> Alan.