On Jan 5, 2010, at 8:07 PM, Jeff Johnson wrote:
>
> On Jan 5, 2010, at 7:15 PM, Todd Rinaldo wrote:
>
>> I'm building rpms from an srpm across multiple distributions. I ran into a problem on FC11 32bit which shows I don't fully understand how rpm filenames are generated. When the rpm file is generating, it's generating it as .i586.rpm but when I do rpm -E '%_arch' on that system, it comes back as i386.
>>
>
> Well there's actually several questions there ...
>
> 1) Q: How are *.rpm filenames generated?
> A: Output filenames are generated using this macro
> $ rpm -E '%{_rpmfilename}'
> which usually has some value (when as a macro) like
> %%{NAME}-%%{VERSION}-%%{RELEASE}.%%{ARCH}
> The doubled %% chars stop macro expansion, the remaining
> string with single % chars is filled in from header tags
> same as
> $ rpm -q --qf '%{NAME{-%{VERSION}-%{RELEASE}.%{ARCH}' bash
>
> 2) Q: How is RPMTAG_ARCH chosen for package headers?
> A: There's multiple ways. E.g.
> BuildArchitecture: noarch
> is what sets architecture for *.noarch.rpm.
> For your specific i386 != i586 architectire question, there's
> a "preferred" output of "i586" and there's a "canonical" architecture of "i386".
> The values are being set in build configuration which you can inspect with
> rpm --showrc
> There's also a means yo specify architecture using --target when building.
>
> 3) Q: What determines the platform architecture?
> A: Usually the CPU-VENDOR-OS-GNU string in /etc/rpm/platform, but (if missing)
> uname(2) values (with a minor translation like s/powerpc/ppc/) is used.
>
Jeff, thanks for your detailed responses
Q1: When running rpm --showrc |grep 86 on FC11/32, the only variable that seems to match the how my RPMs are being generated here is %_target_cpu. Would this be a better predictor of the arch of the RPM that was generated? At the moment, of the 50 Red Hat derivative systems I'm testing, only FC11/32 %_arch does not match the RPM file name %%ARCH.
Q2: Does the arch in the file name serve a purpose? if I rename a i586 RPM to i386, does the install of the RPM file behave differently? Most RPMs are compiled for a specific distribution and major version of that distribution. Was arch name something put into RPM file names long ago that in this day and age lacks a use other than knowing 32/64 bit-ness?
Q3: You mentioned:
A. /etc/rpm/platform (not there)
B. uname(2) -- perl -mPOSIX -e 'print join("\n", POSIX::uname()) . "\n"' implies i686, not i586
if A and B != i586, what is happening here?
I see from other responses that this was an intentional change by the fedora folks, but at the moment, I'm at a loss to figure out what was changed to make rpm file generation behave this way.
Thanks,
Todd
_______________________________________________
Rpm-list mailing list
Rpm-list@xxxxxxxxxxxxx
http://lists.rpm.org/mailman/listinfo/rpm-list
