1 |
On Mon, 2 Jan 2017 12:49:59 -0500 |
2 |
Rich Freeman <rich0@g.o> wrote: |
3 |
|
4 |
> However, in this case why would we want to rule out sets, "and all the |
5 |
> other shenanigans?" We've already established that a single stable |
6 |
> request bug can apply to multiple package-versions, so why not allow |
7 |
> full dependency specifications? If there is a set that describes what |
8 |
> needs to be stabilized in an atomic operation, then what is the value |
9 |
> in breaking it down into a million separate =-only atoms? |
10 |
|
11 |
That rather complicates validation. |
12 |
|
13 |
Mostly, because if you declare a list of dependencies, if any of them is a range, |
14 |
the subgraph of that specific atom becomes variable, not constant. |
15 |
|
16 |
Which means you may have omitted one of the possible dependencies from one of the possible |
17 |
subgraphs that an AT/Keyworder will see. |
18 |
|
19 |
This IMO would mostly negate the utility of submitter defined lists of specifications. |
20 |
|
21 |
In that, by making the submitter resolve it all, its either "good" or "bad" |
22 |
|
23 |
Instead of leaving the person doing the testing in a confused state about which packages |
24 |
are expected to be used. |
25 |
|
26 |
The sets as defined should either work, and the person doing them should get each and every one |
27 |
working as expected, or none of them should, and it should be pushed back to the submitter to |
28 |
fix their specification. |
29 |
|
30 |
Levity in tester interpretation is likely to introduce side effects. |