Re: How is RPMTAG_FILELINKTOS filled?
On Tue, Oct 28, 2008 at 5:54 PM, Jeff Johnson [EMAIL PROTECTED] wrote: On Oct 28, 2008, at 12:36 PM, Eric MSP Veith wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Hello, while creating a package, I renamed in the %install section a bunch of manpages by changing their extension from e. g. 1 to 1ssl. RPM-5.1.6 creates an implicit dependency for each renamed manpage. I got curious on what (1) causes the dependency generation (it is not a (symbolic|hard) link, just a regular file), what (2) actually triggers the dependency generation and how (3) to avoid it. I tried digging through the source code, but I got completely lost. So I wrote this message to the ML. Thanks in advance! What is conceptiually confusing aboy filelinktos (and parentdir) dependencies is that there is no change to package metadata content whatsoever. What is being done internally to rpm during install is to take RPMTAG_FILELINKTOS (and RPMTAG_DIRNAMES) string arrays and wrap the contents as if Requires: had been added to the spec file originally. No matter how hard you look, you will not find any additional metadata, its in the implementation. Add --noparentdirs and --nolinktos to disable the dependencies. The only remaining issue is what the default in rpm should be. Just an opinion on the file linkto and parentdirs deps, already posed in other forms some time ago (http://rpm5.org/community/rpm-devel/2374.html) For me it is a feature of rpm5 and not a incompatibility: for example other, that don't have this feature only define the guideline but don't impose then ( https://fedoraproject.org/wiki/Packaging/UnownedDirectories). Being a feature could be the case for rpm5, to fullfill automatically but giving warning at installation time , of /etc/rpm/sysinfo/Dirnames and /etc/rpm/sysinfo/Filelinktos. Perhaps a rpm macros could be necessary for enable/disable this behaviour. Look like a RFE or a plan ? Opinions ? Since I might as well not have done the implementation (as with --nosignatures and --nodigests disablers) if the default setting is always and routinely turned off, I choose to distribute rpm with the functionality enabled. Exercise left to lusers to find how to disable the functionality. Hint: its very not hard. 73 de Jeff __ RPM Package Managerhttp://rpm5.org User Communication List [EMAIL PROTECTED]
Re: [CVS] RPM: rpm/ CHANGES VENDOR rpm/lib/ depends.c
Could be better also to fullfill automatically /etc/rpm/sysinfo/Dirname e
Filelinkto, giving an hint to the user ? In this case there is the
possibilty to catch this packaging error and to improve the quality of the
system -- and the distro.
Regards
On Tue, Oct 28, 2008 at 4:25 PM, Per Øyvind Karlsen [EMAIL PROTECTED]wrote:
RPM Package Manager, CVS Repository
http://rpm5.org/cvs/
Server: rpm5.org Name: Per Øyvind Karlsen
Root: /v/rpm/cvs Email: [EMAIL PROTECTED]
Module: rpm Date: 28-Oct-2008 16:25:20
Branch: HEAD Handle: 2008102815251901
Modified files:
rpm CHANGES VENDOR
rpm/lib depends.c
Log:
Mandriva: make directory symlink dependencies optional with
%{_check_symlink_deps}
%{_check_dirname_deps}. (optional-dirname-and-symlink-deps)
Summary:
RevisionChanges Path
1.2636 +2 -0 rpm/CHANGES
2.61+14 -1 rpm/VENDOR
1.424 +16 -0 rpm/lib/depends.c
patch -p0 '@@ .'
Index: rpm/CHANGES
$ cvs diff -u -r1.2635 -r1.2636 CHANGES
--- rpm/CHANGES 28 Oct 2008 14:19:44 - 1.2635
+++ rpm/CHANGES 28 Oct 2008 15:25:19 - 1.2636
@@ -1,5 +1,7 @@
5.2a2 - 5.2a3:
+- proyvind: Mandriva: make directory symlink dependencies optional
with
+%{_check_symlink_deps} %{_check_dirname_deps}.
(optional-dirname-and-symlink-deps)
- proyvind: Mandriva: prefer file extension for detecting file
compression.
