1 |
On Friday 13 April 2007, Ciaran McCreesh wrote: |
2 |
> Mike Frysinger <vapier@g.o> wrote: |
3 |
> > > It masks all kinds of programming screwups. doblah should make a |
4 |
> > > blah, not make a blah and possibly make a directory. |
5 |
> > |
6 |
> > name one |
7 |
> |
8 |
> dosym's old behaviour prevented a broken Vim release (upstream screwed |
9 |
> up a Makefile dependency) from getting into the tree unnoticed. Had |
10 |
> that happened after you changed some (but not all) of the do* |
11 |
> utilities, the duff symlinks would probably have gone unnoticed for a |
12 |
> while. |
13 |
|
14 |
improper package testing that was saved by a dosym that did not create |
15 |
directories ... useful mayhaps, but not nearly enough to justify the proposed |
16 |
change |
17 |
|
18 |
> > you're proposing we suddenly bloat all of our src_install functions |
19 |
> > for no gain at all ... sounds like a no brainer to me |
20 |
> |
21 |
> No, I'm proposing that functions not have strange side effects. |
22 |
|
23 |
the behavior here is desired ... you install into a directory, then the |
24 |
directory should exist |
25 |
|
26 |
> Well no, they can't, because there are a whole load of ebuilds that |
27 |
> will break if they do that. But if it's introduced as mandatory |
28 |
> (barring ebuilds RESTRICTing it) for EAPI 1, the tree will slowly move |
29 |
> towards everything that reasonably can do having working test suites, |
30 |
> which will be a huge step forward for QA. |
31 |
|
32 |
that's really the QA's job to enforce, not the package manager |
33 |
|
34 |
if the QA team wants to spear head a tree wide effort at getting src_test up |
35 |
and running, they're certainly free to |
36 |
-mike |