| 1 |
On Thu, 2004-06-03 at 12:27 -0400, Aron Griffis wrote: |
| 2 |
> Yes, that's right, but I would avoid it anyway because it takes some |
| 3 |
> extra thought on the part of the ebuild writer (and reader). For |
| 4 |
> example, drop the semicolon and it's a bash error again. |
| 5 |
> |
| 6 |
> On a different topic... it is still wrong because it falls back on |
| 7 |
> configure to make a default decision for python. This is somewhat of |
| 8 |
> a different topic, but all ebuilds that do this are wrong because they |
| 9 |
> produce a non-deterministic build. Here is what should be done, as I |
| 10 |
> mentioned earlier: |
| 11 |
> |
| 12 |
> if use ppc; then |
| 13 |
> myconf="$myconf --without-python" |
| 14 |
> else |
| 15 |
> myconf="$myconf $(use_with python)" |
| 16 |
> fi |
| 17 |
> |
| 18 |
> At some point I'm going to use mcummings' bugzie client to file a |
| 19 |
> barrage of bugs to get maintainers to fix their ebuilds to produce |
| 20 |
> deterministic builds. |
| 21 |
|
| 22 |
The problem is a lot of upstream packages do not do this right (rely on |
| 23 |
autodetection), so I don't think a bug avalanche is the answer here (cause it's a |
| 24 |
_lot_ of packages). It should be made aware to our devs why this is |
| 25 |
important so that configure scripts need to be fixed and the fixes sent |
| 26 |
upstream. |
| 27 |
|
| 28 |
- foser |
Archives