(extension-based-compression-detection)
- proyvind: Mandriva: disable buildrequires checking for building
src.rpms.
@@ .
patch -p0 '@@ .'
Index: rpm/VENDOR
$ cvs diff -u -r2.60 -r2.61 VENDOR
--- rpm/VENDOR28 Oct 2008 14:19:44 - 2.60
+++ rpm/VENDOR28 Oct 2008 15:25:19 - 2.61
@@ -514,7 +514,7 @@
Change: no-deps-on-building-srpms
Purpose:Do not perform any dependency checking on building
SRPMs.
Reason: Don't really need this as 'digest(...)' and friends
hasn't
- been put to use (yet), making this just annoying.
+ been put to use (so far), making this just annoying.
Change: extension-based-compression-detection
@@ -531,6 +531,19 @@
detecting LZMA we want to do via file extension
only.
+ Change: optional-dirname-and-symlink-deps
+ Purpose:Optional dependencies on symlinks and parent
directories.
+ Reason: Mandriva doesn't store file lists in it's standard
+ synthesis metadata, resulting in any dependencies on
these
+ not met during installation fail. Also always
enforcing
+ these dependencies on third party packages which
rarely
+ complies with distro policies, is of varying quality
wrt.
+ quality and what not will also often lead to these
having
+ files in directories not owned etc. Also in the case
of
+ usage on systems not primarily using rpm, this will
+ ~always be an issue.
+
+
o Name: WINDRIVER
Vendor: Wind River http://windriver.com/
Representative: Jeff Johnson [EMAIL PROTECTED]
@@ .
patch -p0 '@@ .'
Index: rpm/lib/depends.c
$ cvs diff -u -r1.423 -r1.424 depends.c
--- rpm/lib/depends.c 27 Oct 2008 13:18:15 - 1.423
+++ rpm/lib/depends.c 28 Oct 2008 15:25:20 - 1.424
@@ -1551,6 +1551,10 @@
int terminate = 2; /* XXX terminate if rc = terminate
*/
int rc;
int ourrc = 0;
+#if defined(RPM_VENDOR_MANDRIVA) /* optional-dirname-and-symlink-deps */
+int dirname_deps;
+int symlink_deps;
+#endif
requires = rpmdsInit(requires);
if (requires != NULL)
@@ -1626,6 +1630,10 @@
}
}
+#if defined(RPM_VENDOR_MANDRIVA) /* optional-dirname-and-symlink-deps */
+dirname_deps =
rpmExpandNumeric(%{?_check_dirname_deps}%{?!_check_dirname_deps:1});
+if (dirname_deps) {
+#endif
dirnames = rpmdsInit(dirnames);
if (dirnames !=
Re: [CVS] RPM: rpm/ CHANGES VENDOR rpm/lib/ depends.c
2008/10/28 devzero2000 [EMAIL PROTECTED] Could be better also to fullfill automatically /etc/rpm/sysinfo/Dirname e Filelinkto, giving an hint to the user ? In this case there is the possibilty to catch this packaging error and to improve the quality of the system -- and the distro. Enabling it would sure help to spot these and improve package quality in the distro, but as the distro isn't really using this and not all that much attention are paid to these things, fixing every single issue I come across would lead to much work for me doing janitor work.. So that's why I leave it optional in stead of disabling it completely, I intend to deal with this in some way eventually/occationaly. :) -- Regards, Per Øyvind
Re: [CVS] RPM: rpm/ CHANGES VENDOR rpm/lib/ depends.c
On Oct 28, 2008, at 3:46 PM, Per Øyvind Karlsen wrote: 2008/10/28 devzero2000 [EMAIL PROTECTED] Could be better also to fullfill automatically /etc/rpm/sysinfo/ Dirname e Filelinkto, giving an hint to the user ? In this case there is the possibilty to catch this packaging error and to improve the quality of the system -- and the distro. Enabling it would sure help to spot these and improve package quality in the distro, but as the distro isn't really using this and not all that much attention are paid to these things, fixing every single issue I come across would lead to much work for me doing janitor work.. So that's why I leave it optional in stead of disabling it completely, I intend to deal with this in some way eventually/ occationaly. :) FWIW: There is a --orphandirs option to assist with initially populating the Dirnames masking (until vendors get around to fixing their packages). All that's needed to mask the pain of parentdir dependencies is to do mkdir -p /etc/rpm/sysinfo rpm -V --orphandors /etc/rpm/sysinfo/Dirnames immediately after installing rpm-4.4.7 (or later, including rpm-5.x) Pain