The bzip2 tool is not included in default installations.
On freshly installed systems, ./configure succeeds but 'make'
might fail later:
BUNZIP2 pc-bios/edk2-i386-secure-code.fd.bz2
/bin/sh: bzip2: command not found
make: *** [Makefile:305: pc-bios/edk2-i386-secure-code.fd] Error 127
make: *** Deleting file 'pc-bios/edk2-i386-secure-code.fd'
make: *** Waiting for unfinished jobs....
Add a check in ./configure to warn the user if bzip2 is missing.
Fixes: 536d2173b2b
Reported-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
---
configure | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/configure b/configure
index efe165edf9..9957e913e8 100755
--- a/configure
+++ b/configure
@@ -1851,6 +1851,13 @@ python_version=$($python -c 'import sys; print("%d.%d.%d" % (sys.version_info[0]
# Suppress writing compiled files
python="$python -B"
+# Some firmware binaries are compressed with bzip2
+if has bzip2; then
+ :
+else
+ error_exit "bzip2 program not found. Please install it"
+fi
+
# Check that the C compiler works. Doing this here before testing
# the host CPU ensures that we had a valid CC to autodetect the
# $cpu var (and we should bail right here if that's not the case).
--
2.21.0
On Fri, Nov 08, 2019 at 11:28:05AM +0100, Philippe Mathieu-Daudé wrote:
> The bzip2 tool is not included in default installations.
> On freshly installed systems, ./configure succeeds but 'make'
> might fail later:
>
> BUNZIP2 pc-bios/edk2-i386-secure-code.fd.bz2
> /bin/sh: bzip2: command not found
> make: *** [Makefile:305: pc-bios/edk2-i386-secure-code.fd] Error 127
> make: *** Deleting file 'pc-bios/edk2-i386-secure-code.fd'
> make: *** Waiting for unfinished jobs....
>
> Add a check in ./configure to warn the user if bzip2 is missing.
>
> Fixes: 536d2173b2b
> Reported-by: Thomas Huth <thuth@redhat.com>
> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
> ---
> configure | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/configure b/configure
> index efe165edf9..9957e913e8 100755
> --- a/configure
> +++ b/configure
> @@ -1851,6 +1851,13 @@ python_version=$($python -c 'import sys; print("%d.%d.%d" % (sys.version_info[0]
> # Suppress writing compiled files
> python="$python -B"
>
> +# Some firmware binaries are compressed with bzip2
> +if has bzip2; then
> + :
> +else
> + error_exit "bzip2 program not found. Please install it"
> +fi
Nitpick, I'd phrase this as
"The bzip2 program is required for building QEMU"
Either way though
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Regards,
Daniel
--
|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org -o- https://fstop138.berrange.com :|
|: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
On 08/11/2019 11.28, Philippe Mathieu-Daudé wrote:
> The bzip2 tool is not included in default installations.
> On freshly installed systems, ./configure succeeds but 'make'
> might fail later:
>
> BUNZIP2 pc-bios/edk2-i386-secure-code.fd.bz2
> /bin/sh: bzip2: command not found
> make: *** [Makefile:305: pc-bios/edk2-i386-secure-code.fd] Error 127
> make: *** Deleting file 'pc-bios/edk2-i386-secure-code.fd'
> make: *** Waiting for unfinished jobs....
>
> Add a check in ./configure to warn the user if bzip2 is missing.
>
> Fixes: 536d2173b2b
> Reported-by: Thomas Huth <thuth@redhat.com>
> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
> ---
> configure | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/configure b/configure
> index efe165edf9..9957e913e8 100755
> --- a/configure
> +++ b/configure
> @@ -1851,6 +1851,13 @@ python_version=$($python -c 'import sys; print("%d.%d.%d" % (sys.version_info[0]
> # Suppress writing compiled files
> python="$python -B"
>
> +# Some firmware binaries are compressed with bzip2
> +if has bzip2; then
> + :
> +else
> + error_exit "bzip2 program not found. Please install it"
> +fi
It's only required for the edk2 binaries, isn't it? So should this maybe
also check whether we build any of the i386-softmmu, x86_64-softmmu
arm-softmmu or aarch64-softmmu targets and pass otherwise?
