Ingmar Blonk wrote:
> Hello,
>
> I'm trying to get printf working in combination with some
> floats/doubles.
> I use the CodeSourcery toolchain (Lite 2010q1-188 for ARM EABI) and I'm
> developing for an lpc1768.
>
> Everything works fine. I implemented the needed syscalls, I'm able to
> use printf with integers. But when I try to print a float or double
> printf tries to allocate lots of memory.
> It first allocates a few bytes (for internal housekeeping?). Immediately
> following 200+ bytes, following a block of 4K, following another 4K,
> then even 8K!

I guess you check requested memory-size in your sbrk. As far as I 
understand the malloc implementation in newlib requests memory in 4096 
bytes pages. So this should it should be "normal" that you see steps of 
"4k".

> I thought it maybe was trying to check how much memory it could
> allocate, so I set errno to ENOMEM, but this resulted in a HardFault.
>
> The strange thing is that the first printf stops allocating after the
> first/second 4K and just works perfect. But the second printf call needs
> way to much memory.

I guess the 2nd printf-call uses FP formating. If yes, this is also kind 
of "normal", newlib's FP format needs a lot of memory (program and RAM)

> I also tried yagarto. Then it just allocates a little memory, but it's
> printing totally nonsense and doubles aren't working with yagarto.

You may ask the Yagarto packager about this (Michael Fischer).

> Has someone an idea why it's allocating so much memory and/or knows how
> to prevent it from doing so.

If you look into newlib's source-code (file mallocr.c) you can see that 
with the SMALL_MEMORY macro defined the page-size is 128 bytes instead 
of 4096 bytes. Maybe printf with FP just needs little more than 4K so 
there  might be enough RAM for the FP formating when using a newlib 
configured and built with SMALL_MEMORY defined.

You may try to avoid stdio at all and use alternative code instead 
(search for rprintf, xprintf, dtostrf...)