[Nasm-devel] on preprocessor pasting code

[Cyrill G. <gor...@gm...>]

Hi Peter

after reading some pasting tokens code history I think we need to put a stone in
and make a hard rule on how tokens might be joined and re-tokenized then.

Single Macro expansion
----------------------

The current rule is

	(TOK_ID | TOK_PREPROC_ID) + (TOK_ID | TOK_PREPROC_ID | TOK_NUMBER)

but it doesn't cover the case if FLOAT is needed. For anothe example the following
code will not work

	%define tail(a) 0xa
	%define head(b) tail(b)

because parser will treat 0xa as a number, and only explicit concat will help, ie
the usage of %+ token.

Multiline Macro expansion
-------------------------

	(TOK_ID | TOK_FLOAT) + (TOK_ID | TOK_NUMBER | TOK_FLOAT | TOK_OTHER)

or/and

	(TOK_NUMBER) + (TOK_NUMBER)

any of there combination will be concat'ed, but still TOK_NUMBER + TOK_FLOAT is not
covered here. The bug in bugzilla happened to hit exactly the case where TOK_NUMBER + TOK_FLOAT
is needed.

So I thought just to add (TOK_NUMBER | TOK_FLOAT) + (TOK_NUMBER | TOK_FLOAT) in both cases
but I fear this is too much.

We definitely have to put a point somewhere. For example C requires explicit sign for concat'ing
preproc tokens. Should we insist to simply use %+ if user need concat? We could leave

	(PREPROC_ID | ID | NUMBER) + (PREPROC_ID | ID | NUMBER)

implicits in 2.10, dropping all others, for 2.09 I may try to restore old behaviour and
fix the bug in bugzilla, but definitely it become messy otherwise.

Thoughts?

P.S. the bug I'm talking about is bug http://sourceforge.net/tracker/?func=detail&aid=3152189&group_id=6208&atid=106208
-- 
    Cyrill