Thomas
On 11/8/19 11:34 AM, Thomas Huth wrote:
> On 08/11/2019 11.28, Philippe Mathieu-Daudé wrote:
>> The bzip2 tool is not included in default installations.
>> On freshly installed systems, ./configure succeeds but 'make'
>> might fail later:
>>
>> BUNZIP2 pc-bios/edk2-i386-secure-code.fd.bz2
>> /bin/sh: bzip2: command not found
>> make: *** [Makefile:305: pc-bios/edk2-i386-secure-code.fd] Error 127
>> make: *** Deleting file 'pc-bios/edk2-i386-secure-code.fd'
>> make: *** Waiting for unfinished jobs....
>>
>> Add a check in ./configure to warn the user if bzip2 is missing.
>>
>> Fixes: 536d2173b2b
>> Reported-by: Thomas Huth <thuth@redhat.com>
>> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
>> ---
>> configure | 7 +++++++
>> 1 file changed, 7 insertions(+)
>>
>> diff --git a/configure b/configure
>> index efe165edf9..9957e913e8 100755
>> --- a/configure
>> +++ b/configure
>> @@ -1851,6 +1851,13 @@ python_version=$($python -c 'import sys;
>> print("%d.%d.%d" % (sys.version_info[0]
>> # Suppress writing compiled files
>> python="$python -B"
>> +# Some firmware binaries are compressed with bzip2
>> +if has bzip2; then
>> + :
>> +else
>> + error_exit "bzip2 program not found. Please install it"
>> +fi
>
> It's only required for the edk2 binaries, isn't it? So should this maybe
> also check whether we build any of the i386-softmmu, x86_64-softmmu
> arm-softmmu or aarch64-softmmu targets and pass otherwise?
I have this on my TODO somewhere, because we extract the edk2 firmwares
even if we only build MIPS/PowerPC.
On 11/08/19 11:42, Philippe Mathieu-Daudé wrote:
> On 11/8/19 11:34 AM, Thomas Huth wrote:
>> On 08/11/2019 11.28, Philippe Mathieu-Daudé wrote:
>>> The bzip2 tool is not included in default installations.
>>> On freshly installed systems, ./configure succeeds but 'make'
>>> might fail later:
>>>
>>> BUNZIP2 pc-bios/edk2-i386-secure-code.fd.bz2
>>> /bin/sh: bzip2: command not found
>>> make: *** [Makefile:305: pc-bios/edk2-i386-secure-code.fd] Error 127
>>> make: *** Deleting file 'pc-bios/edk2-i386-secure-code.fd'
>>> make: *** Waiting for unfinished jobs....
>>>
>>> Add a check in ./configure to warn the user if bzip2 is missing.
>>>
>>> Fixes: 536d2173b2b
>>> Reported-by: Thomas Huth <thuth@redhat.com>
>>> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
>>> ---
>>> configure | 7 +++++++
>>> 1 file changed, 7 insertions(+)
>>>
>>> diff --git a/configure b/configure
>>> index efe165edf9..9957e913e8 100755
>>> --- a/configure
>>> +++ b/configure
>>> @@ -1851,6 +1851,13 @@ python_version=$($python -c 'import sys;
>>> print("%d.%d.%d" % (sys.version_info[0]
>>> # Suppress writing compiled files
>>> python="$python -B"
>>> +# Some firmware binaries are compressed with bzip2
>>> +if has bzip2; then
>>> + :
>>> +else
>>> + error_exit "bzip2 program not found. Please install it"
>>> +fi
>>
>> It's only required for the edk2 binaries, isn't it? So should this
>> maybe also check whether we build any of the i386-softmmu,
>> x86_64-softmmu arm-softmmu or aarch64-softmmu targets and pass otherwise?
>
> I have this on my TODO somewhere, because we extract the edk2 firmwares
> even if we only build MIPS/PowerPC.
But that applies to all of "BLOBS" in the root-level Makefile, doesn't
it? We also install, say, "vgabios-qxl.bin", when only the MIPS/PowerPC
system emulators are built. The only aspect that's specific to edk2
binaries is that they are kept compressed until installation time, to
save space in the git repo and in the source tarball.
