Drop roms from standard build
PATCH 1/1 Drop roms from standard build
There is no point in generating and copying the optionsrom each time qemu is build. It's a maintenance task given that we carry them as binaries already. And that task could also be triggered explicitly via 'make roms'. Solves the mingw32 build breakage, too.
---
Makefile | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
Makefile +1 -1
@@ -44,7 +44,7 @@ ifdef CONFIG_WIN32 LIBS+=-lwinmm -lws2_32 -liphlpapi endif -build-all: $(TOOLS) $(DOCS) roms recurse-all +build-all: $(TOOLS) $(DOCS) recurse-all config-host.mak: configure ifneq ($(wildcard config-host.mak),)
Quoted 4 lines · Jan Kiszka
There is no point in generating and copying the optionsrom each time qemu is build. It's a maintenance task given that we carry them as binaries already. And that task could also be triggered explicitly via 'make roms'. Solves the mingw32 build breakage, too.
Well, Anthony specifically asked me to make the roms get built automatically whenever possible.
So that behavior is actually on purpose :-)
Alex
Quoted 21 lines
Signed-off-by Jan Kiszka <[email protected]>---
Makefile| 2 +-1 files changed, 1 insertions(+), 1 deletions(-)Makefile +0
-0@@ -44,7 +44,7 @@ ifdef CONFIG_WIN32LIBS+=-lwinmm -lws2_32 -liphlpapi
endifMakefile +1
-1-build-all: $(TOOLS) $(DOCS) roms recurse-all +build-all: $(TOOLS) $(DOCS) recurse-allconfig-host.mak: configure
ifneq ($(wildcard config-host.mak),)
Quoted 11 lines · Alexander Graf
Quoted 4 lines · Jan Kiszka
There is no point in generating and copying the optionsrom each time qemu is build. It's a maintenance task given that we carry them as binaries already. And that task could also be triggered explicitly via 'make roms'. Solves the mingw32 build breakage, too.
Quoted 4 lines
Well, Anthony specifically asked me to make the roms get built automatically whenever possible.
So that behavior is actually on purpose :-)
Then quite a few things need to be fixed:
- host tool generation (signrom, I think there were several approached under discussion - but what is preferred now?)
- disabling the build for incompatible tool chains (mingw32?)
- do not automatically copy stuff into the source dir (absolute no-go if you want to keep it clean from build outputs)
So for now I would simply disable the build - unless I missed that there are already patches pending to fix all this.
Jan
Quoted 5 lines · Jan Kiszka
There is no point in generating and copying the optionsrom each time qemu is build. It's a maintenance task given that we carry them as binaries already. And that task could also be triggered explicitly via 'make roms'. Solves the mingw32 build breakage, too.
Yeah, it needs fixing but we absolutely want to build the roms as part of the normal build process.
When a user types make install from a source tarball, if they have the proper tool chain, we should be building everything from scratch and installing the versions that we built.
Signature · 3 lines
Regards, Anthony Liguori
Quoted 8 lines · Anthony Liguori
Quoted 4 lines · Jan Kiszka
There is no point in generating and copying the optionsrom each time qemu is build. It's a maintenance task given that we carry them as binaries already. And that task could also be triggered explicitly via 'make roms'. Solves the mingw32 build breakage, too.
Yeah, it needs fixing but we absolutely want to build the roms as part of the normal build process.
I disagree. For the most part I'd expect roms to be better maintained as a side project.
Quoted 3 lines
When a user types make install from a source tarball, if they have the proper tool chain, we should be building everything from scratch and installing the versions that we built.
I think the chances of the user having appropriate toolchains readily available (without significant additional per-rom configury) are extremely slim.
Paul
Quoted 9 lines · Paul Brook
Quoted 7 lines · Anthony Liguori
Quoted 5 lines · Jan Kiszka
There is no point in generating and copying the optionsrom each time qemu is build. It's a maintenance task given that we carry them as binaries already. And that task could also be triggered explicitly via 'make roms'. Solves the mingw32 build breakage, too.
Quoted 6 lines
Quoted 3 lines
Yeah, it needs fixing but we absolutely want to build the roms as part of the normal build process.
I disagree. For the most part I'd expect roms to be better maintained as a side project.
