[avr-gcc-list] ... Serious Bug in gcc head 4.1 and 4.0 .

[Haase Bjoern (PT-BEU/EMT) *]

Hi,

Today I have identified today a serious "wrong code" bug in the gcc 4.x.x 
family. The error I have seen shows up in the context of frame-pointer removal: 
In a situation where gcc would need two stack slots to place local variables, I 
have observed that both, a gcc 4.0 snapshot and present head cvs 4.1 ...

1.) Place one local automatic Pmode variable on the stack
2.) Overwrite the frame pointer in the body of the function
3.) Try to access the variable by using the overwritten frame pointer.

Hopefully I'll be able to fix the bug in the near future. I'd have posted the 
bug if it were not contained in an extremely confidential portion of my code. 
Hopefully, I'll succeed also in producing a non-confidential test case.

Meanwhile I'd like to recommend anybody NOT TO USE GCC 4.x.x FOR CRITICAL 
PROJECTS. It is, IMO, however very well worth experimenting with it in order to 
get a feeling of what one will be able to realize in the future: 4.x.x *will* 
be a substantial improvement in comparison to 3.4.4. .

Yours,

Björn