I'm reminded of the "external QEMU blob / firmware repo" idea.
Thanks
Laszlo
On 08/11/2019 12.39, Laszlo Ersek wrote:
> On 11/08/19 11:42, Philippe Mathieu-Daudé wrote:
>> On 11/8/19 11:34 AM, Thomas Huth wrote:
>>> On 08/11/2019 11.28, Philippe Mathieu-Daudé wrote:
>>>> The bzip2 tool is not included in default installations.
>>>> On freshly installed systems, ./configure succeeds but 'make'
>>>> might fail later:
>>>>
>>>> BUNZIP2 pc-bios/edk2-i386-secure-code.fd.bz2
>>>> /bin/sh: bzip2: command not found
>>>> make: *** [Makefile:305: pc-bios/edk2-i386-secure-code.fd] Error 127
>>>> make: *** Deleting file 'pc-bios/edk2-i386-secure-code.fd'
>>>> make: *** Waiting for unfinished jobs....
>>>>
>>>> Add a check in ./configure to warn the user if bzip2 is missing.
>>>>
>>>> Fixes: 536d2173b2b
>>>> Reported-by: Thomas Huth <thuth@redhat.com>
>>>> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
>>>> ---
>>>> configure | 7 +++++++
>>>> 1 file changed, 7 insertions(+)
>>>>
>>>> diff --git a/configure b/configure
>>>> index efe165edf9..9957e913e8 100755
>>>> --- a/configure
>>>> +++ b/configure
>>>> @@ -1851,6 +1851,13 @@ python_version=$($python -c 'import sys;
>>>> print("%d.%d.%d" % (sys.version_info[0]
>>>> # Suppress writing compiled files
>>>> python="$python -B"
>>>> +# Some firmware binaries are compressed with bzip2
>>>> +if has bzip2; then
>>>> + :
>>>> +else
>>>> + error_exit "bzip2 program not found. Please install it"
>>>> +fi
>>>
>>> It's only required for the edk2 binaries, isn't it? So should this
>>> maybe also check whether we build any of the i386-softmmu,
>>> x86_64-softmmu arm-softmmu or aarch64-softmmu targets and pass otherwise?
>>
>> I have this on my TODO somewhere, because we extract the edk2 firmwares
>> even if we only build MIPS/PowerPC.
>
> But that applies to all of "BLOBS" in the root-level Makefile, doesn't
> it? We also install, say, "vgabios-qxl.bin", when only the MIPS/PowerPC
> system emulators are built.
IIRC there was another odd dependency that the PCI devices need their
ROM also on non-x86 systems... but that's another story...
> The only aspect that's specific to edk2
> binaries is that they are kept compressed until installation time, to
> save space in the git repo and in the source tarball.
I noticed that there are also some iotests that use bzip2 ... they are
not used during the build process, but still, it might be better to
simply always require bzip2, also in case some other architectures want
to use it. So I'm also fine if we simply always require bzip2 for the build.
Thomas
PS: Anybody interested in writing a patch to compress the other big
binaries in the pc-bios folder, too? ... skiboot.lid ...
openbios-sparc64 ... ppc_rom.bin ...
On 11/8/19 12:39 PM, Laszlo Ersek wrote:
> On 11/08/19 11:42, Philippe Mathieu-Daudé wrote:
>> On 11/8/19 11:34 AM, Thomas Huth wrote:
>>> On 08/11/2019 11.28, Philippe Mathieu-Daudé wrote:
>>>> The bzip2 tool is not included in default installations.
>>>> On freshly installed systems, ./configure succeeds but 'make'
>>>> might fail later:
>>>>
>>>> BUNZIP2 pc-bios/edk2-i386-secure-code.fd.bz2
>>>> /bin/sh: bzip2: command not found
>>>> make: *** [Makefile:305: pc-bios/edk2-i386-secure-code.fd] Error 127
>>>> make: *** Deleting file 'pc-bios/edk2-i386-secure-code.fd'
>>>> make: *** Waiting for unfinished jobs....
