I found it useful to generalize .local

[Damjan Jovanovic]

On Mon, Apr 11, 2011 at 3:26 PM, Michael Thayer
<michael.thayer at oracle.com> wrote:
> On Mon, 2011-04-11 at 12:08 +0200, Damjan Jovanovic wrote:
>> Most *nix applications hardcode their installation prefix at compile
>> time, so you'd have to have exactly the same $HOME as the account
>> under which an RPM was compiled for that to work.
>>
>> This is not an issue that can really be fixed portably: argv[0] is
>> unreliable, POSIX has no API like GetModuleHandle() and
>> GetModulePath() on Windows, /proc/self/exe is Linux-only and can only
>> be used for application, while searching the lines in /proc/self/maps
>> for an internal symbol is the only method available to a library to
>> discover its own path, only works on Linux, and is obscure and unknown
>> to the vast majority of developers.
> Convention says that argv[0] should point to the executable.  If you are
> going to modify an application anyway to support runtime prefix
> discovery why not just make that a requirement (maybe with a default
> path as a fallback) and let it be the caller's problem if they don't set
> argv[0] as they should.  (Let's not discuss setuid here!)
>
> Regards,
>
> Michael

argv[0] doesn't work for libraries and isn't that easy even in
non-setuid cases (you need to check for relative paths, then search
$PATH, and convert relative paths to absolute). Alternatively why
don't we add an API to POSIX for querying your own path? dladdr()
already semi-works in special cases (-rdynamic), on OSs where it's
available.

Damjan