Re: [Fwd: AW: [avr-gcc-list] Improving in leaps and skips!]

[Björn Haase]

Am Dienstag, 1. März 2005 18:16 schrieb E. Weddington:
> Björn Haase wrote:
> > FYI: I presently observe around 491 testsuite failures only, most of
> > them due to missing nested function support and missing support of
> > library functions like sprintf() for operations that are not important
> > for avr.
>
> That's interesting. avr-libc has the sprintf() function. Do those tests
> need to be conditionally rewritten so they will properly execute for
> avr-libc?
>
> Eric
I would have to look after it in detail for each individual case, but in most 
cases that I have investigated so far, it was an error that showed up when 
passing, e.g., a long double to sprintf or, e.g.,  an unsigned long long. ... 
Instead of handling also these cases in the avr-libc, I'd rather consider to 
start with marking these testcases as "xfail". I, e.g., don't feel that 
anybody whithin the avr community will miss long double support for sprintf 
for quite a while :-) ...

Concerning the test suite it will require some tedious work to remove all the 
obsolete or irrelevant failures. I have started with doing this for 3.4.3, 
but I will have to redo most of the work for 4.0.0: the 4.0.0 testsuite 
harness is much improved and much better documented than it used to. 
I meanwhile already have opened a thread on gcc-patches and already got a 
response on the style the testsuite magic comments are expected to be 
generated in future. 

Most important work will be the definition of a new switch with the working 
title  "no_nested_function_support"  that makes it possible to remove all 
those related bugs. I thing that at least 300 of the remaining 491 failing 
test cases would be skipped this way.

Yours,

Björn