>>>>
>>>> Add a check in ./configure to warn the user if bzip2 is missing.
>>>>
>>>> Fixes: 536d2173b2b
>>>> Reported-by: Thomas Huth <thuth@redhat.com>
>>>> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
>>>> ---
>>>> configure | 7 +++++++
>>>> 1 file changed, 7 insertions(+)
>>>>
>>>> diff --git a/configure b/configure
>>>> index efe165edf9..9957e913e8 100755
>>>> --- a/configure
>>>> +++ b/configure
>>>> @@ -1851,6 +1851,13 @@ python_version=$($python -c 'import sys;
>>>> print("%d.%d.%d" % (sys.version_info[0]
>>>> # Suppress writing compiled files
>>>> python="$python -B"
>>>> +# Some firmware binaries are compressed with bzip2
>>>> +if has bzip2; then
>>>> + :
>>>> +else
>>>> + error_exit "bzip2 program not found. Please install it"
>>>> +fi
>>>
>>> It's only required for the edk2 binaries, isn't it? So should this
>>> maybe also check whether we build any of the i386-softmmu,
>>> x86_64-softmmu arm-softmmu or aarch64-softmmu targets and pass otherwise?
>>
>> I have this on my TODO somewhere, because we extract the edk2 firmwares
>> even if we only build MIPS/PowerPC.
>
> But that applies to all of "BLOBS" in the root-level Makefile, doesn't
> it? We also install, say, "vgabios-qxl.bin", when only the MIPS/PowerPC
> system emulators are built. The only aspect that's specific to edk2
> binaries is that they are kept compressed until installation time, to
> save space in the git repo and in the source tarball.
You are right, other targets could improve this too.
Since this add quite complexity to the buildsys and nobody complained
about that previously, I suggest we clean that *after* we switch the
build machinery to Meson.
> I'm reminded of the "external QEMU blob / firmware repo" idea.
Do you mind starting a new thread asking about it? It would be nice we
clear this during the next dev cycle.
Regards,
Phil.
On 11/08/19 11:28, Philippe Mathieu-Daudé wrote:
> The bzip2 tool is not included in default installations.
> On freshly installed systems, ./configure succeeds but 'make'
> might fail later:
>
> BUNZIP2 pc-bios/edk2-i386-secure-code.fd.bz2
> /bin/sh: bzip2: command not found
> make: *** [Makefile:305: pc-bios/edk2-i386-secure-code.fd] Error 127
> make: *** Deleting file 'pc-bios/edk2-i386-secure-code.fd'
> make: *** Waiting for unfinished jobs....
>
> Add a check in ./configure to warn the user if bzip2 is missing.
We've come full circle. Let me explain:
>
> Fixes: 536d2173b2b
So this makes me kinda grumpy. If you look at the v3 posting of the patch that would later become commit 536d2173b2b:
http://mid.mail-archive.com/20190321113408.19929-8-lersek@redhat.com
you see the following note in the changelog:
- compress FD files with bzip2 rather than xz, so that decompression at
"make install" time succeed on older build OSes too [Peter]
So I couldn't use xz because that was "too new" for some build OSes, but now we also can't take bzip2 for granted because that's "too old" for some other build OSes? This is ridiculous.
To be clear, my disagreement is only with the "Fixes" tag. For me, "Fixes" stands for something that, in retrospect, can be proven to have been a bug at the time the code was *originally* committed. But, at the time, taking "bzip2" for granted was *not* a bug. The conditions / circumstances have changed more recently, and the assumption about bzip2 has been invalidated *after* adding a dependency on bzip2.
Nonetheless, thank you for adapting the code to the potential absence of bzip2. Can you perhaps go in some details in the commit message, near "not included in default installations" and "freshly installed systems"? If we can, we should identify the exact distro release where this problem has been encountered (and I wouldn't mind a link to the BZ or ticket under which people agreed to remove bzip2 from the default package set).
