We have noticed a growing trend of the final symbolic link not being established for libraries. Since the linker auto name mangling breaks because of this issue it causes the platform to be a problematic development host. Either the system administrators have to go an manually establish missing links or build process must be modified to use full paths to library files. Neither of these solutions are desirable. We are seeing this with RedHat Enterprise and other distributions as well.
Question, how long do we think a distribution will be an acceptable development hosting platform because of this issue? We think this maybe due to disagreement over what version of a library should be the default. Taking the current approach is not a good long term solution. We believe the most current version should be the default and if a package needs an older version then that package should have to customize their installation.
This is not a flame, it is meant to be constructive.
On 11/07/2013 04:36 AM, David Highley wrote:
We have noticed a growing trend of the final symbolic link not being established for libraries.
Could you provide an example?
To me, your description sounds like as if you have encountered packaging bugs in some packages.
Ralf
"Ralf Corsepius wrote:"
On 11/07/2013 04:36 AM, David Highley wrote:
We have noticed a growing trend of the final symbolic link not being established for libraries.
Could you provide an example?
To me, your description sounds like as if you have encountered packaging bugs in some packages.
Then there are several; X11, libxml2, libxslt, and others. Its not a simple script to try and identify all the ones missing as there are no hard and fast naming rules for libraries so that you could easily parse the real library and then check for symlink.
Ralf
packaging mailing list packaging@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/packaging
On Thu, 7 Nov 2013 10:13:24 -0800 (PST), David Highley wrote:
"Ralf Corsepius wrote:"
On 11/07/2013 04:36 AM, David Highley wrote:
We have noticed a growing trend of the final symbolic link not being established for libraries.
Could you provide an example?
To me, your description sounds like as if you have encountered packaging bugs in some packages.
Then there are several; X11, libxml2, libxslt, and others. Its not a simple script to try and identify all the ones missing as there are no hard and fast naming rules for libraries so that you could easily parse the real library and then check for symlink.
Sounds a bit as if you're missing the *-devel packages for those libs. libX11-devel libxml2-devel and so on.
"Michael Schwendt wrote:"
On Thu, 7 Nov 2013 10:13:24 -0800 (PST), David Highley wrote:
"Ralf Corsepius wrote:"
On 11/07/2013 04:36 AM, David Highley wrote:
We have noticed a growing trend of the final symbolic link not being established for libraries.
Could you provide an example?
To me, your description sounds like as if you have encountered packaging bugs in some packages.
Then there are several; X11, libxml2, libxslt, and others. Its not a simple script to try and identify all the ones missing as there are no hard and fast naming rules for libraries so that you could easily parse the real library and then check for symlink.
Sounds a bit as if you're missing the *-devel packages for those libs. libX11-devel libxml2-devel and so on.
Ah, now we see the subtlety in the split between development and run time packages. It had not dawned on us that the final symlink is tied to the development package, especially as their are other symlinks being applied.
-- packaging mailing list packaging@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/packaging
On Thu, 7 Nov 2013 18:33:13 -0800 (PST), David Highley wrote:
Sounds a bit as if you're missing the *-devel packages for those libs. libX11-devel libxml2-devel and so on.
Ah, now we see the subtlety in the split between development and run time packages. It had not dawned on us that the final symlink is tied to the development package, especially as their are other symlinks being applied.
There is no "final" symlink. They are two separate types of symlinks: The versioned one is maintained by ldconfig for the runtime. The non-versioned .so is used by build-time tools (and may point directly at the fully versioned lib file - there is no need to create a chain of symlinks, such as .so -> .so.1 -> .so.1.0.0).
Note that the split into run-time and build-time packages is somewhat controversial. There are exceptions, where the non-versioned .so symlink is used at runtime (e.g. via a dlopen method). At Fedora, it is not always tried to patch such code to open the versioned library instead. For example, if the developers insist on opening the non-versioned .so (if installed) and try to support a broad range of library interfaces instead of expecting a specific library version. Sometimes packagers accept the burden of adding and updating a patch, but there are exceptions.
packaging@lists.fedoraproject.org
-
David Highley -
Michael Schwendt -
Ralf Corsepius