Re: How to patsubst the result of an include?

[Gary V. Vaughan]

Hi Stepan,

Stepan Kasal wrote:
> On Fri, Feb 04, 2005 at 05:53:49PM +0000, Gary V. Vaughan wrote:
>>>     define(`include_orig', defn(`include'))
>>>     define(`include', `indirq(`include_orig', $1)')
> 
> ...
> 
>>Actually, with the attached implementation, that won't work because when
>>GNU M4 puts a file on the input stack (ala include) any partial string
>>that doesn't come from the stack is lost :-(  So it still doesn't solve
>>Matthew's problem.
> 
> 
> Do I understand correctly that
>       qindir(`include', `file')
> wouldn't work?

That depends what you mean by not work. Certainly the open and close quotes
added to the input stack by m4_qindir() are dropped and ignored:

$ echo 'a, b' > foo
$ m4
qindir(`include', `foo')
a, b

define(a, z)

qindir(`include', `foo')
z, b

undivert(`foo')
a, b

But as you can see, undivert does the right thing here anyway.  I almost
made m4_qindir() recognise calls to m4_include() and redirect to
m4_undivert(), but that would be a hack so I restrained myself.  And in
anycase, the output of undivert is not rescanned, and doesn't help Matthew any
with sustituting the contents of a file:

patsubst(undivert(`foo'), `b', `x')
a, b

>  (And perhaps would even emit a spurious right quote string?)

No the right quote is lost too.

> Please let me discuss a problem which is not too much unrelated:
> 
> the following code:
>       define(`x', -'-)
>       define(y, defn(`x'))
>       y
> expands to:
>       --'
> while I believe it should expand to:
>       -'-

Agreed.

> I think that there should be a special ``token type'' for treating defn
> and qindir: the token would carry a string, adn as soon as the token is
> expanded, it would immediately expand to the string.
> 
> In other words, you'd place the string on the stack, but when m4 proceeds
> to do the next expansion of the string, it'd be somehow instructed to
> jump over the string (as if it would find a matching right quote there).
> Since we know the expansion will follow immediately, we could implement
> this hack via one global variable, which would tell the ``expand loop''
> that it should jump to the destination; the ``expand loop'' would clear
> the variable as soon as the message is received and the jump is done.

So that the expansion of a defn or qindir is never rescanned?  I guess
that is the case already for defn.  So far, so good...

> The result of this would be that
>       qindir(`include', `file')
> would include the file literally, even if it contains unmatched right
> quotes.

The implementation I wrote last week does that already, so I'm not sure
what this buys?  (except that it would fix your defn bug, above, of course)

$ echo "a-'-b" > foo
$ m4
define(a, z)dnl
include(`foo')dnl
z-'-b
qindir(`include', `foo')dnl
z-'-b
define(y, qindir(`include', `foo'))dnl
y
z-'-b

> What do you think?

I think that fixing the dropped quote bug in the stack handling, and running
with my existing implementation does the right thing without perturbing the
code so much...

Cheers,
        Gary.
-- 
Gary V. Vaughan      ())_.  address@hidden,gnu.org}
Research Scientist   ( '/   http://tkd.kicks-ass.net
GNU Hacker           / )=   http://www.gnu.org/software/libtool
Technical Author   `(_~)_   http://sources.redhat.com/autobook

signature.asc
Description: OpenPGP digital signature