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Thanks
Laszlo
> Reported-by: Thomas Huth <thuth@redhat.com>
> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
> ---
> configure | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/configure b/configure
> index efe165edf9..9957e913e8 100755
> --- a/configure
> +++ b/configure
> @@ -1851,6 +1851,13 @@ python_version=$($python -c 'import sys; print("%d.%d.%d" % (sys.version_info[0]
> # Suppress writing compiled files
> python="$python -B"
>
> +# Some firmware binaries are compressed with bzip2
> +if has bzip2; then
> + :
> +else
> + error_exit "bzip2 program not found. Please install it"
> +fi
> +
> # Check that the C compiler works. Doing this here before testing
> # the host CPU ensures that we had a valid CC to autodetect the
> # $cpu var (and we should bail right here if that's not the case).
>
On 11/8/19 12:01 PM, Laszlo Ersek wrote:
> On 11/08/19 11:28, Philippe Mathieu-Daudé wrote:
>> The bzip2 tool is not included in default installations.
>> On freshly installed systems, ./configure succeeds but 'make'
>> might fail later:
>>
>> BUNZIP2 pc-bios/edk2-i386-secure-code.fd.bz2
>> /bin/sh: bzip2: command not found
>> make: *** [Makefile:305: pc-bios/edk2-i386-secure-code.fd] Error 127
>> make: *** Deleting file 'pc-bios/edk2-i386-secure-code.fd'
>> make: *** Waiting for unfinished jobs....
>>
>> Add a check in ./configure to warn the user if bzip2 is missing.
>
> We've come full circle. Let me explain:
>
>>
>> Fixes: 536d2173b2b
>
> So this makes me kinda grumpy. If you look at the v3 posting of the patch that would later become commit 536d2173b2b:
>
> http://mid.mail-archive.com/20190321113408.19929-8-lersek@redhat.com
>
> you see the following note in the changelog:
>
> - compress FD files with bzip2 rather than xz, so that decompression at
> "make install" time succeed on older build OSes too [Peter]
>
> So I couldn't use xz because that was "too new" for some build OSes, but now we also can't take bzip2 for granted because that's "too old" for some other build OSes? This is ridiculous.
>
> To be clear, my disagreement is only with the "Fixes" tag. For me, "Fixes" stands for something that, in retrospect, can be proven to have been a bug at the time the code was *originally* committed. But, at the time, taking "bzip2" for granted was *not* a bug. The conditions / circumstances have changed more recently, and the assumption about bzip2 has been invalidated *after* adding a dependency on bzip2.
>
> Nonetheless, thank you for adapting the code to the potential absence of bzip2. Can you perhaps go in some details in the commit message, near "not included in default installations" and "freshly installed systems"? If we can, we should identify the exact distro release where this problem has been encountered (and I wouldn't mind a link to the BZ or ticket under which people agreed to remove bzip2 from the default package set).
I am just reading this and already sent a v2.
I can amend these details. Thomas, what distro release were you using?
>
> Reviewed-by: Laszlo Ersek <lersek@redhat.com>
>
> Thanks
> Laszlo
>
>> Reported-by: Thomas Huth <thuth@redhat.com>
>> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
>> ---
>> configure | 7 +++++++
>> 1 file changed, 7 insertions(+)
>>
>> diff --git a/configure b/configure
>> index efe165edf9..9957e913e8 100755
>> --- a/configure
>> +++ b/configure
>> @@ -1851,6 +1851,13 @@ python_version=$($python -c 'import sys; print("%d.%d.%d" % (sys.version_info[0]
>> # Suppress writing compiled files
>> python="$python -B"
>>
>> +# Some firmware binaries are compressed with bzip2
>> +if has bzip2; then
>> + :
>> +else
>> + error_exit "bzip2 program not found. Please install it"
>> +fi
>> +
>> # Check that the C compiler works. Doing this here before testing
>> # the host CPU ensures that we had a valid CC to autodetect the
>> # $cpu var (and we should bail right here if that's not the case).
