Re: demand_empty_rest_of_line and ignore_rest_of_line

[Zack Weinberg <zack at codesourcery dot com>]

Hans-Peter Nilsson <hp@bitrange.com> writes:

> On Mon, 26 Apr 2004, Zack Weinberg wrote:
>> I'm talking about the 192 places in the GCC source code where "APP" or
>> "NO_APP" or other names related to emitting these strings in assembly
>> appear.  If it were one, I'd not be complaining.  Two or three even,
>> ok.  192, not cool.
>>
>> (number got by running "find . -type f | grep -v CVS | grep -v ChangeLog |
>> xargs egrep '(\<APP\>|\<NO_APP\>|APP_ON|APP_OFF|app_on|app_off)' | wc -l"
>> in a CVS checkout - yes, documentation deliberately included).
>
> Seems you're counting occurrences in individual target files as
> well. 

Quite deliberately.  I said I'd not be complaining if this crap
appeared in one place, and I meant one place.  In the entire source
tree.  Which is hyperbole; but I do think the impact on the sources
could be narrowed down quite a bit, along the following lines:

 - completely decouple it from TARGET_ASM_FILE_START. GAS wants to see
   it as the very first thing in the file, and it's supposed to be a
   comment so no one else should care if it e.g. comes ahead of a
   .file directive.  So emit it from init_asm_output before
   targetm.asm_out.file_start.  Actually, fold all of
   default_file_start into init_asm_output.

 - eliminate ASM_APP_ON/OFF entirely; the words "NO_APP" and "APP"
   should be invariant, the comment character should be
   ASM_COMMENT_START (which ought to get target-hook-ized with a
   sensible default, "#" looks good).

 - think of a saner way to do what alpha.h is currently doing with
   ASM_APP_ON/OFF.

 - overhaul final_scan_insn so it's not such a horrible mess.

But I'd rather spend time thinking about how to make GAS faster so it
can all go away.

zw