1 |
On Fri, 09 Nov 2007 18:41:38 +0100 |
2 |
"Marijn Schouten (hkBst)" <hkBst@g.o> wrote: |
3 |
> Use case one: package is completely unversioned upstream. |
4 |
> Have src_fetch add a version as appropriate to the downloaded/mirrored |
5 |
> version. This will work as change of upstream sources will be |
6 |
> detected by all the checksums we do. |
7 |
|
8 |
You're confusing multiple problems here. src_fetch *won't* work for |
9 |
checksums. If we're talking merely badly named tarballs, a better |
10 |
solution is SRC_URI arrow support. |
11 |
|
12 |
> Use case two: package is incompatibly versioned. |
13 |
> smlnj for example releases unversioned files in a versioned |
14 |
> directory. There is currently no way to mirror that in distfiles as |
15 |
> there is nowhere that I could specify that I want files to go to a |
16 |
> separate directory. |
17 |
|
18 |
Again, SRC_URI arrow support is a much better solution. |
19 |
|
20 |
> Right. These use cases are really a bonus. Having src_fetch that we |
21 |
> can redefine is simply the right thing and I can't believe it doesn't |
22 |
> exist already. |
23 |
> |
24 |
> Consider this my vote for an EAPI 2 which adds user-redefinable |
25 |
> src_fetch ASAP. |
26 |
|
27 |
src_fetch is necessary, yes, but it shouldn't be used in the cases you |
28 |
describe. Arrows solve the same problem, don't break mirroring (if |
29 |
implemented correctly) and don't break checksumming. |
30 |
|
31 |
-- |
32 |
Ciaran McCreesh |