1 |
> I actually have to admit that, apparently, I suffer the same mental |
2 |
> disability of the poster -that is, I used to believe that using |
3 |
> --resume-skipfirst was harmless. |
4 |
|
5 |
In most cases it is harmless (e.g. if you use recompile your whole system |
6 |
without installing anything new or upgrading anything). |
7 |
Even in other cases a loop as mentiond on the webpage _can_ be used |
8 |
safely - but this requires extreme care and/or a lot of knowledge. |
9 |
|
10 |
The problem is that after failing of a package, portage does |
11 |
not recalculate the dependencies, i.e. it will attempt to install also |
12 |
those packages which depend on the failed package. Unfortunately, this |
13 |
does not mean that the packages depending on this will fail either: |
14 |
Due to the dubious autotools, they might recognize that the failed |
15 |
package is not available (if it was a new packages which failed) |
16 |
and build/install successfully but actually without support for this |
17 |
package (although by the intention of the ebuild author and/or according |
18 |
to your useflags they should support it). |
19 |
|
20 |
The correct solution to the problem of failing compiles e.g. over night |
21 |
is the proper implementation of the --keep-going option where these |
22 |
problems do not arise, because dependencies are recalculated after a |
23 |
failure. However, in the absence of this option, the above mentioned |
24 |
loop was the _only_ available workaround. For the reason explained above |
25 |
it _might_ give you an inconsistent system (although the chances for this |
26 |
are not very likely) unless _you_ check/know the dependencies and if in |
27 |
doubt recompile those packages which were "falsely" successfully installed |
28 |
after the failed package. |
29 |
So - as all workarounds - the above mentioned loop had pros and cons |
30 |
(the pro is of course that, usually, your machine would compile through |
31 |
the night, and if you are lucky and e.g. only one package fails and you |
32 |
_know_ that nothing depends on it, you have not wasted the whole night |
33 |
of compiling). |
34 |
In the presence of a --keep-going option, it is now fortunately not |
35 |
necessary anymore to weight the pros and cons. Of course, to insult |
36 |
somebody just because he weighted the pros and cons differently is beyond |
37 |
any acceptable limit. |