>>
>
On 08/11/2019 12.54, Philippe Mathieu-Daudé wrote: > On 11/8/19 12:01 PM, Laszlo Ersek wrote: >> On 11/08/19 11:28, Philippe Mathieu-Daudé wrote: >>> The bzip2 tool is not included in default installations. >>> On freshly installed systems, ./configure succeeds but 'make' >>> might fail later: >>> >>> BUNZIP2 pc-bios/edk2-i386-secure-code.fd.bz2 >>> /bin/sh: bzip2: command not found >>> make: *** [Makefile:305: pc-bios/edk2-i386-secure-code.fd] Error 127 >>> make: *** Deleting file 'pc-bios/edk2-i386-secure-code.fd' >>> make: *** Waiting for unfinished jobs.... >>> >>> Add a check in ./configure to warn the user if bzip2 is missing. >> >> We've come full circle. Let me explain: >> >>> >>> Fixes: 536d2173b2b >> >> So this makes me kinda grumpy. If you look at the v3 posting of the >> patch that would later become commit 536d2173b2b: >> >> http://mid.mail-archive.com/20190321113408.19929-8-lersek@redhat.com >> >> you see the following note in the changelog: >> >> - compress FD files with bzip2 rather than xz, so that >> decompression at >> "make install" time succeed on older build OSes too [Peter] >> >> So I couldn't use xz because that was "too new" for some build OSes, >> but now we also can't take bzip2 for granted because that's "too old" >> for some other build OSes? This is ridiculous. >> >> To be clear, my disagreement is only with the "Fixes" tag. For me, >> "Fixes" stands for something that, in retrospect, can be proven to >> have been a bug at the time the code was *originally* committed. But, >> at the time, taking "bzip2" for granted was *not* a bug. The >> conditions / circumstances have changed more recently, and the >> assumption about bzip2 has been invalidated *after* adding a >> dependency on bzip2. >> >> Nonetheless, thank you for adapting the code to the potential absence >> of bzip2. Can you perhaps go in some details in the commit message, >> near "not included in default installations" and "freshly installed >> systems"? If we can, we should identify the exact distro release where >> this problem has been encountered (and I wouldn't mind a link to the >> BZ or ticket under which people agreed to remove bzip2 from the >> default package set). > > I am just reading this and already sent a v2. > > I can amend these details. Thomas, what distro release were you using? I encountered this problem with a freshly installed Fedora 31. Thomas
On 11/8/19 5:01 AM, Laszlo Ersek wrote: >> >> Add a check in ./configure to warn the user if bzip2 is missing. > > We've come full circle. Let me explain: > >> >> Fixes: 536d2173b2b > > So this makes me kinda grumpy. If you look at the v3 posting of the patch that would later become commit 536d2173b2b: > > http://mid.mail-archive.com/20190321113408.19929-8-lersek@redhat.com > > you see the following note in the changelog: > > - compress FD files with bzip2 rather than xz, so that decompression at > "make install" time succeed on older build OSes too [Peter] > > So I couldn't use xz because that was "too new" for some build OSes, but now we also can't take bzip2 for granted because that's "too old" for some other build OSes? This is ridiculous. > > To be clear, my disagreement is only with the "Fixes" tag. For me, "Fixes" stands for something that, in retrospect, can be proven to have been a bug at the time the code was *originally* committed. But, at the time, taking "bzip2" for granted was *not* a bug. The conditions / circumstances have changed more recently, and the assumption about bzip2 has been invalidated *after* adding a dependency on bzip2. > > Nonetheless, thank you for adapting the code to the potential absence of bzip2. Can you perhaps go in some details in the commit message, near "not included in default installations" and "freshly installed systems"? If we can, we should identify the exact distro release where this problem has been encountered (and I wouldn't mind a link to the BZ or ticket under which people agreed to remove bzip2 from the default package set). bzip2 is no longer a favored compression. If we are trying to pick a compression that is most likely to be present on any system, go with gzip. If we are trying to pick a compression that packs tighter and uncompresses faster, pick xz or zstd. But bzip2 does neither: it packs slightly tighter than gzip but has slower performance in doing so, and thus is no longer used as a default compression. -- Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
