Re: [MiNT] GCC testsuite

[Keith Scroggins <kws@radix.net>]

Hello Miro,

> So I installed expect + tcl RPMs + compiled dejagnu (no problems at 
> all), went into $top_src/gcc (here is stage1-xgcc), ran make check and a 
> looooot and lot of FAILs! I played also with RUNTESTFLAGS (to get 
> -m68060 code tests) and the same result. How can our gcc produce 815 
> expected PASSes and 30 000 unexpected FAILs ?

If memory serves me right (and we might need to buzz Mark Duckworth for 
this) the problem with Dejagnu is that our version of expect was too old 
for the tests, hence why all of the failures.  And, building the newer 
one, some problems were encountered with reading from the terminal 
correctly.  It has been some time though since I think we looked at this 
(3 years maybe?).

I suppose I can dig into it.  I was going to try building a 68060 version 
of GCC today (along with a binutils rpm and a Cross compiler with the 
68060 target).  I'll have to see how my time goes as far as starting to 
build things at different locations.

I did install your archive on my Falcon, thanks.  Only suggestion I can 
make is that the binaries should be stripped, would greatly reduce the 
size of the tbz2.  Have not used it much yet (I did try ScummVM with a 
-m68060 and the resulting binary spits out Privilege Violation, but I need 
to build a debug version to examine closer in gdb).  If I play with 
expect, I'll be using it.

I did build the upcoming ScummVM release with the latest Cross Compiler 
tools.  -O2 produced a 6.5 Meg binary, -Os produced a 5.5 Meg binary.  The 
-Os binary ran roughly 10 to 15% slower than the -O2 produced binary.  Not 
alot of experimentation, but since I had the ability, I figured I would 
test it. All libraries that went into the binary, with the exception of 
libc and the gcc libs, were built with the same optimizations as well when 
I did this.  -O2 was the submitted release version (due out 2/29).

Sorry about the semi off topic babbling....  :)

Keith