[RFC] Deprecate @srcdir in pkg_install manifests

[Florent Thoumie]

On Sun, May 9, 2010 at 5:11 AM, Garrett Cooper <yanefbsd at gmail.com> wrote:
> On Sat, May 8, 2010 at 8:44 PM, Garrett Cooper <yanefbsd at gmail.com> wrote:
>> On Sat, May 8, 2010 at 8:25 PM, Garrett Cooper <yanefbsd at gmail.com> wrote:
>>> Hi Arch and Portmgr,
>>>    Found another item that I'm proposing for removal -- @srcdir. Now,
>>> first off, here's what it does (from pkg_create):
>>>
>>>     @srcdir directory
>>>             Set the internal directory pointer for _creation only_ to
>>>             directory.  That is to say that it overrides @cwd for package
>>>             creation but not extraction.
>>>
>>> This construct:
>>>
>>> 1. Isn't used anywhere in /usr/ports's pkg_plist* files.
>>> 2. Isn't used in /usr/src .
>>> 3. Is semi-broken in pkg_create as it's treated as a really awkward
>>> special case, like so:
>>>
>>>        else if (p->type == PLIST_CWD || p->type == PLIST_SRC)
>>>            fprintf(totar, "-C\n%s\n", p->name);
>>>
>>>    So if I specify @srcdir multiple times, pkg_create will fall in on itself.
>>> 4. It over-complicates things, as the -p option basically already sort
>>
>> Correction -- -s and -S handle this functionality:
>>
>>     -s srcdir
>>             srcdir will override the value of @cwd during package creation.
>>
>>     -S basedir
>>             basedir will be prefixed to all @cwd during package creation.
>
> Grr... this isn't nearly as simple as I hoped. -s is the controlling
> factor for @srcdir:
>
>    /* If a SrcDir override is set, add it now */
>    if (SrcDir) {
>        if (Verbose && !PlistOnly)
>            printf("Using SrcDir value of %s\n", SrcDir);
>        add_plist(&plist, PLIST_SRC, SrcDir);
>    }
>
> I'll talk this over with flz, but it's definitely obfuscated design.
>
>>> of provides this level of functionality; the only pro for doing this
>>> that I can think of is if someone had tainted vs untainted files that
>>> they wanted to install, then using @srcdir with a custom manifest and
>>> directory would simplify things. I argue that if they're doing that,
>>> they should be using a chroot or a jail anyhow because package
>>> maintainers would potentially unnecessarily taint the system with
>>> their environment and the packages wouldn't be necessarily as safe to
>>> redistribute.
>>>    Another item I'll be talking about with flz and other folks at
>>> BSDCan, but I wanted to see if anyone had any concerns that they
>>> needed to air here before a final decision was made by portmgr.

As discussed the other day, I don't see any good reason to use -s (aka
@srcdir). The -S option is a different story, even though I'm not sure
anybody actually uses it...

-- 
Florent Thoumie
flz at FreeBSD.org
FreeBSD Committer