with filelinktos deps is several orders of magnitude smaller than with parentdir dependencies, so I did not bother adding a popt alias like --orpandirs to jumpstart supplying symlink end-point dependencies. After populating /etc/rpm/sysinfoDirnames, I find (but YMMV) that the manual effort involved with editing in additional orphan directory paths into /etc/rpm/sysinfo/Dirnames when, say, an upgrade to a package adds new orphan directories to be quite modest and acceptable. Sure annoying, but using rpm is always annoying. ;-) Maintaining the list of orphandirs automagically is certainly doable. OTOH, the packages, not rpm, are what really need fixing. Automagically maintaining a list of orphan directory paths automagically is unlikely to fix *.rpm packages this decade imho. As much as I hate nag ware, any utility gain by automagically maintaining a list of orphan directories by rpm hardly seems worth the implementation effort. Improving the quality of *.rpm packages is the goal, consistent policies for how to package directories correctly have never been completely specified afaik. Note that the issue of orphan directories has a huge and complicated history. The problem was (quite correctly imho) diagnosed by Enrico Scholtz like 5+ years ago. Wiring a dependency on parent directories directly is the only implementation (that I know of) that has a prayer of accomplishing the necessary (and yes, annoying) goal of improving the quality of *.rpm packages. 73 de Jeff
Re: [CVS] RPM: rpm/ CHANGES VENDOR rpm/lib/ depends.c
On Oct 28, 2008, at 2:09 PM, devzero2000 wrote: Could be better also to fullfill automatically /etc/rpm/sysinfo/ Dirname e Filelinkto, giving an hint to the user ? In this case there is the possibilty to catch this packaging error and to improve the quality of the system -- and the distro. I did miss one item in the original question, permit me a direct reply. My paraphrasing of what was requested is Can rpmbuild detect orphan directories or dangling symlinks? The RFE is perfectly sensible, based on a well-known principle The earlier you detect an error, the easier it is to fix. When producing *.rpm packages, the only earlier is upstream in most cases. Fixing a problem upstream certainly removes the need to do anything in RPM whatsoever. The problem that is (so far) unsolved is that rpmbuild does not have sufficient information to detect orphan directories or dangling symlinks while building. Packages are unit builds, and largely free of the distro context for which they are being built. So rpmbuild would either need a larger context (i.e. an everything rpmdb for the distro which is intended to include the build), or a narrower (and more restrictive) packaging policy would need to be attempted. I tend to dislike policies implemented by tools that are insufficiently general. But there are more restrictive policies like All package(s) in build set will satisfy parentdir filelinkto dependencies. that could be enforced by rpmbuild without additional context. OTOH, if an rpmdb were maintained incrementally to reflect the current target distro context, where it would be possible to check if __SOME__ package has the necessary parentdir/linkto dependency resolution, that is more general (and less restrictive) a policy than what I just mentioned. However, there is an additional cost associated maintaining an everything rpmdb so that rpmbuild would be permitted a more relaxed (and more general) build policy, with detction or madatory enforcement, simply restated as __SOME__ package(s) in a distro need to satisfy parentdir/filelinkto dependencies. I can go either way, more than willing to do an implementation. My sole interest is improving the quality of *.rpm packages. Package managers should manage package contents, the policies above just force additional items, parent directories symlink end-points, into packages so that the information can be managed just like all other package information. hth 73 de Jeff __ RPM Package Managerhttp://rpm5.org Developer Communication Listrpm-devel@rpm5.org
