Hello,
I think that the http://fedoraproject.org/wiki/PackagingDrafts/StaticLinkage could be ameliorated, and also I am opposed to one point.
* I think that in the motivation a link to Ulrich page could be a good thing, since there are other valid arguments listed there: http://people.redhat.com/drepper/no_static_linking.html
* As showed by the thread https://www.redhat.com/archives/fedora-devel-list/2006-November/msg00713.htm... there is a valid use of static libraries, namely in trusted environements statically linking executables enhance their portability (although, sadly since FC-5 this portability is limited to kernel 2.6.9). Doing the same with dynamically linking is also possible (by providing the libs and using ld hacks), but much less unconvenient. I think that this should be explained in the draft.
* I think that asking for FESCO permission to ship a static lib is wrong, for 3 reasons. One is that packagers may know better than FESCO members if the package is in his area of expertise. Second because I think it is not the FESCO role to participate in reviews. For me FESCO is about general issues, or last resort arbitrage in case of dispute, and there is enough work for FESCO already with those issues. If FESCO is meant to be involved in reviews, it should grow in size over time. And the third reason is that it unnecessarily slow down things and add work to reviewers/submitters.
-- Pat
On Mon, 2006-11-27 at 10:11 +0100, Patrice Dumas wrote:
Hello,
I think that the http://fedoraproject.org/wiki/PackagingDrafts/StaticLinkage could be ameliorated,
Well, FPC's vote on this proposal had been unanimous.
Nevertheless, I am open to listen to further proposals.
and also I am opposed to one point.
- I think that in the motivation a link to Ulrich page could be a good thing, since there are other valid arguments listed there:
Ulrich is approaching this issue from a completely different angle.
We were addressing this issue from a "distro's point of view" having maintainability of the distro itself in mind.
- As showed by the thread
https://www.redhat.com/archives/fedora-devel-list/2006-November/msg00713.htm... there is a valid use of static libraries, namely in trusted environements statically linking executables enhance their portability (although, sadly since FC-5 this portability is limited to kernel 2.6.9).
As many others already pointed out, this claim simply is not true.
IMO, the fact static linkage appears to work for you ("scientific apps"), probably only stems from the simplicity of such applications, because such kind of applications typically don't use much of the OS's resources. If they were, you'd probably notice the brokenness of this approach.
- I think that asking for FESCO permission to ship a static lib is wrong,
for 3 reasons. One is that packagers may know better than FESCO members if the package is in his area of expertise. Second because I think it is not the FESCO role to participate in reviews. For me FESCO is about general issues, or last resort arbitrage in case of dispute, and there is enough work for FESCO already with those issues. If FESCO is meant to be involved in reviews, it should grow in size over time. And the third reason is that it unnecessarily slow down things and add work to reviewers/submitters.
Our motivation for getting FESCO involved is us wanting to collect a list of precedences of cases "when to allow "*-static*" packages".
Technically, there never should be any necessity to allow any static libraries - I.e. "when to allow exceptions" essentially is a _political_ decision - That's why we want to get "political organs" involved.
Ralf
On Mon, Nov 27, 2006 at 10:38:50AM +0100, Ralf Corsepius wrote:
We were addressing this issue from a "distro's point of view" having maintainability of the distro itself in mind.
Indeed, but the fact that some nss and iconv things won't work with statically linked apps seems of relevance to me here.
- As showed by the thread
https://www.redhat.com/archives/fedora-devel-list/2006-November/msg00713.htm... there is a valid use of static libraries, namely in trusted environements statically linking executables enhance their portability (although, sadly since FC-5 this portability is limited to kernel 2.6.9).
As many others already pointed out, this claim simply is not true.
As many others have already pointed out, this claim is true. To be more precise, this claim may be true or not depending on the precise case, there is no general case (if I understand well, there are some things that prevent for sure static linking, that should of course be taken into account).
IMO, the fact static linkage appears to work for you ("scientific apps"), probably only stems from the simplicity of such applications, because such kind of applications typically don't use much of the OS's resources. If they were, you'd probably notice the brokenness of this approach.
Of course, I have never said something different, and that's why I think that static libraries shouldn't be packaged for all the packages. I think it doesn't makes much sense for kde, gnome apps, for example. But for data processing and numerical computations, linking statically enhance portability a lot, and allow to run binaries on computer the user doesn't administer.
Our motivation for getting FESCO involved is us wanting to collect a list of precedences of cases "when to allow "*-static*" packages".
Ok, that makes sense (more generally, documenting difficult packaging issues would be interesting), but then it shouldn't be to ask permission, only to register the packagers decisions.
Technically, there never should be any necessity to allow any static libraries - I.e. "when to allow exceptions" essentially is a _political_ decision - That's why we want to get "political organs" involved.
I disagree, it is a technical decision, not a political one.
-- Pat
packaging@lists.fedoraproject.org
-
Patrice Dumas -
Ralf Corsepius