On 11/8/19 4:43 PM, Eric Blake wrote: > On 11/8/19 5:01 AM, Laszlo Ersek wrote: > >>> >>> Add a check in ./configure to warn the user if bzip2 is missing. >> >> We've come full circle. Let me explain: >> >>> >>> Fixes: 536d2173b2b >> >> So this makes me kinda grumpy. If you look at the v3 posting of the >> patch that would later become commit 536d2173b2b: >> >> http://mid.mail-archive.com/20190321113408.19929-8-lersek@redhat.com >> >> you see the following note in the changelog: >> >> - compress FD files with bzip2 rather than xz, so that >> decompression at >> "make install" time succeed on older build OSes too [Peter] >> >> So I couldn't use xz because that was "too new" for some build OSes, >> but now we also can't take bzip2 for granted because that's "too old" >> for some other build OSes? This is ridiculous. >> >> To be clear, my disagreement is only with the "Fixes" tag. For me, >> "Fixes" stands for something that, in retrospect, can be proven to >> have been a bug at the time the code was *originally* committed. But, >> at the time, taking "bzip2" for granted was *not* a bug. The >> conditions / circumstances have changed more recently, and the >> assumption about bzip2 has been invalidated *after* adding a >> dependency on bzip2. >> >> Nonetheless, thank you for adapting the code to the potential absence >> of bzip2. Can you perhaps go in some details in the commit message, >> near "not included in default installations" and "freshly installed >> systems"? If we can, we should identify the exact distro release where >> this problem has been encountered (and I wouldn't mind a link to the >> BZ or ticket under which people agreed to remove bzip2 from the >> default package set). > > bzip2 is no longer a favored compression. If we are trying to pick a > compression that is most likely to be present on any system, go with > gzip. If we are trying to pick a compression that packs tighter and > uncompresses faster, pick xz or zstd. But bzip2 does neither: it packs > slightly tighter than gzip but has slower performance in doing so, and > thus is no longer used as a default compression. The problem was with OpenBSD 6.1 which hadn't xz available. In commit 12745eaa02 Gerd updated the VM to OpenBSD 6.5 and we now have access to xz. IIRC OSX supported versions also provide xz. If we want to revert Laszlo's patches and apply his first version (using xz), we should do that during 5.0 dev cycle, now it is too late. I'd prefer we simply fix bzip2 for the next release.
On Fri, 8 Nov 2019 at 17:07, Philippe Mathieu-Daudé <philmd@redhat.com> wrote: > On 11/8/19 4:43 PM, Eric Blake wrote: > > bzip2 is no longer a favored compression. If we are trying to pick a > > compression that is most likely to be present on any system, go with > > gzip. If we are trying to pick a compression that packs tighter and > > uncompresses faster, pick xz or zstd. But bzip2 does neither: it packs > > slightly tighter than gzip but has slower performance in doing so, and > > thus is no longer used as a default compression. > > The problem was with OpenBSD 6.1 which hadn't xz available. > > In commit 12745eaa02 Gerd updated the VM to OpenBSD 6.5 and we now have > access to xz. IIRC OSX supported versions also provide xz. > > If we want to revert Laszlo's patches and apply his first version (using > xz), we should do that during 5.0 dev cycle, now it is too late. > I'd prefer we simply fix bzip2 for the next release. I don't think we should try to use 'xz' because I don't see the point. We should use something that's generally available, whether that's bzip2 or gzip. Life's too short to deal with yet another file compression tool and format. thanks -- PMM
On 08/11/2019 18.10, Peter Maydell wrote: > On Fri, 8 Nov 2019 at 17:07, Philippe Mathieu-Daudé <philmd@redhat.com> wrote: >> On 11/8/19 4:43 PM, Eric Blake wrote: >>> bzip2 is no longer a favored compression. If we are trying to pick a >>> compression that is most likely to be present on any system, go with >>> gzip. If we are trying to pick a compression that packs tighter and >>> uncompresses faster, pick xz or zstd. But bzip2 does neither: it packs >>> slightly tighter than gzip but has slower performance in doing so, and >>> thus is no longer used as a default compression. >> >> The problem was with OpenBSD 6.1 which hadn't xz available. >> >> In commit 12745eaa02 Gerd updated the VM to OpenBSD 6.5 and we now have >> access to xz. IIRC OSX supported versions also provide xz. >> >> If we want to revert Laszlo's patches and apply his first version (using >> xz), we should do that during 5.0 dev cycle, now it is too late. >> I'd prefer we simply fix bzip2 for the next release. > > I don't think we should try to use 'xz' because I don't see > the point. We should use something that's generally available, > whether that's bzip2 or gzip. Life's too short to deal with > yet another file compression tool and format. FWIW, on the weekend, I accidentially came accross this page: https://www.nongnu.org/lzip/xz_inadequate.html After reading that, I also don't think anymore that we should switch to 'xz'. Thomas
