qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [PATCH v2] Drop --whole-archive and static libraries


From: Andreas Färber
Subject: Re: [Qemu-devel] [PATCH v2] Drop --whole-archive and static libraries
Date: Wed, 6 Jan 2010 19:51:05 +0100


Am 04.01.2010 um 21:47 schrieb Blue Swirl:

On Sun, Jan 3, 2010 at 12:31 PM, Blue Swirl <address@hidden> wrote:
2009/12/31 Andreas Färber <address@hidden>:
From: Andreas Färber <address@hidden>

Juan has contributed a cool Makefile infrastructure that enables us to drop
static libraries completely:

Move shared obj-y definitions to Makefile.objs, prefixed {common-,hw-,user-},
and link those object files directly into the executables.

Replace HWLIB by HWDIR, specifying only the directory.

Drop --whole-archive and ARLIBS in Makefiles and configure.

Drop GENERATED_HEADERS dependency in rules.mak, since this rebuilds all common objects after generating a target-specific header; add dependency
rules to Makefile and Makefile.target instead.

v2:
- Don't try to include /config.mak for user emulators
- Changes to user object paths ("Quickfix for libuser.a drop") were obsoleted
 by "user_only: compile everything with -fpie" (Kirill A. Shutemov)

Breaks build:
 CC    i386-softmmu/i386-dis.o
make[1]: *** No rule to make target `/loader.o', needed by `qemu'. Stop.

I found out that completely removing the object directory and
recreating it from scratch helps. This sucks bisectability-wise, is
there any way to avoid it?

Sounds as if configure was not re-run and thus $(HWDIR) not set in config-target.mak. No idea why, more likely a general issue. Do you have a way to reproduce? What about `make clean` or `make distclean`?

There is still one problem, tools need config-host.h but the
dependencies are not correct:
 CC    qemu-nbd.o
In file included from /src/qemu/qemu-nbd.c:19:
/src/qemu/qemu-common.h:32:25: error: config-host.h: No such file or directory

Fix coming up. Basically, $(GENERATED_HEADERS) were previously used on _all_ object files via rules.mak. That does not work when referencing objects from another Makefile, so the modelling needs to be done explicitly.

Andreas



reply via email to

[Prev in Thread] Current Thread [Next in Thread]