Re: frozen files format 2

[Gary V. Vaughan]

Hi Eric,

Eric Blake wrote:
> Eric Blake <ebb9 <at> byu.net> writes:
> 
>> Ouch.  We aren't compatible here, without adding the feature of partial 
>> input 
>> across EOF.  With Solaris 8:
>>
>> $ echo "m4wrap(\`abc)
>>> ')len(" | m4
>> NONE:0: m4: ERROR: EOF in argument list
>> $ echo "m4wrap(\`abc)
>>> ')len(" | /usr/ccs/bin/m4
>> 3
> 
> It looks like Solaris doesn't mind EOF during argument collection, but does 
> mind EOF during strings:
> 
> $ echo "define(a,\`1" > a.m4   
> $ echo "2')a" > b.m4
> $ m4 a.m4 b.m4
> NONE:0: m4: ERROR: EOF in string
> $ /usr/ccs/bin/m4 a.m4 b.m4
> 
> /usr/ccs/bin/m4:a.m4:2 EOF in quote
> define(a,1
> )
> $ 
> 
> But if we are going to allow partial input, I would argue that partial 
> strings 
> should not behave any differently than partial arguments

Agreed.

> when we are not in -G/--traditional mode.

Why the distinction for --traditional?  I don't see a problem with
supporting EOF in quotes or definitions regardless of mode.. in much the
same way that we handle 10 or more arguments to macros.

We do, however, need to be careful to still throw correct diagnostics
when we genuinely fall off the end of the last file.  Perhaps we should
also be clever about EOF across boundaries between named files and
stdin.  For example:

  $ echo "len(" > a.m4
  $ echo "abc)" > b.m4
  $ m4 a.m4 b.m4
  3
  $ m4 a.m4 -
  NONE:0: m4: ERROR: EOF in argument list

The rationale being that the 'm4 a.m4 -' idiom is for slurping setup
from a.m4 for an interactive session.

Cheers,
        Gary.
-- 
Gary V. Vaughan      ())_.  address@hidden,gnu.org}
Research Scientist   ( '/   http://blog.azazil.net
GNU Hacker           / )=   http://trac.azazil.net/projects/libtool
Technical Author   `(_~)_   http://sources.redhat.com/autobook

signature.asc
Description: OpenPGP digital signature