For things like gpxe where there's a well defined boundary between QEMU and the rom, yes, that makes sense.
For something like the PC BIOS, it's proven to be very difficult to maintain it out of tree. It's difficult for people to develop patches for it, difficult for people to test those patches, and then difficult for the maintainers to pull in those patches.
Making it part of the tree and a standard part of the build system will tremendously simplify that.
Quoted 2 lines
I think the chances of the user having appropriate toolchains readily available (without significant additional per-rom configury) are extremely slim.
Actually, for x86, all of our ROMs will just use GCC so when building x86 on x86, it's very likely the right tool chain is available.
Regards,
Anthony Liguori
Quoted 6 lines
Quoted 3 lines
I think the chances of the user having appropriate toolchains readily available (without significant additional per-rom configury) are extremely slim.
Actually, for x86, all of our ROMs will just use GCC so when building x86 on x86, it's very likely the right tool chain is available.
Except when it isn't, for example when building with a windows binary. Also, x86 is only one of many targets qemu supports. If you're serious about rebuilding roms as standard, then at bare minimum you need to handle sparc and ppc as well as x86.
Paul
Quoted 10 lines · Paul Brook
Quoted 7 lines
Quoted 4 lines
I think the chances of the user having appropriate toolchains readily available (without significant additional per-rom configury) are extremely slim.
Actually, for x86, all of our ROMs will just use GCC so when building x86 on x86, it's very likely the right tool chain is available.
Except when it isn't, for example when building with a windows binary.
That's a simple problem that has to do with the way ld is being used. A patch was posted that actually works with the mingw ld.
Quoted 2 lines
Also, x86 is only one of many targets qemu supports. If you're serious about rebuilding roms as standard, then at bare minimum you need to handle sparc and ppc as well as x86.
Yup. That's why we need to probe for toolchains in order to determine whether roms can be built. If you're a developer making changes to OpenBIOS, clearly you are going to have the appropriate tool chains installed.
Quoted 2 lines
Paul
Signature · 3 lines
Regards, Anthony Liguori
Quoted 8 lines
Quoted 3 lines
Also, x86 is only one of many targets qemu supports. If you're serious about rebuilding roms as standard, then at bare minimum you need to handle sparc and ppc as well as x86.
Yup. That's why we need to probe for toolchains in order to determine whether roms can be built. If you're a developer making changes to OpenBIOS, clearly you are going to have the appropriate tool chains installed.
IMO a developer changes to a rom is a special case.
If you're arguing that roms should be built by default, for everyone, then I disagree and don't think this is feasible. Most users simply don't have the prerequisites, and don't want to have to cofigure everything even when they do. If you care about this (for reproducibility or licencing reasons) then it should be all or nothing. I don't buy "It works for x86-on-x86 most of the time" as a valid argument for anything. We either build things by default, or we include binaries, not both.
If you're talking about developers modifying the roms, then most of the arguments above still apply. Chances are you're only actually interested in one rom anyway, so building it separately really isn't significant extra hassle.
Having everything in one source tree maybe makes sense. Except that most roms are already used by several projects. What you really want is to coordinate releases with the upstream maintainers, having a qemu branck/fork if it's really that hard to get changes integrated back. In the case of the pc-bios it probably wants to be split into a sparate project, rather than being part of bochs.
Paul
Quoted 7 lines · Paul Brook
Quoted 6 lines
Quoted 3 lines
I think the chances of the user having appropriate toolchains readily available (without significant additional per-rom configury) are extremely slim.
Actually, for x86, all of our ROMs will just use GCC so when building x86 on x86, it's very likely the right tool chain is available.
Quoted 6 lines
Except when it isn't, for example when building with a windows binary. Also, x86 is only one of many targets qemu supports. If you're serious about rebuilding roms as standard, then at bare minimum you need to handle sparc and ppc as well as x86.
Paul
Actually windows binaries of MinGW can built this just fine, you just have to link it to the native platform format (PE-COFF) and then objcopy it to binary. The rest should work just like on Linux. I'm building the BIOS from time to time on MinGW and while it is annoying to modify the build files to work for MinGW I'd be happy to maintain it for QEMU.
Best regards,
Filip Navara