On Thursday 13 July 2006 07:36, Steve Dibb wrote:
> Henrik Brix Andersen wrote:
> > On Wed, Jul 12, 2006 at 11:11:01PM +0200, Jakub Moc wrote:
> >> Uh... Sorry but it's pretty hard to imagine something more annoying than
> >> an ebuild that dies after a couple of hours compile just because
> >> upstream decided to rename Changelog.txt to ChangeLog.txt and noone
> >> noticed  during version bump, or because someone made a typo there. Fail
> >> to see any benefit from this... :S Ditto for manpages.
> >
> > How could that slip through the initial testing of the ebuild
> > performed by the developer doing the version bump?
>
> Well, it could happen while testing an ebuild. :)  I'd be pretty ticked
> if I were testing Qt and I didn't realize they did change the doc files
> around before doing a test run.

there is always,

1. oh golly I made a mistake
2. rm -rf /var/tmp/portage/{package}/image
3. (vi|emacs|nano) {package}.ebuild
4. FEATURES="noauto" ebuild  {package}.ebuild install
5. victory

sure the may be some packages that don't like going through an install twice 
but I'd hope most of the bigger ones would oblige happily.

A FEATURE like dowarn as an alternate to what John Myers said maybe be good 
for the time conscious dev.

> Besides that though, imho, a simple function with a boolean return type
> shouldn't kill the script executing it.  Throw a warning, yes, but not
> stop everything.
>
> Steve

As a general implementation strategy I think portage should
initially:
die on dobin, doexe

warn on
doins, doman, doexe

After the QA stuff is fixed in the tree, and the general feeling that mass 
hysteria won't break out because of it, change the warns to a die.

ECLASSES

if these use doins/man/doc then the should probably check them before 
installing:
[ -f ${doc} ] && dodoc ${doc}



-- 
Daniel Black <[EMAIL PROTECTED]>
Gentoo Crypto/dev-embedded/Forensics/NetMon

Attachment: pgpvHx9mQGdVB.pgp
Description: PGP signature

Reply via email to