1 |
On Thu, 15 Jun 2017 19:38:48 +0200 |
2 |
Michał Górny <mgorny@g.o> wrote: |
3 |
|
4 |
> On czw, 2017-06-15 at 18:07 +0200, Alexis Ballier wrote: |
5 |
> > On Thu, 15 Jun 2017 17:59:13 +0200 |
6 |
> > Michał Górny <mgorny@g.o> wrote: |
7 |
> > |
8 |
> > > On śro, 2017-06-14 at 16:09 +0200, Alexis Ballier wrote: |
9 |
> > > > On Wed, 14 Jun 2017 15:57:38 +0200 |
10 |
> > > > Michał Górny <mgorny@g.o> wrote: |
11 |
> > > > [...] |
12 |
> > > > > > [...] |
13 |
> > > > > > > > > > > [1]:https://wiki.gentoo.org/wiki/User:MGorny/GLEP:ReqUse |
14 |
> > > > > > > > > > |
15 |
> > > > > > > > > > I really don't like the reordering thing. Even the |
16 |
> > > > > > > > > > restricted syntax does not fix the issue with '^^ |
17 |
> > > > > > > > > > ( a b ) b? ( a )' already mentioned here. It'd be |
18 |
> > > > > > > > > > much better and simpler for the spec just to assign |
19 |
> > > > > > > > > > a fixed value and use the solving rules with |
20 |
> > > > > > > > > > those. |
21 |
> > > > > > > > > |
22 |
> > > > > > > > > You're not going to convince me by providing examples |
23 |
> > > > > > > > > that are utterly broken by design and |
24 |
> > > > > > > > > meaningless ;-). |
25 |
> > > > > > > > |
26 |
> > > > > > > > Well... if it's so obvious that the example is broken by |
27 |
> > > > > > > > design that you don't even bother to explain why, I |
28 |
> > > > > > > > assume you have an algorithm for that. Where is the |
29 |
> > > > > > > > code ? What are the numbers ? How many ebuilds might |
30 |
> > > > > > > > fail after reordering ? How can this be |
31 |
> > > > > > > > improved ? |
32 |
> > > > > > > |
33 |
> > > > > > > Are you arguing for the sake of arguing here? I just |
34 |
> > > > > > > presumed that this example is so obviously broken there |
35 |
> > > > > > > is no point wasting any more time on it. The code of |
36 |
> > > > > > > nsolve clearly detects that, so I don't really understand |
37 |
> > > > > > > what you're trying to prove here. |
38 |
> > > > > > |
39 |
> > > > > > Those are real questions. You should take breath, think a |
40 |
> > > > > > bit about it, and try to run the 2 possible orderings of |
41 |
> > > > > > the ^^ through nsolve or even solve.py. They both are very |
42 |
> > > > > > happy (and are right to be) with the above ordering. You |
43 |
> > > > > > might want to think a bit more about what is the relation |
44 |
> > > > > > between this broken 10 chars example and the 10 lines |
45 |
> > > > > > python targets one below. |
46 |
> > > > > > |
47 |
> > > > > > You should also realize that all the above questions have |
48 |
> > > > > > already been answered in length if you do as I |
49 |
> > > > > > suggest. |
50 |
> > > > > |
51 |
> > > > > No. I have already spent too much time on this. We're already |
52 |
> > > > > long past all useful use cases, and now I feel like you're |
53 |
> > > > > going to argue to death just to find a perfect algorithm that |
54 |
> > > > > supports every absurd construct anyone can even write, if |
55 |
> > > > > only to figure out the construct is completely useless. |
56 |
> > > > |
57 |
> > > > I'm not going to argue to death. It's already proven reordering |
58 |
> > > > is broken. |
59 |
> > > > |
60 |
> > > > > If you want to play with it more, then please by all means do |
61 |
> > > > > so. |
62 |
> > > > |
63 |
> > > > There is nothing to do for reordering. It's broken by design. |
64 |
> > > > |
65 |
> > > > > However, do not expect me to waste any more of my time on it. |
66 |
> > > > > I've done my part, the code works for all reasonable use |
67 |
> > > > > cases and solves all the problems I needed solving. If you |
68 |
> > > > > want more, then it's your job to do it and solve the |
69 |
> > > > > resulting issues. |
70 |
> > > > |
71 |
> > > > Like... writing code handling all the cases and describing how |
72 |
> > > > it works ? We're past that. The only thing we're not past is |