On Monday, November 11, 2019, Thomas Huth <thuth@redhat.com> wrote: > On 08/11/2019 18.10, Peter Maydell wrote: > > On Fri, 8 Nov 2019 at 17:07, Philippe Mathieu-Daudé <philmd@redhat.com> > wrote: > >> On 11/8/19 4:43 PM, Eric Blake wrote: > >>> bzip2 is no longer a favored compression. If we are trying to pick a > >>> compression that is most likely to be present on any system, go with > >>> gzip. If we are trying to pick a compression that packs tighter and > >>> uncompresses faster, pick xz or zstd. But bzip2 does neither: it packs > >>> slightly tighter than gzip but has slower performance in doing so, and > >>> thus is no longer used as a default compression. > >> > >> The problem was with OpenBSD 6.1 which hadn't xz available. > >> > >> In commit 12745eaa02 Gerd updated the VM to OpenBSD 6.5 and we now have > >> access to xz. IIRC OSX supported versions also provide xz. > >> > >> If we want to revert Laszlo's patches and apply his first version (using > >> xz), we should do that during 5.0 dev cycle, now it is too late. > >> I'd prefer we simply fix bzip2 for the next release. > > > > I don't think we should try to use 'xz' because I don't see > > the point. We should use something that's generally available, > > whether that's bzip2 or gzip. Life's too short to deal with > > yet another file compression tool and format. > > FWIW, on the weekend, I accidentially came accross this page: > > https://www.nongnu.org/lzip/xz_inadequate.html > > After reading that, I also don't think anymore that we should switch to > 'xz'. Off topic, but, from the page you linked to: There are two ways of constructing a software design: One way is to make it so simple that there are obviously no deficiencies and the other way is to make it so complicated that there are no obvious deficiencies. The first method is far more difficult. -- C.A.R. Hoare Perfection is reached, not when there is no longer anything to add, but when there is no longer anything to take away. -- Antoine de Saint-Exupery Perhaps we should display these quotes at some highly visible place on our dev QEMU website. ;) Aleksandar > > Thomas > > >
On Fri, Nov 08, 2019 at 12:01:16PM +0100, Laszlo Ersek wrote: > On 11/08/19 11:28, Philippe Mathieu-Daudé wrote: > > The bzip2 tool is not included in default installations. > > On freshly installed systems, ./configure succeeds but 'make' > > might fail later: > > > > BUNZIP2 pc-bios/edk2-i386-secure-code.fd.bz2 > > /bin/sh: bzip2: command not found > > make: *** [Makefile:305: pc-bios/edk2-i386-secure-code.fd] Error 127 > > make: *** Deleting file 'pc-bios/edk2-i386-secure-code.fd' > > make: *** Waiting for unfinished jobs.... > > > > Add a check in ./configure to warn the user if bzip2 is missing. > > We've come full circle. Let me explain: > > > > > Fixes: 536d2173b2b > > So this makes me kinda grumpy. If you look at the v3 posting of the patch that would later become commit 536d2173b2b: > > http://mid.mail-archive.com/20190321113408.19929-8-lersek@redhat.com > > you see the following note in the changelog: > > - compress FD files with bzip2 rather than xz, so that decompression at > "make install" time succeed on older build OSes too [Peter] > > So I couldn't use xz because that was "too new" for some build OSes, but now we also can't take bzip2 for granted because that's "too old" for some other build OSes? This is ridiculous. We're not saying bzip2 is too old / missing from the OS. Every OS we care about has bzip2. The problem is that a minimal installation migt not have installed it. This kind of problem is increasingly common with use of minimal container images for example. So we're just ensuring we validate that what we want is actuall present. Regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
© 2016 - 2024 Red Hat, Inc.