svn commit: r348866 - head/ports-mgmt/poudriere-devel

[Bryan Drewery]

On 3/23/2014 1:31 PM, Adam Weinberger wrote:
> On 23 Mar 2014, at 14:11, Bryan Drewery wrote:
> 
>> On 3/23/2014 12:49 PM, Adam Weinberger wrote:
>>> On 23 Mar 2014, at 12:20, Bryan Drewery wrote:
>>>
>>>> On 3/23/2014 11:18 AM, Mathieu Arnold wrote:
>>>>>
>>>>>
>>>>> +--On 23 mars 2014 11:15:51 -0500 Bryan Drewery <bdrewery at FreeBSD.org>
>>>>> wrote:
>>>>> | On 3/23/2014 11:11 AM, Mathieu Arnold wrote:
>>>>> |> +--On 23 mars 2014 15:40:11 +0000 Bryan Drewery
>>>>> <bdrewery at FreeBSD.org>
>>>>> |> wrote:
>>>>> |> |  .include <bsd.port.options.mk>
>>>>> |> |
>>>>> |> | +post-install:
>>>>> |> |  .if ${PORT_OPTIONS:MZSH}
>>>>> |> | -PLIST_FILES+=    share/zsh/site-functions/_poudriere
>>>>> |> | -PLIST_DIRSTRY+=    share/zsh/site-functions
>>>>> |> | -PLIST_DIRSTRY+=    share/zsh/
>>>>> |> | -.else
>>>>> |> | -MAKE_ENV+=    NO_ZSH=yes
>>>>> |> | +    @${MKDIR} ${STAGEDIR}${PREFIX}/share/zsh/site-functions/
>>>>> |> | +    ${INSTALL_DATA} ${WRKSRC}/completions/zsh/_poudriere \
>>>>> |> | +        ${STAGEDIR}${PREFIX}/share/zsh/site-functions/
>>>>> |> |  .endif
>>>>> |>
>>>>> |> I think you could do without the include and always install the
>>>>> file.
>>>>> |>
>>>>> |
>>>>> | What?
>>>>>
>>>>> Replace:
>>>>>
>>>>> .include <bsd.port.options.mk>
>>>>>
>>>>> post-install:
>>>>> .if ${PORT_OPTIONS:MZSH}
>>>>>     @${MKDIR} ${STAGEDIR}${PREFIX}/share/zsh/site-functions/
>>>>>     ${INSTALL_DATA} ${WRKSRC}/completions/zsh/_poudriere \
>>>>>         ${STAGEDIR}${PREFIX}/share/zsh/site-functions/
>>>>> .endif
>>>>>
>>>>> With:
>>>>>
>>>>> post-install:
>>>>>     @${MKDIR} ${STAGEDIR}${PREFIX}/share/zsh/site-functions/
>>>>>     ${INSTALL_DATA} ${WRKSRC}/completions/zsh/_poudriere \
>>>>>         ${STAGEDIR}${PREFIX}/share/zsh/site-functions/
>>>>>
>>>>> The package won't register the file if the option is disabled, and it
>>>>> removes the need for make to parse bsd.port.mk twice.
>>>>>
>>>>
>>>> And then there is a stage orphan.
>>>>
>>>> No.
>>>
>>> Staging only installs things listed in the plist. If the ZSH option is
>>> enabled, that file is in the plist. Stage orphans are files that never
>>> get installed, no matter which options are selected.
>>
>> Right, there are no *leftovers*, but there are *orphans*. Files which
>> are in stagedir and never installed. I invented the whole idea of stage
>> orphans and checking for them in poudriere, before it was a target in
>> ports. Installing a file to the stagedir and never installing in the
>> package creates a fatal error in poudriere as an orphan. Leaving
>> false-positive orphans creates a very hard to maintain situation.
>> Someone will come along in the future and update, see orphans, and not
>> know if they are new or not. The point is to find files you forgot to
>> add to the plist.
>>
>> Sorry but the stagedir cannot have orphans in it, it's not intended for
>> that. It's intended to make leftovers impossible and allow packaging as
>> non-root. Leaving untracked files in there is a big red flag that your
>> package is missing something. You must handle them either by not
>> installing them or by removing them.
>>
>> People are missing this point. You must keep the stagedir clean if you
>> want to pass poudriere testing. Many people have been giving incorrect
>> advice on how to treat the stagedir and it's going to be a big problem
>> once we start doing daily QAT/poudriere on the tree to find leftovers
>> and orphans. Right now orphans are going unfound because QAT/tinderbox
>> doesn't check for orphans. That is a big problem right now. When staging
>> a port, you now have no idea if there are files missing from plist if
>> you are depending on QAT.
>>
>> I really wish people would stop giving the awfully incorrect advice
>> here. People are making it into some kind of style or opinion matter,
>> but it's simply not the case.
>>
>> There are ideas and plans to only consider an orphan as something that
>> is not in any plist (for subpackages), but we don't have subpackages
>> yet. So we are left with keeping the stagedir clean.
> 
> The ability to dramatically simply a port's Makefile and let simple
> plist %%OPTIONS%% figure out the rest is an incredibly powerful, useful,
> and advantageous thing. I am having trouble understanding why we should
> give it all up so "to pass poudriere testing." Ports work perfectly
> right now. They install exactly what we intend them to install, and it
> works perfectly for users. If poudriere is unhappy, then it seems to me
> that the problem is with poudriere.
> 
> Maybe it wasn't how you originally envisioned that staging would work.
> But people have adopted it not because of "awfully incorrect advice,"
> but because that's how we want it to work.
> 
> Hitachi doesn't recall Magic Wands just because they wanted a back
> massager.

I also want it work like that. However, having false-positives leads to
ignoring them entirely and causing breakage.

I am working to merge the poudriere and check-orphans script into
check-orphans, enable in tinderbox, and also cleanup EXAMPLES/DOCS
false-positives with check-orphans.

I also have work in progress for a way to mark files as explicitly
ignored so they stop showing in orphans.

> 
> # Adam
> 
> 
> -- 
> Adam Weinberger
> adamw at adamw.org
> http://www.adamw.org


-- 
Regards,
Bryan Drewery

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 553 bytes
Desc: OpenPGP digital signature
URL: <http://lists.freebsd.org/pipermail/svn-ports-head/attachments/20140329/9194f69c/attachment.sig>