| 1 |
On Tue, May 16, 2006 at 03:56:38PM -0700, Brian Harring wrote: |
| 2 |
> If your parent parsing implementation handled N parents on a single |
| 3 |
> line (rather then parent per line as you do now), portage would |
| 4 |
> explode rather then silently use the left most. Your implementation |
| 5 |
> isn't doing that however... |
| 6 |
|
| 7 |
And that won't even work. Woot. About the only guranteed "portage |
| 8 |
will choke and not invalidly use N parent profiles if it doesn't |
| 9 |
support it are"- |
| 10 |
1) specify the parents prefixed with #, fex |
| 11 |
# .. |
| 12 |
# ../blah |
| 13 |
End result is that python will throw a TypeError (might be |
| 14 |
ValueError). Not incredibly clean. |
| 15 |
2) insane profile inheritance version trick. Require the |
| 16 |
leftmost parent to point at a custom profile, say profiles/version1, |
| 17 |
which has a nice little bashrc along the lines of |
| 18 |
die "upgade your $PKG_MANAGER- your current version cannot parse this |
| 19 |
profile properly". |
| 20 |
If the package manager knows of version1, it just filters that parent |
| 21 |
out as it goes- if it doesn't, any/all ebuild actions result in a |
| 22 |
forced bail as soon as control is transfered over to ebuild env. |
| 23 |
|
| 24 |
Both solutions suck, and the alternatives of "just ignore it" or |
| 25 |
"require left parsed profile to be minimally sane" don't cut it all |
| 26 |
that well. |
| 27 |
|
| 28 |
~harring |
Archives