1 |
Am Montag, 29. Mai 2017, 19:32:48 CEST schrieb Walter Dnes: |
2 |
|
3 |
> |
4 |
> The irc log mentions problems with some ebuilds under some versions of |
5 |
> GCC. In a similar vein, are there any anticipated issues with binaries |
6 |
> built under older GCC versions running on a system that is otherwise |
7 |
> all-GCC-6.3.0? I'm thinking of pre-built binaries like Pale Moon (from |
8 |
> their website), firefox-bin, seamonkey-bin, openoffice-bin, |
9 |
> libreoffice-bin, libreproject-bin, etc. |
10 |
|
11 |
Yes, but you already should have these issues (and it's a different problem). |
12 |
|
13 |
A] Here, we've been talking about changes in the default language standard. |
14 |
-- GCC up to 5.* defaults to -std=gnu98, while GCC 6 defaults to -std=gnu14 |
15 |
-- ICU 59 >requires< its reverse dependencies to build with at least -std=c+ |
16 |
+11 |
17 |
-- So all packages that a) use gcc-4 or gcc-5, and b) do not in the ebuild |
18 |
"manually" add something like -std=c++11 or -std=c++14 or -std=gnu14 will fail |
19 |
to *build*. |
20 |
|
21 |
B] What you are talking about is a change in the ABI, the binary interface of |
22 |
the libraries. That happened as well (when gcc-5.4 was stabilized), and was |
23 |
the reason for the "rebuild your stuff" newsitem. |
24 |
-- GCC-5 and later internally has a different representation of C++ stuff, |
25 |
compared with gcc-4 |
26 |
-- Building a library with gcc-4, and an executable that links to it with |
27 |
gcc-5, or vice versa, may fail with "missing symbols". |
28 |
-- If you have a binary app built with gcc-4, and your libraries e.g. are |
29 |
built with gcc-5, you may run into the same problem (and your app won't |
30 |
start). |
31 |
-- This is why e.g. unbundling of libraries in vmware-workstation was at some |
32 |
point not possible anymore. |
33 |
-- I don't know of any workaround. |
34 |
|
35 |
HTH |
36 |
|
37 |
-- |
38 |
Andreas K. Hüttel |
39 |
dilfridge@g.o |
40 |
Gentoo Linux developer (council, perl, libreoffice) |