RE: [avrdude-dev] Verify errors,

[Alex Shepherd]

> Right, I've looked through the archived for the thread called
> program error?
>
> It seems to end with alex saying he had Win2K SP2 on a Dual
> Celeron 466,
> I'm running a 1.1 GHz Athlon but the same OS.

Hmmm... that would tend to suggest my dual processor being the problem. That
just leaves the timers in cygwin or some other subtle flaw!

> What was the conclusion, apart from doubling the timing on
> the processor
> defenition?

We have not reached a conclusion yet. Not many people seem to have this
problem and it probably works fine on the code maintainers PCs - there in
lies the problem. :(

I have had a play around with the 4.0.0 source code bundle and can
.configure and recompile it under the cygwin install I have but my system is
not up to the mark to do a CVS checkout and build. It has problems building
the configure scripts.

I have made several changes already and will continue to play with it. I
wondered if there was an issue with timing and so someone came up with an
alternative timing mechanism using the select() function, but so far I have
not been able to apply this patch locally. I guess I will have to do it
manually. :(

I did read somewhere that the cygwin people had done some changes to the
implementation of usleep and family to now call nanosleep() which must have
been recently (earlier this year) added to cygwin. I was not able to find
any comments about why they changed to this so I wonder if there was an
issue that someone has fixed with the new nanosleep() implementation.

My observations are that the verify error is always at the beginning of a 64
byte block so I now suspect that the code is not waiting long enough between
blocks. Can you note the verify error locations and preferably have a look
at the location in the AVR FLASH with PonyProg and see what the value is in
FLASH? I am pretty sure mine were all still at FF which is a bit suspicious.

The overall programming time under Win2K seems to be faster compared to
running under Linux on the exact same hardware. Under Win2K it takes about 8
seconds to program my bit of test code (6.8k long) but on Linux it take
about 11-12 seconds. So this tells me that either delays are shorter on
Win2K or somehow the code path on Linux is much longer - maybe doing the LPT
port IOCTLs.

Cheers

Alex