[oe] Why not use config.guess in conf/bitbake.conf ?

[Esben Haabendal]

What is the reason that conf/bitbake.conf is relying directly on the 'uname' output for generating the BUILD_SYS string?

I am trying to get OE building on Cygwin, and on my machine bitbake.conf generates a BUILD_SYS that is not supported by config.sub, which breaks a autotools packages.

$ uname -a
CYGWIN_NT-5.1 PC_K_EHA 1.5.24(0.156/4/2) 2007-01-31 10:57 i686 Cygwin

As bitbake.conf takes the first field as BUILD_OS, and the 5th field as BUILD_ARCH and then set BUILD_SYS to ${BUILD_ARCH}-${BUILD_OS} I end up with
i686-cygwin_nt-5.1
which breaks config.sub due to the '-' in the os string.

Using a current config.guess on this system, I get
i686-pc-cygwin
which is fine.
On my laptop (running gutsy) it would give me a BUILD_SYS
i686-pc-linux-gnu

Another workaround is to use the last field (also output from 'uname -o') as BUILD_OS which give
i686-cygwin
On my laptop (still running gutsy) it would also give me a BUILD_SYS
i686-pc-linux-gnu

To keep paths short and nice, I can see it would be prefereable to end up with i686-cygwin and i686-linux.
On the other hand, using the output from config.guess migt warrant some more consistency.

Anyway, the current solution just don't work for cygwin, so I would like to hear what kind of solution would fit OE the best.

P.S. No, I cannot just use coLinux.

/Esben
-- 
Esben Haabendal
Embedded Software Consultant
Doré Development ApS