Re: configure.lineno?

[Paul Eggert]

bonzini <address@hidden> writes:

> bash reports a $LINENO relative to the start of the function inside
> shell functions.

That bug is fixed in Bash 2.05b, released 2002-07-17.

The bug is also present in ksh M-11/16/88i (still quite popular, e.g.,
it's /bin/ksh on Solaris 9), but was fixed some time before
M-12/28/93d came out.

The bug is not present in PD KSH v5.2.14 99/07/13.2.

> 1) revert to the old __oline__ mechanism that was used in autoconf
> 2.13 (not very aesthetic)
> 
> 2) always use configure.lineno (slow)
> 
> 3) implement __oline__ in m4 2.0, and rever to configure.lineno when
> configure scripts are created with an older m4 (who knows the timeline
> for m4 2.0 and libtool 1.6, on which m4 depends?).
> 
> What's the best one?

How about the following possibilities as well?

4) Don't use $LINENO inside shell functions that we write, and warn users
   that $LINENO isn't portable when used inside shell functions.

5) Dynamically test whether $LINENO is working inside shell functions,
   and prepend the function name to $LINENO if the bug is present.

6) Go ahead and use $LINENO within shell functions, but warn users that
   the line numbers will be bogus in older shells.

7) Rewrite the code to prefer a shell that supports both shell
   functions and proper LINENO (e.g., bash 2.05b, ksh93) to a shell
   that doesn't.

Personally I'd prefer (6) + (7).