| Summary: | libtasn1 shared library is miss on the target | ||
|---|---|---|---|
| Product: | [Build System, Metadata & Runtime] Meta-yocto | Reporter: | Lianhao Lu <lianhao.lu> |
| Component: | meta-yocto | Assignee: | Qing He <qing.he> |
| Status: | VERIFIED FIXED | QA Contact: | |
| Severity: | normal | ||
| Priority: | Medium | CC: | dongxiao.xu, jessica.zhang, kevin.tian, liping.ke, mark.hatle, poky.bs.watcher, poky.watcher, qing.he, richard.purdie, sgw, xianchao.zhang |
| Version: | 0.9 | ||
| Target Milestone: | 0.9 M4 | ||
| Hardware: | x86 | ||
| OS: | ppc | ||
| Whiteboard: | |||
| OS type for building Yocto: | --- | Type of Regression: | --- |
| Verified: | Program Management | Documentation change: | --- |
|
Description
Lianhao Lu
2010-10-12 20:22:43 UTC
Is it missing from all 4 arches? qemumips/qemuarm/qemux86 are verified to have that shared library installed in rootfs. atom-pc also missed that libruary. This is very likely due to the use of rpm_rootfs and debian renaming of shared
libraries.
According to debian renaming rule, the shared library package of libtasn1 (note
the ending digit) is renamed to libtasn1-3, thus the package file is now:
libtasn1-3-2.7-r0.core2.rpm
It is confirmed that libtasn1-3 is written as dependencies in other libraries,
but when resolve_package() in rootfs_rpm.bbclass runs, this package is not
inclluded.
ipk packaging doesn't suffer from this issue.
Two options are:
1. let rpm-native handles package names with `-', but is it even possible?
2. disable debian renaming for libtasn1, just like what libxml2 does:
python populate_packages_prepend() {
# autonamer would call this libxml2-2, but we don't want that
if bb.data.getVar('DEBIAN_NAMES', d, 1):
bb.data.setVar('PKG_libxml2', 'libxml2', d);
}
Cc Mark Hatle for question 1
RPM allows -'s in the names of packages, so that should not be an issue. Dependencies are really not handled by package name in RPM.. The name is only one facet of the dependency. Mostly they are handled by requires and provides on individual files. I think the key thing is what is the dependency on? What package is providing the dependency and to trace that. I suspect the dependency (based on the info below) is to "libtasn1.so" (a file name). The package provides that file, so the dependency is met. Instead the dependency should have been on libtasn1.so(VERSION). This is captured within the SONAMES of the files. If the libtasn1.so doesn't have an internal SONAME, then there is no way, other then purely via a filename to find the right file to include it. --- What is the command line to reproduce the failure condition? I can attempt to investigate this. In the mean time, you can follow what was installed on the target and trace that back to the RPM manifest list of what's to be installed. That stuff should all be logged as part of the build process. From there you can use rpm query commands to determine what dependencies were searched for and who the providers were. I was debugging it on a different setup with a slightly different phenomenon. It's sato image instead of sdk, where libtasn1-dev doesn't get installed, so instead all libtasn1.so* are missing in this case. I ran "rpm -qi -p libgnutls26... -R", and saw "libtasn1-3 >= 2.7" in the Requirement list, but not libtasn1.so.3, any hints? The issue appears to be this line in the recipe:
RREPLACES_${PN}-bin = "libtasn1 (<< 2.7)"
"<<" has no meaning in RPM. So the code generated becomes:
Obsoletes: libtasn1-3
So the bin package simply replaces the library package due to this flag.
Does << have meaning within Poky? If not, it's simply a typo and should be corrected. Otherwise its an un-implemented "feature". Once the behavior of "<<" is know, we can fix the RPM translation.
(In reply to comment #7) > The issue appears to be this line in the recipe: > RREPLACES_${PN}-bin = "libtasn1 (<< 2.7)" > "<<" has no meaning in RPM. So the code generated becomes: > Obsoletes: libtasn1-3 > So the bin package simply replaces the library package due to this flag. That makes sense. I did see libtasn1-bin around. The intention is to smooth the upgrading as the packaging has changed, the previous package libtasn1 is split to libtasn1 and libtasn1-bin. > Does << have meaning within Poky? If not, it's simply a typo and should be > corrected. Otherwise its an un-implemented "feature". Once the behavior of > "<<" is know, we can fix the RPM translation. In .deb, "<<" means strict precedence, while "<" has the same meaning of "<=", it looks misleading at the first glance, possibly due to some historic reason. Not really sure about opkg, but I assumed it inherits similar semantics. Actually, I don't think "package upgrading in target" works that well, so how about simply remove that line? btw, another question: if libtasn1-bin is to replace libtasn1-3 <= 2.6, while it dependes on libtasn1-3 >= 2.7, does that work with RPM semantics? >> Does << have meaning within Poky? If not, it's simply a typo and should be >> corrected. Otherwise its an un-implemented "feature". Once the behavior of >> "<<" is know, we can fix the RPM translation. > In .deb, "<<" means strict precedence, while "<" has the same meaning of "<=", > it looks misleading at the first glance, possibly due to some historic reason. > Not really sure about opkg, but I assumed it inherits similar semantics. In order to use these syntaxes within RPM, we need to find them and translate (or ignore) them in some way. The << should at a minimum be caught and filtered out of the system. > if libtasn1-bin is to replace libtasn1-3 <= 2.6, while it dependes on > libtasn1-3 >= 2.7, does that work with RPM semantics? Yes, those sematics will work within RPM. They seem odd to me that -bin would replace the library package of a previous version though. I vote for just removing this line iven where we're at in the release cycle. *** Bug 473 has been marked as a duplicate of this bug. *** fd59449f5b906bf5941fef0e940f5b651e4cf284 works around the issue. Before closing this bug, a bug for 1.0 to deal with the << operator should be filed with a note about this bug and the libtasn recipe included. In RC3 and master Verified in RC3 build. |