73 |
> > > > that you fail to understand it and attempt to block it. |
74 |
> > > > |
75 |
> > > |
76 |
> > > Then please provide a single valid example that: |
77 |
> > |
78 |
> > app-text/wklej-0.2.1-r1 ^^ ( python_single_target_pypy |
79 |
> > python_single_target_pypy3 python_single_target_python2_7 |
80 |
> > python_single_target_python3_4 python_single_target_python3_5 |
81 |
> > python_single_target_python3_6 ) python_single_target_pypy? |
82 |
> > ( python_targets_pypy ) python_single_target_pypy3? |
83 |
> > ( python_targets_pypy3 ) python_single_target_python2_7? |
84 |
> > ( python_targets_python2_7 ) python_single_target_python3_4? |
85 |
> > ( python_targets_python3_4 ) python_single_target_python3_5? |
86 |
> > ( python_targets_python3_5 ) python_single_target_python3_6? |
87 |
> > ( python_targets_python3_6 ) vim? ( ^^ |
88 |
> > ( python_single_target_python2_7 ) ) |
89 |
> > |
90 |
> > |
91 |
> > Simplified as: |
92 |
> > ^^ ( a b ) c? ( b ) |
93 |
> > |
94 |
> > (see the pattern now ? :) ) |
95 |
> > |
96 |
> > > a. is completely 'correct' (that is, provides a valid, predictable |
97 |
> > > and acceptable solution) with the default ordering O_a, |
98 |
> > |
99 |
> > c? ( b ) ^^ ( b a ) |
100 |
> > |
101 |
> > |
102 |
> > > b. is not 'correct' with at least one reordering O_b (assuming |
103 |
> > > only |
104 |
> > > > > , ^^, ?? is subject to reordering), |
105 |
> > |
106 |
> > c? ( b ) ^^ ( a b ) |
107 |
> > |
108 |
> > > |
109 |
> > > c. nsolve reports O_a as all good, and O_b as not good. |
110 |
> > |
111 |
> > I'll let you run this. It does. |
112 |
> > |
113 |
> > > The best way to convince me is through valid examples. |
114 |
> > |
115 |
> > |
116 |
> > It is also easier to be convinced when you try to understand and ask |
117 |
> > for clarifications instead of just rejecting without thinking :) |
118 |
> > |
119 |
> |
120 |
> Ok, now I get your point. Not that I like it but I don't see any sane |
121 |
> way around it. |
122 |
> |
123 |
> The question then is, how can you reliably ensure that developers will |
124 |
> use the same ordering in cross-relevant packages? For example, |
125 |
> consider the following: |
126 |
> |
127 |
> ^^ ( provider_ssl_openssl provider_ssl_libressl ) |
128 |
> |
129 |
> Since those providers block each other, all packages will have to have |
130 |
> either openssl or libressl consistently (or another provider). |
131 |
|
132 |
Yep that'd be a problem for things having a global impact but since |
133 |
REQUIRED_USE is merely local there's not much to be done at that level |
134 |
I think. |
135 |
|
136 |
> The reordering idea was mostly addressing this. However, it just |
137 |
> occurred to me it only solved some the case when user selected both |
138 |
> and not the one where neither was preferred over the other. |
139 |
> |
140 |
> Without reordering, I think we need to enforce the specific ordering |
141 |
> consistently across the tree. Any idea how to achieve that? |
142 |
|
143 |
There will always be the case where one developer uses the implication |
144 |
form with one ordering and another one the form with the other |
145 |
ordering, so I don't think there would be any solution on the spec or |
146 |
algorithmic side. It should rather be some policy or guideline that |
147 |
aims to that. If there's too much duplication then an eclass defining |
148 |
the proper order and the deps to use could be added (like python |
149 |
eclasses do); we could add some repoman warning that e.g. openssl must |
150 |
be preferred over libressl (e.g. there is never a 'libressl? !openssl' |
151 |
nor '!openssl? libressl' sub-implication in the implication form). For |
152 |
the eclass case, it might be that one day we will want to reverse the |
153 |
order, potentially breaking the solver, but hopefully your CI checks |
154 |
will catch that. |
155 |
|
156 |
|
157 |
Alexis. |