view TODO @ 203:3a25180d3a5c

Abort on line numbering or column numbering overflow. Line numbers are limited to values that fit in "unsigned int". Also reject input lines longer than 2^32-1 characters. It seems reasonable to presume that any input that violates these constraints is someone screwing around and not a serious attempt to compile or preprocess anything useful. Done in response to n2129, but without getting into any of the silliness found there.
author David A. Holland
date Tue, 01 Aug 2017 14:51:04 -0400
parents 6947cd2cde79
children
line wrap: on
line source

not implemented:
	- mode.input_allow_dollars.
	- column counts do not take tabstops into account.
	- mode.output_linenumbers.
	- mode.do_depend.
	- mode.do_macrolist.
	- mode.do_trace.
	- warns.endiflabels. (they cause errors)
	- warns.unused.	
	- the -iremap option.
	- $CPP_RESTRICTED
	- other environment variables

tidy up:
	- get rid of inlinedefs.h
	- use of places in and below macro.c is pretty bogus.
	- macro code should be reworked.
	- place_changefile is manky and wastes memory. Also, in an ideal
	  world we'd remember the place #line changed the name and refer
	  to it when printing errors.

fix:
	- "#if 0 && 1/0" should not crash; fix eval method.
	- quote characters and comment delimiters that are emitted by
	  macros are not recognized. See:
		t34 (should produce a quote and FOO Q)
		t35 (similarly, this test may be redundant once it's fixed)
		t36 (C(3) should produce nothing)
		t37 (BC foo EC should produce nothing)
	  Joerg says comments like in t36 should be stripped exactly
	  twice, once when the macro is defined and again when it's
	  expanded. Note that gcc's cpp -traditional is getting t37
	  wrong, and it gets t36 wrong with -C.
	- remove the intentionally undocumented -p option and generate
	  proper linenumber output. (also, in this code what happens if
	  a comment spans files? I bet currently it will emit the line
	  number into the comment.)