Subject: Re: USE_NEW_TOOLCHAIN toolchain build not parallel-safe
To: Havard Eidnes <email@example.com>
From: Todd Vierling <firstname.lastname@example.org>
Date: 10/01/2001 16:40:09
On Thu, 27 Sep 2001, Havard Eidnes wrote:
: Not only that -- the normal toolchain build is also busted in the same
: way (no big surprise there, I guess). I tried to do a "make -j2" of
: the tree recently -- my tree was updated Sep 23.
: The first time my build blew up was with
: --- dependall-gcc ---
: In file included from /usr/src/gnu/usr.bin/gcc/cc1/../../../dist/toolchain/gcc/c-iterate.c:34:
: /usr/src/gnu/usr.bin/gcc/cc1/../../../dist/toolchain/gcc/expr.h:353: insn-codes.h: No such file or directory
: /usr/tools/bin/mkdep: compile failed.
: *** Error code 1
: 1 error
: The reason appears to be that "backend" needs to be built before "cc1"
: can be descended into, i.e. a missing .WAIT between those two in
.WAIT is, IMNSHO, just plain silly. However, since this seems to be the
preferred method to do parallelism, I'll add the necessary goop to
<bsd.subdir.mk> to make this work Right[*]. I will also fix the other
reported problems with parallelism....
[*] Before the parallelism overhaul, subdir Makefiles would contain rules
like "all-foo: all-bar all-baz" to express the dependencies of programs,
so that a *single* program could be built from a clean tree by building
its dependencies first. What I didn't know, when creating the new trees
for the toolchain, was that this practice has become *completely broken*
thanks to the "dependall" target being used directly for recursion.
The Right way to fix this is to have dependency directives that will
work for both "all" and "dependall"; barring that, we could create
these dependencies on the fly, using the .WAIT directives as marks for
-- Todd Vierling <email@example.com> * Wasabi NetBSD: Run with it.
-- CDs, Integration, Embedding, Support -- http://www.wasabisystems.com/