On Wed, 2014-10-15 at 23:36 +0100, Ian Jackson wrote: > Actually, the problem is indeed in policy. In its resolution of > #539158 the TC decided unanimously (but unfortunately slightly > implicitly) that printf ought to be provided by our /bin/sh. > > Unfortunately the policy has not been properly clarified. This leaves > us in the somewhat unsatisfactory situation where our real > compatibility requirement is de facto rather than de jure. > > As the maintainer of a minority shell, Thorsten has the most interest > in regularising this. Perhaps Thorsten would like to propose a > suitable policy wording (with a view to changing posh to match). > > Obviously that wording ought to be consistent with the TC's decision > in #539158 - ie, it should specify printf as a builtin. The arguments about printf in #539158 also applies to '['. POSIX does not say '[' must be a built in (in POSIX's terminology is part of the 'Special Built-In Utilities'). Thus if the shell didn't implement '[' udev would fail since uses [ and sets PATH to be /bin:/sbin. The reality is in a POSIX (or a minimal Policy 10.4) world shell scripts must have access to bulk of the stuff that is both covered in the man1p pages and is required in Debian. Turns out only three commands fall into that category: [, printf, and test. And yes, to me the obvious fix is say in policy /bin/sh have those commands as builtins.
Description: This is a digitally signed message part