Re: [PATCH] Line synchronisation output in comments

[Eric Blake]

Eric Blake <ebb9 <at> byu.net> writes:

> 
> However, the issue is not just multiline comments, but also multiline
> strings, so I think creating a new TOKEN_COMMENT is not strictly
> necessary, since the fix for one should be applied to the other.

For comparison, here is how Solaris m4 behaves:

=> #line 1 "-"
changecom(/*,*/)
=> 
hello
=> hello
/* multi
=> /* multi
line */
=> line */
`multi
=> multi
line'
=> line
define(foo,``multi
line'')
=> 
=> #line 9
foo
=> multi
=> line
=> #line 10
define(bar, `/*multi 
line*/')
=> 
=> #line 12
dnl
dnl
hi
=> hi
=> #line 15
bar
=> /*multi
=> line*/
=> #line 16

Their algorithm is a bit off (it prints the line directive _before_ collecting 
an input line, so anything that follows omitted lines is off; for example, 
#line 12 should have been #line 14, and then #line 15 was unnecessary).  But 
that output never broke up a multi-line token.

On the other hand:

=> #line 1 "-"
define(foo,a   
b)
=> 
=> #line 3
one foo two
=> one a
=> #line 3
=> b two
define(bar,``a
b'')
=> 
=> #line 6
one bar two
=> one a
=> #line 6
=> b two
hi
=> hi

The second #line 3 is expected; the output of foo is three tokens (a, newline, 
b).  But the second #line 6 broke up a multiline token (the single 
string 'a\nb').  Which means Solaris m4 has the same bug, that it can insert 
line directives inside a multiline comment and mess up a preprocessor's notion 
of the current line for the subsequent line.

-- 
Eric Blake