| 1 |
On Fri, 2008-08-08 at 09:40 +0200, Fabian Groffen wrote: |
| 2 |
> On 31-07-2008 13:48:06 +0200, Michael Haubenwallner wrote: |
| 3 |
> > When using GNU getopt, these warnings go away on solaris ;) |
| 4 |
> |
| 5 |
> gnulib is in the tree now, and it seems to work. Great! |
| 6 |
> |
| 7 |
> However, looking at the ebuild changes, I immediately went wondering... |
| 8 |
|
| 9 |
The current somehow large ebuild changes are a first shot to get some |
| 10 |
experience and because of exactly this discussion missing. I've waited |
| 11 |
four you to get a look at these changes ;) |
| 12 |
|
| 13 |
> |
| 14 |
> In Prefix we are very much GNU-minded. Some platforms (e.g. Solaris 10 |
| 15 |
> and Darwin) have mostly sufficient implementations of e.g. getopt. |
| 16 |
> However, they are not GNU. Do we want to use them? Sometimes they |
| 17 |
> appear to be a little bit different, or just insufficient. Interix |
| 18 |
> seems to miss them entirely. |
| 19 |
|
| 20 |
I'd say use the GNU ones if gnulib detects they're better. |
| 21 |
|
| 22 |
> |
| 23 |
> In other words, for those platforms which do not use elibc_glibc, why |
| 24 |
> don't we install gnulib as part of @system, |
| 25 |
|
| 26 |
Yes, so there's no need to add an explicit platform specific DEPEND. |
| 27 |
|
| 28 |
> and have the linker wrapper inject -lgnu? |
| 29 |
|
| 30 |
No (but maybe it's worth a try). |
| 31 |
Any package that requires libgnu should activate it explicitly. |
| 32 |
But it's fine to have some eclass function which does nothing for |
| 33 |
elibc_glibc, and adds proper flags for others, so the activation is a |
| 34 |
one-liner per ebuild. |
| 35 |
|
| 36 |
Iff it is added by default, we definitely need a way to disable this, |
| 37 |
see below. |
| 38 |
|
| 39 |
Question is how many packages break when it's added by default, compared |
| 40 |
to how many packages need it. |
| 41 |
|
| 42 |
> (Assuming we would change the install location to |
| 43 |
> /usr/lib. That means we would automatically always take the GNU |
| 44 |
> replacements, without any extra efforts. Drawback, if a platform |
| 45 |
> implements vasprintfi, dup2 or something, we add extra bloat to the |
| 46 |
> binary, and a possibly not as efficient implementation as the platform's |
| 47 |
> native implementation. Thoughts? |
| 48 |
|
| 49 |
gnulib does not put any function into libgnu.a if the host os' libc is |
| 50 |
sufficient. Fex you get an empty libgnu.a with elibc_glibc. |
| 51 |
|
| 52 |
Now, when -lgnu is there always, we cannot rebuild gnulib itself, as |
| 53 |
it's configure would detect sufficient "libc", and build an empty |
| 54 |
libgnu.a. |
| 55 |
This is the main reason why not to add -lgnu always, or the need to |
| 56 |
disable it. |
| 57 |
|
| 58 |
/haubi/ |
| 59 |
-- |
| 60 |
Michael Haubenwallner |
| 61 |
Gentoo on a different level |
Archives