[net.lang.c] C cross-compilers for micros

jeff (12/14/82)

I have an ever-increasing need for C cross-compilers for various uP's.
I'm writing this note to solicit any advice or suggestions you may have to
offer.

Here's the picture:  My company runs 4.1BSD on several machines (and plans to
acquire some SUNS in short order).  We use UNIX as a development environment to
produce embedded software for a variety of 8 and 16-bitters.  We're wanting to
write this uP software in C because writing assembly code is boring.

So far we've gotten by with C compliers purchased from Whitesmiths'.  We've
hacked these up to emit code for the TMS9900 and the Z80, but we're not keen
on spending six months grinding out another code generator every time a new chip
pops out of Silicon Valley.

We're even more concerned about the licensing morass when we start bringing in
numerous SUN workstations.  Multiply dozens of host machines by the multiplicity
of compilers and our licensing costs grow unpleasantly large.  Now add another
degree of freedom by considering multiple vendors (each with a half-dozen
licensing options) and we see ourselves spending a goodly fraction of a man-
year just trying to keep all the lawyers placated.

So, dear readers, here's the Question of the Day:  Anybody else out there in
the same boat?  Do you have a plan for orderly growth and acquisition of new
cross-compilers?  How do you deal with the combinatorial explosion of
    HOSTS x VENDORS x VERSIONS x LICENSING OPTIONS x TARGET MACHINES?

How do you acquire cross-compilers for new $1.98 target machines?  Buy 'em?
>From whom?  Roll your own?  Starting from what base - PCC or something else?
Any advice from One Who Knows?

Drop me a line - I'd love to talk with you.  If nothing else, we can commis-
erate.
	-- Jeff Stearns,  John Fluke Mfg. Co, Inc.  (206) 356-5064
	....decvax!microsoft!fluke!jeff

tihor (12/23/82)

#R:fluke:-61900:cmcl2:15400001:000:860
cmcl2!tihor    Dec 22 00:41:00 1982

Ideally you would like a setup such as the one Whitesmiths designed with 
one multipass C compiler and several asm phases one per machine.  At this
point you have only the cost of recoding some small set of high performance
or kernal flavored library functions, a standard assembler or assembler
front end, and, if desired, object level tools like asmopt and adb.

There is none the less that cost in adding another C system to your list,
although if the numbers seem to grow fast enough then you ought to be
able to afford source licenses for all the necessary pieces from whomever.
By using a common middle ground (the common intermediary code) you
reduce a N * M problem to a N + M problem.  If there are enough machines
and chips involved then it certainly makes more sense to hire a compiler/
assembler writer rather than a lawyer to take care of it all.