Re: [Autogen-users] Fw: 'make check' for autogen-5.18.2 fails

[Bruce K. <bk...@gn...>]

On 01/18/14 16:22, Sergei Steshenko wrote:
> "If your shell does not support "read -u $n", then it is deficient." - probably.
>
> I am using 'dash' not to gain microseconds. As I've upgraded to opensus-13.1, I can't build 'bash' anymore - something is wrong with include files - a type is defined twice.
>
> Since I'm building as much as possible from sources, I can't use 'bash' because I can't build it, so I've switched to 'dash'. There is some problem with 'dietlibc' <-> 'dash' - the latter can supposedly be built with the former, but in reality there is an apparent  problem in 'dietlibc' - a macro is not defined.
>
> Luckily, I can still build 'dash' using stock 'glibc' and I can link it statically.
>
> Now on 'dash' itself and on whether it's broken or not - 'dash' is supposedly POSIX-compliant, and scripts which run under 'bash', but not under 'dash' supposedly have 'bashisms' in them. I am not shell programmer (I mostly program in Perl, "C", GNU Octave), so I honestly do not know whether 'read -u ...' should be supported by a POSIX-compliant shell.
>
> I know that 'dash' has been used by Debian for years, and I know it's their policy to have POSIX-compliabt /bin/sh scripts.
>
> Maybe my test case discovered a 'bashism' - if so, the effort was worth it.


"read -u" is in ksh, bash and most recent (read: since 1987) shells.
It hasn't been "standardized" and dash will include absolutely nothing
that has not been mandated by the Open Group.  ergo, no "-u", despite
being nearly ubiquitous.  ("dash" and Solaris' /bin/sh being the
exceptions.)  So the option is not a "bashism" -- its absence is a dash-ism.

WRT building bash, I've had no trouble building a farily recent version
(a year ago or so).  It's pretty straight forward.  Weird.

Anyway, writing shell scripts is not easy and the extra somersaults
required by stripped down shells aren't worth it.